Computation offloading strategy in mobile edge computing can help users decide how to execute tasks, which is related to user experience, and has become a research hotspot in mobile edge computing. At present, most computation offloading strategies are carried out under the condition of overall task offloading, and only consider a single indicator of delay or energy consumption, and do not combine the two for optimization. To solve this problem, this paper takes the weighted sum of task processing delay and energy consumption as the optimization goal, and proposes a partial offloading algorithm based on reinforcement learning. We divide the processing of a single task into local computing and partial offloading computing, and introduce a variable to determine the offloading weight in partial offloading. Finally, we use reinforcement learning Q-learning to complete the computation offloading and resource allocation of all tasks. Experimental results show that the proposed algorithm can effectively reduce the delay and energy consumption of task processing.

In the field of indoor applications of UWB(Ultra Wide Band) positioning technology, it is important to establish an efficient and accurate 3D coordinate positioning system to overcome signal interference. Machine learning methods are used to investigate the problem of accurate positioning of indoor UWB signals under interference. Firstly, various statistical analysis models are used to clean up invalid or error measurements, then the a priori knowledge of TOF ( Time Of Flight) algorithm is combined with neural network and XGBoost algorithm to build a neural XGB(Exterme Gradient Boosting) 3D oriented system. The system can accurately predict the coordinate value of the target point by “ normal data冶 and “ abnormal data冶 ( disturbed), the coordinates of four anchor points, and the final error is as low as 5. 08 cm in two鄄dimensional plane and 8. 03 cm in three鄄dimensional space. A neural network classification system is established to determine whether the data is disturbed or not, with an accuracy of 0. 88. Finally, by combining the above systems, continuous and regular motion trajectories are obtained, which proves the effectiveness and robustness of the systems.

Aiming at the privacy protection requirements of existing image datasets, a privacy-preserving scenario of image datasets and a privacy-preserving image alternative data generation method is proposed. The scenario is to replace the original image dataset with an alternative image dataset processed by a privacy-preserving method, where the substitute image is in one-to-one correspondence with the original image. And humans can not identify the category of the substitute image, the substitute image can be used to train existing deep learning images classification algorithm, having a good classification effect. For this scenario, the data privacy protection method based on the PGD ( Project Gradient Descent) attack is improved, and the attack target of the original PGD attack is changed from the label to the image, that is the image-to-image attack. A robust model for image-to- image attacks as a method for generating alternative data. On the standard testset, the replaced CIFAR(Canadian Institute For Advanced Research 10)dataset and CINIC dataset achieved 87. 15% and 74. 04% test accuracy on the image classification task. Experimental results show that the method is able to generate an alternative dataset to the original dataset while guaranteeing the privacy of the alternative dataset to humans, and guarantees the classification performance of existing methods on this dataset. 

In order to improve the recognition performance and accuracy of traditional object detection and to accelerate the computation speed, a CNN( Convolutional Neural Network) model with more powerful feature learning and representation capabilities and with related deep learning training algorithms is adopted and applied to large-scale recognition tasks in the field of computer vision. The characteristics of traditional object detection algorithms, such as the V-J(Viola-Jones) detector, HOG(Histogram of Oriented Gradients) features combined with SVM( Support Vector Machine) classifier, and DPM ( Deformable Parts Model) detector are analyzed. Subsequently, the deep learning algorithms that emerged after 2013, such as the RCNN ( Region-based Convolutional Neural Networks) algorithm and YOLO(You Only Look Once) algorithm are introduced, and their application status in object detection tasks is analyzed. To detect fallen individuals, the YOLOv5(You Only Look Once version 5) model is used to train the behavior of individuals with different heights and body types. By using evaluation metrics such as IoU(Intersection over Union), Precision, Recall, and PR curves, the YOLOv5 model is analyzed and evaluated for its performance in detecting both standing and fallen activities. In addition, by pre- training and data augmentation, the number of training samples is increased, and the recognition accuracy of the network is improved. The experimental results show that the recognition rate of fallen individuals reaches 86% . The achievements of this study will be applied to the design of disaster detection and rescue robots, assisting in the identification and classification of injured individuals who have fallen, and improving the efficiency of disaster area rescue.

The connection between mobile internet devices based on the Android platform and people’s lives is becoming increasingly close, and the security issues of mobile devices have become a major research hotspot. Currently, many visual Android malware detection methods based on convolutional neural networks have been proposed and have shown good performance. In order to better utilize deep learning frameworks to prevent malicious software attacks on the Android platform, a new application visualization method is proposed, which to some extent compensates for the information loss problem caused by traditional sampling methods. In order to obtain more accurate software representation vectors, this study uses the Swin Transformer architecture instead of the traditional CNN(Convolutional Neural Network) architecture as the backbone network for feature extraction. The samples used in the research experiment are from the Drebin and CICCalDroid 2020 datasets. The research experimental results show that the proposed visualization method is superior to traditional visualization methods, and the detection system can achieve an accuracy of 97. 39% , with a high ability to identify malicious software.

 In order to improve the invulnerability and working ability of the satellite constellation network after being attacked, a simulated annealing wolf pack algorithm is proposed. We use the subjective and objective weight method combined with the TOPSIS( Technique for Order Preference by Similarity) to Ideal Solution to evaluate the importance of nodes in the network, and attack the network according to the order of node importance. The network connection efficiency is the optimization goal, and the satellite constellation network communication limitation is the constraint condition. The idea of motion operator is adopted to realize the walking, summoning and sieging of wolves with adaptive step size. The network structure is optimized using the edge-adding scheme obtained through optimization. Experiments show that compared with other optimization algorithms, this algorithm has superiority. It solves the problem that the satellite constellation network爷s working ability declines after being attacked, and improves its invulnerability after being attacked. Key words: satellite network; invulnerability optimization; simulated annealing algorithm; improved wolf colony algorithm

 The design of the loop filter in noise-shaping SAR ADC(Successive Approximation Register Analogto Digital Converter) is the key to the effect of noise shaping and is also an important module to achieve high accuracy performance. Compared with the active lossless integral loop filter, the traditional passive lossy integral loop filter has the characteristics of low power consumption and simple circuit design, but its NTF(Noise Transfer Function) is smooth and the noise shaping effect is weak. To solve this problem, a passive lossless second-order integral loop filter is proposed, which retains the advantages of the passive lossy integral loop filter and has a good noise shaping effect. A hybrid architecture noise-shaping SAR ADC with a resolution of 16 bits and a sampling rate of 2 Ms/ s is also designed. The simulation results show that high SNDR( Signal to Noise and Distortion Ratio) (91. 1 dB), high accuracy ( 14. 84 bits), and low power consumption ( 285 uW) are achieved when the bandwidth is 125 kHz and the oversampling ratio is 8.

To solve the problem that Q function overestimation may cause SAC ( Soft Actor Critic) algorithm trapped in local optimal solution, SCSAC ( Stable Constrained Soft Actor Critic) algorithm is proposed for perfectly resolving the above weakness hidden in maximum entropy objective function improving the stability of Stable Constrained Soft Actor Critic algorithm in trailing process. The result of evaluating Stable Constrained Soft Actor Critic algorithm on the suite of OpenAI Gym Mujoco environments shows less Q value overestimation appearance and more stable results in trailing process comparing with SAC algorithm.

The wide application of the Internet is likely to lead to various kinds of privacy data leakage. In order to solve the problems, a cloud computing down-centric dual differential privacy data protection algorithm is proposed. First, the purpose of accurate collection of private data is achieved by learning the network model of private data transmission channel, and then the method of reconstructing the spatial characteristics of private data is used to obtain the ontological characteristics of private data. Finally, the collected private data is accurately noised through the characteristics of private data to achieve the purpose of accurate protection of private data, and the decentralized dual differential privacy data protection is completed. The experimental results show that the proposed algorithm has high real-time and good security for privacy data protection, and can accurately protect privacy data in different noise environments.

 FBG(Fiber Bragg Grating) sensing technology has achieved rapid development in scientific research and engineering applications, but it is rarely used in undergraduate experimental teaching. Currently, there are few devices available in the market that can be directly used for FBG sensing experimental teaching, and cutting- edge scientific research technology is disconnected from undergraduate experimental teaching. To address this situation, a teaching experiment device for temperature stress sensing based on FBG has been designed. The device consists of three parts: a fiber laser, a spectrometer, and upper computer control software. The fiber laser enables laser output of about 1 550 nm. The spectrometer measures the change of FBG center wavelength and collects data into the computer. The upper computer control software is used for graphic display and data storage. The experimental device has the advantages of simple operation, flexible assembly, good repeatability, and stability, and can be used for undergraduate experimental teaching. We introduce cutting-edge science and technology into undergraduate experimental teaching, promote the integration of scientific research and experimental teaching, and realize the synchronous improvement of scientific research and teaching levels. 

To address the issue of frequent faults in direct current electric vehicle charging piles and the difficulty of precise diagnosis, a fault diagnosis method based on an improved BP(Back Propagation) neural network is proposed. Firstly, the operation data set of the charging pile is preprocessed, such as normalization and filling in missing values, and the processed data set is input into the BP model for training. Secondly, an optimization method based on the BOA-SSA ( Butterfly Optimization Algorithm improved Sparrow Search Algorithm) is introduced to optimize the weights and thresholds of the BP model to obtain the optimal model. Finally, the fault status of the charging pile is diagnosed based on the optimized BP model. The simulation results show that the proposed BP method has good computational advantages in terms of MAE(Mean Absolute Error), MAPE(Mean Absolute Percentage Error), and RMSE(Root Mean Square Error). Compared to the traditional BP algorithm, the diagnostic accuracy of the improved BP method has increased by 14. 85% , which can diagnose the state of the charging pile accurately, providing a strong guarantee for the fault diagnosis of electric vehicles.

 In order to solve the problem that Yolov3-Tiny algorithm has insufficient feature extraction in gas station monitoring scene detection, which results in low detection accuracy, a new target detection algorithm based on gas station scene is proposed. This method first introduces Mosaic data enhancement algorithm to make the picture contain more feature information. Secondly, InceptionV2 and PSConv ( Poly-Scale Convolution) multiscale feature extraction methods are used to improve the network multiscale prediction ability. Finally, combined with the scSE(Concurrent Spatial and Channel ‘ Squeeze & Excitation’) attention mechanism, the output characteristics of the backbone network are reconstructed. The experimental results show that the algorithm has high detection accuracy and the detection speed meets the actual needs. The performance of the optimized algorithm is greatly improved and can it be applied to other target detection. 

A novel approach is proposed to address the issues of high accuracy but slow speed or low accuracy but appropriate training speed in traditional image recognition methods for fault diagnosis of oil pumps. The proposed method is based on an enhanced residual network model, with several improvement strategies. Firstly, the first-layer convolution kernel of the model is replaced with a smaller one. Secondly, the order of residual modules is changed. Thirdly, the fully connected layer of ResNet50( a Residual Network model) is replaced with an RBF( Radial Basis Function) network as an additional classifier. Finally, data augmentation techniques are used to expand the dataset, and transfer learning is utilized to obtain pre-trained weight parameters on ImageNet for the improved ResNet50-RBF model. Experimental results demonstrate that the proposed model achieves 98. 86% accuracy in pump curve recognition, exhibiting stronger robustness and improved speed compared to other networks. This provides some reference for fault diagnosis of oil pumps. The proposed method can significantly enhance the efficiency and accuracy of image recognition in fault diagnosis for oil pumps, which is of great significance for practical applications in the industry.

As one of the important components of the detection system, the performance of photoelectric detector is directly related to the quality of system data acquisition. In order not to affect the final detection result, it is essential to ensure the detector performance. The performance of high performance PtS₂ / MoTe₂ heterojunction infrared photodetector is studied. First, the materials, reagents and equipment are prepared to make PtS₂ / MoTe₂ heterojunction infrared photodetectors. The detector performance test environment, the four indicators of light response, detection rate, response time and photoconductivity gain are set up, and the detector performance is analyzed. The results show that the optical responsivity of PtS₂ / MoTe₂ heterojunction infrared photodetector is always above the 5 A/ W limit with the passage of test time. The detection rate of the detector is greater than 10 cm·Hz1 / 2 W ^-1 regardless of the infrared light reflected from any material. Whether the photocurrent is in the rising time or the falling time, its response time is always below the limit of 150 μs; The photoconductivity gain value has been kept above 80% . 

The ATMADDPG ( Attention Mechanism based Multi-Agent Deep Deterministic Policy Gradient) algorithm is proposed to improve the navigation ability of a multi-unmanned ship formation system. In the training phase, the algorithm trains the best strategy through a large number of experiments, and directly uses the trained best strategy to obtain the best formation path in the experimental phase. The simulation experiment uses four ' Baichuan' unmanned ships as experimental objects. The experimental results show that the formation maintenance strategy based on the ATMADDPG algorithm can achieve stable navigation of multiple unmanned ship formations and meet the requirements of formation maintenance to some extent. Compared to the MADDPG (Multi-Agent Depth Deterministic Policy Gradient ) algorithm, the developed ATMADDPG algorithm shows superior performance in terms of convergence speed, formation maintenance ability, and adaptability to environmental changes. The comprehensive navigation efficiency can be improved by about 80% , which has great application potential.

Considering that the randomness of the new power load is enhanced, the traditional accurate forecasting methods can not meet the requirements, an EMD-BiLSTM-ANFIS (Empirical Mode Decomposition Bi-directional Long Short Term Memory Adaptive Network is proposed based Fuzzy Inference System) quantile method to predict the load probability density. It replaces the accurate value of point prediction with the load prediction interval, which can provide more data for power System analysis and decision-making, The reliability of prediction is enhanced. First, the original load sequence is decomposed into several components by EMD, and then divided into three types of components by calculating the sample entropy. Then, the reconstructed three types of components and the characteristics of external factors screened by correlation. And they are used together with the Bilstm and ANFIS models for prediction training and QR(Quantile Regression), and accumulate the results of the prediction interval of the components to obtain the prediction interval of the final load. Finally, the kernel density estimation is used to output the user load probability density prediction results at any time. The validity of this method is proved by comparing the point prediction and interval prediction results with CNN- BiLSTM(Convolutional Neural Network-Bidirectional Long Short-Term Memory) and LSTM ( Long Short-Term Memory)models. 

Microseismic monitoring technology has been widely used in unconventional oil and gas development. The microseismic signal has weak energy and strong noise, which makes the follow-up work difficult and requires high-precision and accurate data. To solve the problem of extracting weak microseismic signals, an EM-KF (Expectation Maximization Kalman Filter)-based method is proposed for denoising microseismic signals. By establishing a state space model that conforms to the laws of microseismic signals and using the EM(Expectation Maximization) algorithm to obtain the optimal solution of the parameters for the Kalman filter, the signal-to-noise ratio of microseismic signals can be effectively improved while retaining the effective signals. The experimental results of synthetic data and real data show that this method has higher efficiency and better accuracy than traditional wavelet filtering and Kalman filtering.

In view of the problem of urban traffic congestion and traffic accidents caused by cabs stopping at will, it is very necessary to reasonably delineate the fixed parking areas for cabs. Using the cab GPS(Global Position System) data and crawled POI ( Point of Interest) data in the actual area of Chengdu, DBSCAN (Density-Based Spatial Clustering of Application with Noise) clustering algorithm is used to cluster the pick-up and drop-off points to get the hotspots of cabs, the types of hotspots are delineated according to the types of POIs, and the travel demand of cabs at different times is analyzed, so as to delineate the fixed parking area of cabs. The results of the study show that the setting of the fixed parking area of cabs is related to the travel demand of travelers, so that the fixed parking area is set in the area where the travel demand of travelers is high, which can satisfy the different travel demands of travelers. The method of combining cab passenger hotspots and crawling POI data to delineate fixed parking areas is highly practical and can provide theoretical and practical significance in urban transportation safety. 

 Aiming at the characteristic that the vehicle is more prone to instability in the snow and ice environment, the stable tracking problem of the vehicle to the reference trajectory under the low adhesion and uneven distribution condition of the road surface is studied. To address this, a fuzzy PID(Proportional-Integral- Differential) controller model based on neural network regulation and MPC ( Model Predictive Control ) a linearized vehicle model are designed. The controller takes the road adhesion coefficient and vehicle speed as input to construct a BP(Back-Propagation)neural network and outputs the adjustment coefficient to optimize the control performance of the PID controller. A ten-degree-of-freedom model is designed to characterize the dynamic characteristics of the vehicle in snow and ice-covered environments, and the lateral stability control of the vehicle is realized by using MPC. CarSim / Simulink is used for co-simulation experiments. Results show that the controller can significantly improve the performance of vehicle trajectory tracking. The dynamic characteristics of the vehicle under snow and ice are analyzed, and good simulation results are obtained.

Fault identification plays an important role in geological exploration, reservoir description, structural trap and well placement. Aiming at the problem that traditional coherence attribute and machine learning are poor in complex fault recognition, a fault recognition method based on UNet++ convolutional neural network is proposed. The weighted cross entropy loss function is used as the objective function to avoid the problem of data sample imbalance in the training process of the network model. Attention mechanism and dense convolution blocks are introduced, and more jump connections are introduced to better realize the feature fusion between the semantic information of deep faults and the spatial information of shallow faults. Furthermore, the UNet ++ network model can realize fault identification better. The experimental results show that the F1 value increased to 92. 38% and the loss decreased to 0. 012 0, which can better learn fault characteristic information. The model is applied to the identification of the XiNanZhuang fault. The results show that this method can accurately predict the fault location and improve the fault continuity. It is proved that the UNet ++ network model has certain research value in fault identification. 

fMRI ( functional Magnetic Resonance imaging) is an efficient research method for brain imaging technique. In order to reduce the redundancy of the fMRI data and transform the fMRI data to the constructed features with more classification potential, a feature construction method based on the siamese network named as SANet(Siamese Network) is proposed. It engineered the brain regions features under multiple scanning points of an fMRI image. The improved AlexNet is used for feature engineering, and the incremental feature selection strategy is used to find the best feature subset for the encephalopathy prediction task. The effects of three different network structures and four classifiers on the SANet model are evaluated for their prediction efficiencies, and the ablation experiment is conducted to verify the classification effect of the incremental feature selection algorithm on the SANet features. The experimental data shows that the SANet model can construct features from the fMRI data effectively, and improve the classification performance of original features.

In order to improve the approximate performance of CVI(Copula Variational Inference), the CHVI (Copula Hierarchical Variational Inference) method is proposed. The main idea of this method is to combine the Copula function in the CVI method with the special hierarchical variational structure of the HVM(Hierarchical Variational Model), so that the variational prior of the HVM obeys the Copula function in the CVI method. CHVI not only inherits the strong ability of the Copula function in CVI to capture the correlation of variables, but also inherits the advantage of the variational prior structure of HVM to obtain the dependencies of the hidden variables of the model, so that CHVI can better capture the relationship between hidden variables. correlation to improve the approximation accuracy. The author validates the CHVI method based on the classical Gaussian mixture model. The experimental results on synthetic datasets and practical application datasets show that the approximate accuracy of the CHVI method is greatly improved compared to the CVI method. 

Aiming at the problems of difficult modeling and low calculation efficiency of equivalent harness method, the effect of electromagnetic pulse radiation on the vehicle harness is studied using CST ( Computer Simulation Technology). The influence of the number of vehicle cables on the electromagnetic coupling effect of the harness is analyzed. By controlling the variables, we changed the number of cables in the harness and observed the maximum value of the coupling voltage in the harness. We also studied the maximum coupling voltage and current in the harness by varying the cable size and load resistance. The simulation results show that the peak value of the coupling voltage decreases linearly with an increase in the number of cables and increases linearly with an increase in cable size. The peak value of the coupling current decreases with an increase in load resistance, which follows a power series relationship. Finally, we combined the simulation results and fitted the maximum coupling voltage and current under different parameters, drawing a conclusion about the relationship between them, which provides a reference for the electromagnetic protection of vehicle wiring harnesses. 

Existing industrial time series data anomaly detection algorithms do not fully consider the temporal data on time dependence. An improved HTM(Hierarchical Temporal Memory)-Attention algorithm is proposed to address this problem. The algorithm combines the HTM algorithm with the attention mechanism to learn the temporal dependencies between data. It is validated on both univariate and multivariate time series data. By introducing the attention mechanism, the algorithm can focus on the important parts of the input data, further improving the efficiency and accuracy of anomaly detection. Experimental results show that the proposed algorithm can effectively detect various types of time series anomalies and has higher accuracy and lower running time than other commonly used unsupervised anomaly detection algorithms. This algorithm has great potential in the application of industrial time series data anomaly detection.

In order to solve the problem of communication mismatch between autopilot system using CAN (Controller Area Network)bus and navigation equipment using serial port communication, a communication conversion interface module based on Cortex-M3 is designed and the function of data conversion between serial port and CAN bus is realized. Aiming at the problems of poor signal stability and low baud rate accuracy of traditional CAN transceiver circuit CTM1050, an alternative hardware scheme is proposed and implemented to improve the timeliness and stability of data communication. Based on the CAN2. 0B extension frame, the internal CAN bus protocol of the autopilot system is designed to ensure the scalability and stability of the bus. The protocol can assign identity frames according to the priority of message information to ensure the orderly transmission of bus data. The actual test results indicate that the communication module is normal and the communication effect is good. The communication module has a certain universality and can be used in a variety of equipment systems. 

 In order to improve the quality of denoised images, an algorithm based on 2D-VMD ( Two Dimensional Variational Mode Decomposition ) and BD ( Bhattacharyya Distance ) is proposed for image denoising. Firstly, the algorithm uses 2D-VMD algorithm to decompose the image into several IMFs ( Intrinsic Mode Functions), and then BD is used to measure the geometric distance between the PDF (Probability Density Function) of each IMF and the original image to distinguish the signal-dominated IMF and the noise-dominated IMF. Finally, the denoising noise-dominated IMF through wavelet threshold denoising and the signal-dominated IMF are reconstructed to obtain the denoised image. The proposed algorithm is applied to medical images. The theoretical analysis and simulation result show that, compared with ROF ( Rudin Osher Fatemi) algorithm, median filter and wavelet threshold algorithm, the algorithm of combining 2D-VMD and BD has better denoising effect in both subjective and objective evaluation, and it effectively improves the quality of denoised images. 

To address the problem that RGB ( Red Green Blue ) modal and thermal modal information representations are inconsistent in form and feature information can not be effectively mined and fused, a new joint attention reinforcement network-FCNet ( Feature Sharpening and Cross-modal Feature Fusion Net ) is proposed. Firstly, the image feature mapping capability is enhanced by a two-dimensional attention mechanism. Then, a cross-modal feature fusion mechanism is used to capture the target region. Finally, a layer-by-layer decoding structure is used to eliminate background interference and optimize the detection target. The experimental results demonstrate that the improved algorithm has fewer parameters and shorter operation times, and the overall detection performance of the model is better than that of existing multimodal detection models.

BNNs( Binarized Neural Networks) are popular due to their extremely low memory requirements. While BNNs can be further compressed through pruning techniques, existing BNN pruning methods suffer from low pruning ratios, significant accuracy degradation, and reliance depending on fine-tuning after training. To overcome these limitations, a filter-level BNN pruning method is proposed based on evolution from ternary to binary, named ETB ( Evolution from Terry to Binary). ETB is learning-based, and by introducing trainable quantization thresholds into the quantization function of BNNs, it makes the weights and activation values gradually evolve from ternary to binary or zero, aiming to enable the network to automatically identify unimportant structures during training. And a pruning ratio adjustment algorithm is also designed to regulate the pruning rate of the network. After training, all zero filters and corresponding output channels can be directly pruned to obtain a simplified BNN without fine-tuning. To demonstrate the feasibility of the proposed method and the potential for improving BNN inference efficiency without sacrificing accuracy, experiments are conducted on CIFAR-10. ETB is pruned the VGG-Small model by 46. 3% , compressing the model size to 0. 34 MB, with an accuracy of 89. 97% . The ResNet-18 model is also pruned by 30. 01% , compressing the model size to 1. 33 MB, with an accuracy of 90. 79% . Compared with some existing BNN pruning methods in terms of accuracy and parameter quantity, ETB has certain advantages.

The optical crossbar chip is the core device used to realize optical routing in the field of optical communication. A control and driver system is designed based on a multi-channel DAC ( Digital to Analog Converter) array to achieve optical routing through the optical crossbar chip. The system consists of a control system module, a multi-channel drive circuit module, and a host computer control module. This system has several advantages, including simple adjustment, bipolar output, more output channels, and higher power accuracy. It solves the problems of the current driving circuit, such as complex operation, single power polarity, fewer output channels, and poor accuracy. The host computer control module can control the driving circuit to apply the control voltage and receive the optical power signal collected from the data acquisition device as the feedback signal of the control driving system. By analyzing the relationship between the control voltage and the received optical power, the best control driving voltage of the optical crossbar chip can be obtained. The system test results show that the system can provide high-precision bipolar driving voltage to effectively drive the optical crossbar chip and can calibrate the control voltage of the optical switch in a short time, fully meeting the requirements of the driving voltage in the active optical crossbar chip control. We believe that this system could be useful for optical crossbar chip control.

 Due to the influence of changes in lighting conditions and pedestrian height differences, there are large cross modal differences in surveillance video images at different times. In order to accurately identify pedestrians in cross modal images, a pedestrian recognition algorithm based on generalized transfer depth learning is proposed. The cross modal image is formed through Cyele GAN(Cycle Generative Adversarial Network), and the reference map is segmented using single object image processing to obtain candidate human body regions. The matching regions are searched in the matching map to obtain the disparity of human body regions, and the depth and perspective features of human body regions are extracted through the disparity. The attention mechanism and cross modal pedestrian recognition are combined to analyze the differences between the two types of images. The two subspaces are mapped to the same feature space. And the generalized migration depth learning algorithm is introduced to learn the loss function measurement, automatically screen the pedestrian features of the cross modal images, and finally complete pedestrian recognition through the modal fusion module to fuse the filtered features. The experimental results show that the proposed algorithm can quickly and accurately extract pedestrians from different modal images, and the recognition effect is good. 

Big data clustering processing has the problem of poor clustering effect and long clustering time for different types of data. Therefore, a big data clustering processing method based on the improved PSO-Means (Particle Swarm Optimization Means) clustering algorithm is proposed. The particle swarm optimization algorithm is used to determine the flight time and direction of unit particles during a cluster, preset the selection range of the initial cluster center, and appropriately adjust the inertia weight of unit particles. It eliminates the clustering defects caused by particle oscillation and successfully obtains the clustering center based on large-scale data. Combined with the spanning tree algorithm, the PSO algorithm is optimized from two aspects: sample skewness and centroid skewness. The optimized clustering center is then input into the k-means clustering algorithm to realize the clustering processing of big data. The experimental results show that the proposed method can effectively cluster different types of data, and the clustering time is only 0. 3 s, which verifies that the method has good clustering performance and clustering efficiency.

The obstacle avoidance control problem of UAVs( Unmanned Aerial Vehicles) and UGVs ( Unmanned Ground Vehicles) named heterogeneous unmanned swarm system is studied using a game-theoretic approach combined with the artificial potential field method. Unlike most multi-agent formation control schemes that only consider the group formation objective, we allow each agent to have individual objectives, such as individual tracking and obstacle avoidance. Specifically, when the heterogeneous unmanned swarm system completes the cooperative formation, each agent needs to track the target point and avoid obstacles in real-time according to its own interests. The heterogeneous unmanned swarm system formation problem is transformed into a non- cooperative game problem between agents because there may be conflicts between the individual and group objectives of the agents. Real-time obstacle avoidance is realized by adding an obstacle avoidance term based on the artificial potential field function into the cost function. And the controller is designed based on the Nash equilibrium seeking strategy to achieve a balanced formation mode of individual and group objectives. Finally, the correctness of the theoretical results is verified through simulation experiments. The proposed method can enable heterogeneous unmanned swarm system to achieve formation motion and real-time obstacle avoidance.

To achieve optimal efficiency and constant voltage output when the electric vehicle is charged wirelessly even after the load resistance value is changed, a synchronous Sepic converter is connected on the load side to identify different load resistance values, and impedance matching is performed by changing the duty cycle to achieve optimal transmission efficiency. The phase shift angle of therectifier is closed-loop controlled using a phase shift full bridge to achieve constant voltage output. Finally, simulation experiments using Matlab / Simulink software demonstrates the feasibility of this impedance-matching method and closed-loop control scheme.

In order to improve the problem of large trajectory tracking control error and low control accuracy of ball and plate system, a differential flatness-based backstepping sliding mode control method is proposed for enhancing the tracking accuracy in ball and plate system. Firstly, based on the Euler-Lagrange equation, the kinematic model of ball and plate system is established, the decoupled linear state-space model is obtained by reasonable simplification. Taking the X-direction controller design as an example, the target state quantity and feedforward control quantity of the system are obtained by using differential flatness technology, and an error system is constructed. Then, the backstep method is used to realize the sliding mode control of the error system, and the stability of the closed-loop system is proved by Lyapunov stability theory. The hyperbolic tangent function is used in the algorithm to suppress the jitter of the sliding mode, and the trajectory tracking control of ball and plate system is realized with high accuracy. Simulation results show that the proposed control strategy has high control accuracy and better control performance.

Traditional ROV(Remote Operated Vehicle) attitude control methods, operated underwater by ROVs, have suffered from chattering and poor stability. A new cooperative control law is proposed and ROV attitude controlleris designed based on thrust vector distribution. Firstly, the ROV kinematics model and dynamics model are established, and the thrust vector distribution model and decoupling dynamics model are carried out. Then, a new cooperative control law is proposed. By constructing appropriate macro variables, the macro variables converge exponentially to provide continuous control rate for the ROV attitude control system and eliminate chattering. Finally, a new cooperative control law is used to design the ROV attitude controller based on thrust vector allocation. The results of Matlab / Simulink simulation show that the proposed new cooperative control law can improve the control accuracy and stability of the ROV attitude control system. The control strategy provides a new feasible scheme for ROV attitude control. 

Aiming at the problem that it is difficult to extract and recognize the dynamic features of facial expression due to local occlusion, a dynamic recognition algorithm of facial expression with local occlusion based on deep learning is proposed, a deep belief network model is established, taking the output value of the previous layer as the input value of the next layer, a feature stacking unit is designed, the distribution of state variables of neurons in the visible layer, and the state variables of hidden neurons are calculated by taking the state value of the visible layer as the input value of the hidden layer according to the dynamic correlation of facial features. The recognition process is divided into two steps: training and forward propagation. The feature change rule is output. In the forward propagation process, the pixel point that conforms to the rule change is found, and the weight of the pixel point is solved. And as a loss function standard, the recognition weight of multiple positions on the face is used to constrain the recognition rate, and the dynamic recognition of facial partial occlusion expression is completed. Experimental data show that the proposed method can reduce image distortion and detail loss, improve image resolution, and achieve high recognition rate. It can complete efficient recognition for different local occlusion situations.

In order to maintain the consistency of parameters such as current, voltage, frequency, and phase during wind power grid connection, and to improve the safety and stability of wind power grid connection, a method for identifying control parameters of wind power grid connected converters based on optimization algorithms is proposed. The control model is established for the wind power grid connected converter, and the power control command of the PI(Proportional Integral) regulator is changed based on the power voltage support. Using differential function equations, to set the control conditions of the PI regulator, to calculate the functional relationship in the complex frequency domain, and to clarify the logical relationship between the adjustment integral coefficients. The identifiability of the control parameters is obtained through the control transfer function, and the parameter control output value and characteristic data are analyzed. Finally, the optimization and identification results of the control parameters are completed. The experimental results show that the proposed method can complete the identification of control parameters in various environments, with small identification error and high accuracy.

To address the shortcomings of traditional model robustness evaluation methods, such as low description consistency and difficulty in obtaining accurate scene matching data, a new power artificial intelligence index algorithm model of multi scenario robustness evaluation method is proposed. The multi scene data is extracted, the disturbance range interval of multi scene data in local space is set, the interval movement distance of spatial range is controlled, and the data acquisition results of sample points within the interval range are predicted. The basic feature parameters of the algorithm model are input, the multiple scene data is selected to obtain distance range values while increasing the input parameter dimension, and the initial data evaluation operations are performed based on the selected values. Based on the characteristics of uncertain control objectives, conduct data foundation analysis to ensure that the system is in a stable state and maintains its dynamic characteristics. Effectively analyze the differences between different system parameters, construct a range of deviation values, judge the multi scenario characteristics of the algorithm model, and achieve data evaluation. The experimental results show that the multi scenario robustness evaluation method of the electric power artificial intelligence index algorithm model can effectively transform the coordinates of sampling points, ensure the invariance of multi scenario sampling point data images, overcome the problem of scene data rotation sensitivity, and improve response speed. Compared with traditional evaluation methods, the proposed evaluation method has strong advantages in interference robustness and affine deformation robustness. 

To ensure timely distribution of emergency power supplies, enable them to quickly restore power supply, and reduce economic losses, a planning method for the distribution path of emergency power supplies is proposed based on geographic information systems. Firstly, based on the Map X component in the GIS (Geographic Information System) geographic information system, a preprocessing model for geospatial data is constructed. Then, based on the processed data, a mathematical model and constraint conditions for distribution path planning are established. Finally, genetic algorithm, mountain climbing algorithm, and ant colony algorithm are integrated, and the mathematical model is iteratively operated to obtain the optimal distribution path. The experiment is based on a power equipment emergency. When the material demand is met, the delivery time of the planned path under normal and abnormal road conditions is reduced by 14 minutes and 30 minutes respectively, and the cost is reduced by 10. 9 yuan and 5. 09 yuan respectively. This proves that the designed planning method has significant superiority. 

Aiming at the problem that the current D2D ( Device-to-Device) communication applied in the Internet of Vehicles, is only considering the resource utilization rate or system traversal capacity of the D2D multiplexing mode, but does not include other D2D modes into the system, an algorithm that takes into account resource allocation and mode selection is proposed. The communication resource utilization rate is improved by the priority multiplexing mode, the cellular mode is adopted for the D2D vehicle pair (D-UE:Device-to-device UsErs) that does not meet the multiplexing mode, the C-UE (Cellular UsErs) is adopted to meet the basic requirements of D-UE, and different resource access modes are allocated to D-UE considering factors such as vehicle and cellular users, BS ( Base Station) distance and signal-to-noise ratio between D-UE. Theoretical calculation and simulation results show that the proposed resource allocation and mode selection scheme can effectively improve the system capacity and D-UE QoS (Quality of Service).

In the process of assembly and production, due to certain errors in geometric parameters, the linkage and joints will inevitably have slight differences, resulting in some errors when the robot operates. In order to reduce the influence of environment on robot motion accuracy, a design scheme of robot motion error compensation algorithm based on improved weight function distance is proposed. The twist angle is added before positioning the robot position to obtain the transformation matrix between the two coordinate systems of the robot. The absolute error of the robot motion positioning is calculated according to the linear calibration. The mathematical model of the robot distance error is established using the improved weight function, and preliminarily compensate the motion error. The deviation of the center point position and attitude of the robot end effector is calculated. The compensation problem is transformed into the robot motion optimization problem, and the objective function of the motion deviation optimization problem is obtained. The final compensation result is obtained through multiple iterations. The experimental results show that the error compensation effect of the proposed method is good, and the motion stability of the robot after center of gravity compensation is good. 

In order to solve the problems of low lip contour detection accuracy and poor visual speech synthesis effect, a Tibetan-driven visual speech synthesis algorithm based on audio matching is proposed. This algorithm extracts short-term energy and short-term zero-crossing rate from Tibetan-language-driven visual speech signal, establishes short-term autocorrelation function of speech signal, and extracts feature information in speech signal, so as to obtain the pitch track of Tibetan speech signal. Secondly, the temporal and spatial analysis model of lip is established to analyze the changing trend of lip contour in the pronunciation process, and the feature of lip contour is extracted by principal component analysis. Finally, the correlation between audio features and lip contour features is obtained through the input-output hidden Markov model, and Tibetan-driven visual speech is synthesized on the basis of audio matching. Experimental results show that the proposed method has high lip contour detection accuracy and good visual speech synthesis effect. 

 Most of the existing detection methods for concept evolution are essentially based on supervised learning and are often used to solve the problem that only one novel class appears in a period of time. However, they can not handle the task of a class disappearing and recurring in streaming data. To address the above problems, an adaptive detection method for concept evolution based on weakly supervised ensemble (AD_WE) is proposed. The weakly supervised ensemble strategy is used to construct an ensemble learner to make local predictions on the training samples in the data block. Similar data with strong cohesion in the feature space are detected and clustered using local density and relative distance. The similarity of the clustering results is then compared to detect novel class instances and distinguish between different novel classes. And a dynamic decay model is established according to the characteristics of data change over time. The vanished class is eliminated in time, and the recurring class is detected through similarity comparison. Experiments show that the proposed method can respond to concept evolution in a timely manner, effectively identify vanished classes and recurring classes, and improve the generalization performance of the learner.

Dynamic load balancing is an indispensable link to ensure the normal operation of digital trunking system, but it is easy to be disturbed by communication failures and other problems in the control process. Therefore, a dynamic load balancing algorithm for digital trunking based on Kent mapping is proposed. The virtual machine system based on cloud platform collects data information such as the number of node connections, response time, dynamic load of the digital cluster, and analyzes the load of the digital cluster system. Secondly, a resource utilization model of digital trunking is constructed, and the resource utilization of digital trunking is obtained by solving the model with the Grey Wolf algorithm based on Kent map. Finally, the resource utilization rate is input into the LQR( Linear Quadratic Regulator) control loop, and the dynamic load balancing of the digital cluster is realized by controlling the migration of the server. The experimental results show that the digital trunking processed by the proposed algorithm has short response time, large fitness value, and strong fault tolerance ability.

In order to address the problem that the number of connected devices in the OGIoT(Oil and Gas IoT) has increased dramatically, resulting in insufficient computing power of the edge nodes in the EC ( Edge Computing) system, and it is difficult to effectively identify the service collapse caused by malicious attacks from other edge nodes, an EMLDI(Efficient Machine Learning method for Improved Data Contamination Detection of Oil and Gas IoT algorithm) is proposed, which solves the problem of fluctuating and inaccurate results of edge nodes due to their poor robustness, data distortion or mild qualitative changes. The problem of large and inaccurate edge node results due to robustness of edge nodes and data distortion or mild qualitative changes is solved. The network is trained by adding GN(Gaussian Noise) to the expanded data set through randomly selected batch samples, which enables the network to have broader data fitting and prediction capabilities, and solves the problem of systemic collapse due to the difficulty of implementing correct operations at the edge nodes when the data is severely corrupted. The algorithm is able to identify noise contaminated and random label contaminated samples more effectively and the algorithm achieves the best results within the specified training batches.

Incomplete big data, due to its irregular structure, has a large amount of computation and low interpolation accuracy when interpolation misses values. Therefore, a missing value interpolation algorithm for incomplete big data based on spatial autoregressive model is proposed. Using a migration learning algorithm to filter out redundant data from the original data under dynamic weights, to distinguish abnormal data from normal data, and to extract incomplete data. Using least square regression to repair the incomplete data. The missing value interpolation is divided into three types, namely, first order spatial autoregressive model interpolation, spatial autoregressive model interpolation, and multiple interpolation. The repaired data is interpolated to the appropriate location according to the actual situation, implementing incomplete big data missing value interpolation. Experimental results show that the proposed method has good interpolation ability for missing values. 

To effectively reuse URLLC(Ultra-Reliable Low-Delay Communications) and eMBB(enhanced Mobile Broad Band) on the same carrier band and improve the performance of the hybrid traffic system, a dynamic bandwidth allocation strategy based on scalable transmission time interval is proposed. The system bandwidth is dynamically divided according to the traffic type. URLLC scheduling priority is promoted in time domain. And in the frequency domain, different lengths of TTI ( Transmission Time Interval) are adopted to carry out user- centered wireless resource allocation. The dynamic system-level simulation shows that, compared to the traditional wireless resource allocation algorithm, the proposed scheme can effectively meet the delay requirements of URLLC users and optimize the throughput consumption of eMBB users under different load levels. The maximum delay gain of URLLC users is 83. 8% . The quality of service for different types of traffic in the 5G hybrid traffic system is satisfied.

 In order to ensure data security and alleviate data storage, a distributed data fault-tolerant storage algorithm based on density partitioning is proposed. High-density data areas of distributed data are filtered, highly similar targets are divided into different areas, the density distribution of data is described through data source sample points, the data elasticity is set, probability and data granularity is used to calculate the corresponding storage gradient and intensity index, and data storage gradient and data elasticity is introduced into information storage to complete distributed data fault-tolerant storage. Experiments show that the proposed algorithm has high fault tolerance, stable bandwidth throughput, small average path length, and can improve the security of network data. 

Aiming at the problem of information loss, such as details of remote sensing images reconstructed by a super-resolution algorithm, in order to ensure that remote sensor reconstruction images contain more texture and high-frequency information, a remote-sensitive image super resolution network is proposed based on a pyramid- based attention mechanism and the generation of confrontational networks. Firstly, a new pyramidal dual attention module is designed, including channel attention network and spatial attention network. Pyramid pooling is used instead of average pooling and maximum pooling in the channel attention network structure to enhance the feature representation capability from the perspective of global and local information. The spatial attention network structure adopts large scale convolution to expand the integration capability of local information, which can effectively extract texture, high frequency and other information. Secondly, the dense multi-scale feature module is designed to extract feature information at different scales using asymmetric convolution, and the extraction accuracy of texture, high frequency and other information is enhanced by fusing multi-level scale features through dense connection. Experimental validation is performed on the publicly available NWPU- RESISC45 dataset, and the experimental analysis shows that the algorithm outperforms the comparison methods in both subjective visual effect and objective evaluation metrics, and the reconstruction performance is relatively good. 

In order to avoid more duplicate data in the encryption process of IoT privacy data, which leads to higher computational complexity and reduces computational efficiency and security, an encryption method of IoT privacy data that combines DES(Data Encryption Standard) and ECC(Ellipse Curve Ctyptography) algorithms is proposed. Firstly, the TF-IDF(Tem Frequency-Inverse Document Frequency) algorithm is used to extract feature vectors from the privacy data of the Internet of Things. They are input into the BP(Back Proragation) neural network and are trained. The IQPSO( Improved Quantum Particle Swarm Optimization) algorithm is used to optimize the neural network and complete the removal of duplicate data from the privacy data of the Internet of Things. Secondly, the Data Encryption Standard and ECC algorithm are used to implement the primary and secondary encryption of the privacy data of the Internet of Things. Finally, a fusion of DES and ECC algorithms is adopted for digital signature encryption to achieve complete encryption of IoT privacy data. The experimental results show that the proposed algorithm has high computational efficiency, security, and reliability.

Aiming at the many causes and difficult features of internal network security threat nodes, an internal network security threat detection method based on XGBoost algorithm is proposed. Using the state differences between the internal network communities as an indicator, the edge weights of the nodes within different community types are calculated to find the nodes associated with the target values. Eigenvalues extracted through multiple assignments are taken as the initial input value XGBoost decision tree to construct the threat feature objective function, solve the corresponding Taylor coefficient of each node, and realize internal network security threat detection. The experimental data show that the proposed method has high feature extraction accuracy and can achieve accurate detection under various network attack conditions.

 Due to its low background noise, solar-blind ultraviolet photodetection technology is widely used in fields such as fire monitoring, missile detection, and military communication. Compared to other solar-blind ultraviolet sensitive materials, amorphous gallium oxide offers several advantages, including a bandgap that matches the solar-blind ultraviolet region, structural stability, and good mechanical strength. The horizontal metal-semiconductor-metal structured photodetectors are known for their simple production processes, ease of integration, and suitability for industrialization. Given the non-uniform distribution of the internal electric field and the photo-generated carriers along the thickness direction in horizontal devices, the thickness of the photoreceptive layer plays a crucial role in the performance of the photodetectors. In order to fabricate high- performance solar-blind ultraviolet photodetectors, amorphous gallium oxide thin films were prepared using low- temperature metal organic chemical vapor deposition method. Structural characterization of the films confirmed their amorphous nature, and the film surfaces were found to be relatively flat, with the optical absorption edge located within the deep ultraviolet spectral range. Solar-blind ultraviolet photodetectors were subsequently developed. As the thickness of the photoreceptive layer increased from 33. 2 nm to 133. 6 nm, the dark-current of the photodetector rose from 2. 33*10^-10 A to 2. 12*10^-8 A, and the photo-current under 254 nm illumination increased from 1. 66 * 10^-7 A to 3. 2 * 10^-5 A. Additionally, both the responsivity and the external quantum efficiency of the photodetectors increased by orders of magnitude with the increase in the photoreceptive layer thickness, reaching maximum values of 2. 91 A/ W and 1 419. 12% , respectively. The thickness-dependent characteristics of the photodetectors can be attributed to the interfacial high-defect layers, light absorption intensity, and the geometric parameters of the photodetectors. The photodetectors exhibited excellent wavelength selectivity, the current of each photo-detector under 365 nm illumination and the photo-current under 254 nm illumination differ by more than two orders of magnitude. Moreover, over the tested 5 cycles, the response / recovery behavior of each photodetector consistently demonstrates good repeatability and stability.

To improve the efficiency of path planning and avoid ant colony algorithm outputting non optimal paths, a multi-agent path planning model is proposed. The grid method is used to establish the environment awareness model of agents, improving the local and global pheromone update rules in the ant colony algorithm, and constraining the ants to travel by adjusting the number of turns and pheromone concentration. The algorithm can intelligently enlarge or reduce the pheromone concentration in the path. When the number of iterations reaches the set maximum, the output value is the optimal path planning result. Experimental results have shown that the improved algorithm achieves shorter planning paths and faster iterative convergence speed.

In order to solve the problem that traditional density peak clustering method without considering similarity of bands in information theory and number of bands in band selection, an adaptive density peak band selection method based on spectral angle mapping and spectral information divergence (SSDPC: Spectral angle mapping and Spectral information divergence Density Peaks Cluster)is proposed. SSDPC combines spectral angle mapping and spectral information divergence for density peak clustering band selection in hyperspectral images, replacing the traditional Euclidean distance to construct a band similarity matrix. By constructing a band scoring strategy, an important subset of spectral bands can be selected automatically and effectively. Using RX(Reed- Xiaoli) algorithm for anomaly detection on three sets of hyper-spectral datasets, the accuracy of anomaly detection is 1. 16% ,1. 18% and 0. 07% higher than that of Euclidean distance measurement under the similarity measure of SSDPC. Under the adaptive SSDPC band selection method, the accuracy of anomaly detection is 6. 49% ,2. 71% and 0. 05% higher than that of the original RX algorithm, respectively. The experimental results show that the SSDPC is robust, can improve the performance of hyper-spectral image anomaly detection and reduce its false alarm rate.

To improve the efficiency of manual detection of illegal and irregular collection of personal information, an APP(Application) personal information evaluation system for illegal and irregular collection is developed based on techniques such as regular expression semantic analysis and machine learning. We conducted illegal and irregular detection on online apps, generated detection algorithms and rules, and focused on solving technical difficulties such as semi automated access to privacy policies, app detection engines, and dynamic sandboxes for custom ROM(Read Only Memory) . The developed prototype system is used to conduct regular technical testing on the apps listed on major application platforms. The testing results show that the system significantly improves the efficiency of comprehensive governance and judgment of illegal and irregular collection of personal information apps, and effectively supports the relevant work of higher-level management departments.

Approximate matching is an indispensable link in the normal use of multimodal data technology, but the process of approximate matching is vulnerable to data redundancy, heterogeneous components and other issues. Firstly, parallel wavelet algorithm is used to eliminate the noise in multimodal data to avoid the impact of noise on the matching process. Secondly, tensor decomposition clustering algorithm is used to divide the data with different similarity into different clusters to eliminate the data difference of different clusters. Finally, the preprocessed data is input into the data matching model based on spatial direction approximation, The approximate matching of multimodal data is completed by calculating the spatial direction approximation and editing the distance between the reference data and the data to be matched. The experimental results show that the proposed method has high matching precision, high recall and short matching time. 

For the uncertainty decision-making problem of privacy risk, based on the theories of intuitionistic fuzzy and set pair analysis, a set pair relationship of information weights is established for privacy certainty & uncertainty. The intuitionistic fuzzy set pair operator is provided, and the relevant concepts, operations, properties, expected values, size ranking, and several intuitionistic fuzzy set pair information aggregation operators are defined, including Intuitionistic fuzzy set pair analysis operators, intuitionistic fuzzy set pair analysis weighted average operators, intuitionistic fuzzy set pair analysis weighted geometric operators, intuitionistic fuzzy set pair analysis ordered weighted average operators, intuitionistic fuzzy set pair analysis ordered weighted geometric operators, intuitionistic fuzzy set pair analysis hybrid aggregation operators, intuitionistic fuzzy set pair analysis hybrid geometric operators and their related properties. On this basis, the intuitionistic fuzzy set pair information aggregation method for privacy risk multi-attribute decision-making is analyzed, and it shows that the proposed method has feasibility and rationality. 

In order to solve the problem of droplet transmission risk of breath alcohol detector, a portable and pollution-free blood alcohol concentration detection scheme based on near-infrared spectroscopy is proposed. we successively completed the signal acquisition, amplification, filtering and re-amplification with hardware, and obtained the standard pulse wave signal with software algorithms such as Kalman filtering and three-sample interpolation method to remove the base. Later, the mathematical model is established in the upper computer by the measurement results of the expiratory type and the amplitude of the pulse wave signal, and the mathematical model is written into the microcontroller to realize the offline measurement. The accuracy of the mathematical model is tested through validation experiments. The results show that it meets the accuracy requirement, the relative error is less than 10% and meets the practical requirements.

In order to solve the problem that using traditional technologies such as Java and .NET to develop and deploy data services are complex and difficult to integrate with advanced cloud and container technologies, a lightweight 3D visualization data service solution for drilling based on Web is proposed, providing data interface support for front-end visualization applications. Based on NodeJS、 Angular、 TypeScript and other open source lightweight technologies, a lightweight drilling database system is designed, which can be used as an auxiliary tool for front-line technical managers and providing the most concerned data items in the fastest way with high efficiency and practicability. With the data loading tool, drilling technicians can easily load data into the database, including surface and seismic slices, measurements, events and well logs of blocks. And the system provides a comprehensive data security mechanism, including JWT ( JSON Web Token ) based identity authentication and JWE ( JSON Web Encripytion ) based data encryption, to ensure data security. The application results show that this solution can provide efficient data transmission services for drilling 3D visualization systems. 

Due to the complexity of digital information, massive and multi-angle unstructured big data, and external interference, data structure damage and other factors cause its information loss, a missing value interpolation algorithm for unstructured big data based on transfer learning is proposed. Through the migration learning algorithm, the missing parts of unstructured big data are predicted, and the naive Bayesian algorithm is used to classify data features, to measure the weight value between attributes, to clarify the feature difference vector of data categories, and to identify the degree of feature difference. The kernel regression model is used to implement nonlinear mapping for the missing part of the data, and the polynomial change coding is used to describe the cross-space complementary condition of the data, completing the interpolation of the missing value of unstructured big data. The experimental results show that the proposed algorithm can effectively complete the interpolation of missing values of unstructured large data, has good interpolation effect and can improve the interpolation accuracy.

To provide more efficient solutions for handling high-dimensional matrices and applying SVD(Singular Value Decomposition) in the context of big data, with the aim of accelerating data analysis and processing, how to quickly calculate the eigenvalues and eigenvectors ( singular value singular vectors) of high-dimensional matrices is studied. By studying random projection and Krylov subspace projection theory, six efficient calculation methods are summarized, making comparative analysis and related application research. Then, the six algorithms are applied, and the algorithms in related fields are improved. In the application of spectral clustering, the algorithm reduces the complexity of the core step SVD( Singular Value Decomposition), so that the optimized algorithm has similar accuracy to the original spectral clustering algorithm, but significantly shortens the running time. The calculation speed is more than 10 times faster than the original algorithm. When this work is applied in the field of image compression, it effectively improves the operation efficiency of the original algorithm. Under the condition of constant accuracy, the operation efficiency is improved by 1 ~ 5 times.

The low-light level digital sighting telescope encounters strong stray light interference, which causes imaging overexposure and submerges useful information in the image. To address this issue, a set of strong stray light suppression technology solutions is proposed. First, absorbance flannelette is pasted to the inner surface of the objective lens, and then several algorithms such as cumulative integration of adjacent images, histogram statistics, and wide dynamic gray enhancement are used in software image processing to suppress strong stray light. In outdoor environments with night sky illumination below 1 伊10 -3 lx, the experiment is conducted with added strong stray light interference. The results show that the technical solution can effectively suppress strong stray light and enhance image details, thereby improving image quality. The software runs based on FPGA(Field Programmable Gate Array), with a maximum processing time of 2 ms, meeting the real-time requirements of the system.

To improve the stability and accuracy of satellite electromagnetic detection target tracking, a twin network based satellite electromagnetic detection target tracking algorithm is proposed to avoid the tedious target acquisition process. Firstly, a multi-satellite scheduling model is established for electromagnetic detection satellites, matching suitable satellites and working modes for electromagnetic detection targets, in order to complete the collection of target electromagnetic signals. Secondly, a twin network is used to train the target signal, obtaining the electromagnetic feature information and true position information of the target by eliminating interfering clutter in the target signal. Finally, a particle filter algorithm is used to achieve stable tracking of satellite electromagnetic detection targets. The test results show that the proposed algorithm can effectively improve the efficiency of target tracking, and has high stability and accuracy.

A node deployment and scheduling algorithm based on fitness function and zero tolerance coverage is proposed to solve the problems of sensor blind area and poor connectivity between sensor nodes in wireless sensor network coverage. The network coverage is considered as a two-dimensional plane, the relationship between the maximum coverage range of node sensing and the distance value is analyzed to obtain the attribute values of the target points with hot spot distribution and overlapping coverage. Then, according to the deployment indicators such as wireless sensor target point coverage, connectivity and candidate locations, the fitness function is used to calculate the optimal deployment relationship of indicators, and to obtain the redundant parameters of nodes. The redundant complementary nodes are found within the same sensing range to achieve replacement scheduling. The experimental results show that the algorithm performs well in terms of network coverage and scheduling effectiveness, and has strong comprehensive performance.

 In response to the chattering phenomenon that traditional SMO(Sliding Mode Observer) faces during switching functions, novel sliding mode observer using saturation function instead of switching function is proposed to weaken chattering. In the process of extracting position information, a phase-locked loop is selected to replace the traditional arctangent method, there by improving the observation accuracy of PMSM(Permanent Magnet Synchronous Motor) rotor position. In the Matlab environment, by comparing traditional SMO and new SMO, it can be observed that the speed error of the rotor has increased by about 14 r/ min, and the position error of the rotor has increased by about 0. 03 rad.

To address the problem of significantly degraded image quality in hazy weather, a new multi-task learning method is proposed based on the classical atmospheric scattering model. This method aims to jointly learn the transmission map and dehazed image in an end-to-end manner. The network framework is built upon a new biconditional generative adversarial network, which consists of two improved CGANs( Conditional Generative Adversarial Network). The hazy image is inputted into the first stage CGAN to estimate the transmission map. Then, the predicted transmission map and the hazy image are passed into the second stage CGAN, which generates the corresponding dehazed image. To improve the color distortion and edge blurring in the output image, a joint loss function is designed to enhance the quality of image transformation. By conducting qualitative and quantitative experiments on synthetic and real datasets, and comparing with various dehazing methods, the results demonstrate that the dehazed images produced by this method exhibit better visual effects. The structural similarity index is measured at 0. 985, and the peak signal-to-noise ratio value is 32. 880 dB.

A control method based on the combination of the CSA ( Cuckoo Search Algorithm) and the conductivity incremental method INC( Incremental Conductivity method) is proposed to improve the speed and accuracy of the maximum power point tracking as well as to reduce the loss and harmonic content of the output power of the PV power generation system when the PV array is locally shaded. To prevent the algorithm from settling on the local optimal solution in the early stages, the cuckoo algorithm is used for global search. Later, a thorough search within a limited range is carried out using the incremental conductivity method in order to lock the maximum power point. And to see if it satisfies the criteria for grid connected harmonic content, this algorithm applied to grid connected control. Then a different strategy suggested. The results of a simulation model created in Matlab / Simulink demonstrated that the composite algorithm based on CSA paired with the conductivity increment approach has a faster tracking speed, less error, and satisfies the grid connected harmonic content requirements.

Unlike single noise, mixed noise has inconsistent characteristics and is difficult to suppress. In order to improve the noise suppression effect and image clarity, a digital image mixed noise suppression algorithm based on lifting wavelet is proposed. By using probabilistic neural networks, digital image noise is divided into pulse noise and Gaussian noise. The median filtering method is used to remove pulse noise from the digital image, and the lifting wavelet method is used to remove Gaussian noise from the digital image, achieving mixed noise suppression. The experimental results show that the proposed algorithm achieves higher image clarity and signal-to-noise ratio, and significantly improves the ENOB( Effective Number Of Bits) value of the digital image after denoising, indicating that the hybrid noise suppression effect of the algorithm is better.

Under the current situation of normalized management of epidemic situation, in order to ensure the normal development of computer experiment courses in colleges and universities, a virtual laboratory for computer basic experiments is established based on Docker technology. Students can access the server through a browser to obtain an independent experimental environment. The Docker-Compose tool is used to create, open, stop, delete and other multi-dimensional management of students' experimental environments, and to ensure their performance, which is equivalent to moving the offline laboratories online. This scheme can meet the needs of online computer basic experiments and provide high-quality experimental services for corresponding theoretical teaching.

To increase the coverage of MIoTs(Maritime Internet-of-Things) devices, a coverage Optimization algorithm based on Deployment of MEC-UAV(UMCO: MEC-UAV-based Coverage Optimization algorithm) is proposed. In UMCO, MEC(Mobile Edge Computing) empowered UAVs(Unmanned Aerial Vehicles) is used to meet the network coverage demand for MIoT, and to maximize the network profit. We formulate a problem of joint MEC-UAVs deployment and their association with MIoT devices as an ILP(Integer Linear Programming) to maximize the network profit. An iterative algorithm is developed based on the Bender decomposition to solve the ILP. Finally, numerical results demonstrate that the proposed UMCO algorithm achieves a near-optimal solution.

In the process of virtual network layout, there are a lot of repetitive features and features with less correlation, which affect the efficiency of its layout. Therefore, a hierarchical layout algorithm of virtual network clustering features under big data redundancy technology is proposed. A weighted undirected graph is used to establish a virtual network graph, and the community structure of the virtual network is divided by communities, so that the clustering characteristics of the virtual network are eliminated to the maximum extent under the premise of keeping the original characteristics unchanged, and the characteristics with high correlation are obtained. According to the repulsion of Coulomb force, the distance between communities is increased, and the distance between network nodes and central points is reduced by the gravity of Hooke's law. Combined with FR (Flecher-Reeves) algorithm, the relationship between repulsion and gravity of virtual network clustering feature layer nodes is adjusted, and the hierarchical layout algorithm is realized. The experimental results show that the proposed algorithm can more clearly show the internal structure characteristics of each community, and the layout time is the shortest. 

Aiming to the problem that the original manifold learning algorithm only utilizes spectral characteristics without incorporating spatial information, a locality preserving projections algorithm based on spatial-spectral (SS-LPP: Spatial-Spectral Locality Preserving Projections) union is proposed. Firstly, the weighted mean filtering algorithm is used to filter the dataset, fuse the spatial information with the spectral information, and eliminate the interference of noise, to increase the smoothness of similar data. Then, the label set is used to construct intra-graph and inter-graph. Through the intra-graph and inter-graph, identification features can be effectively extracted, and the classification performance can be improved. The effectiveness of the algorithm is verified on the Salinas dataset and the PaviaU dataset. Experimental results show that the algorithm can effectively extract data features and improve the accuracy of classification.

Rapidity and accuracy are most important in medical treatment. Traditional blood tests rely on experienced physicians, which leads to low efficiency and accuracy. For the first time, a comprehensive determination method based on image recognition technology, named P-HSV ( Perimeter-Hue, Saturation, Value), is proposed for microfluidic blood sample chips. Size and color are used for integrated interpretation of blood types. Size interpretation is based on the contour perimeter and number of agglutination clusters within the reaction chamber, while color interpretation is based on categorization of the color saturation (S: Saturation) to brightness ( V: Value ) ratio of agglutination clusters within the reaction chamber. The grade of blood agglutination reaction is synthetically determined by size and color results. In this method, machine vision is used to determine the grade of blood agglutination reaction, resulting in accurate and rapid blood type determination. This reduces the subjective judgment of artificial judgment, improving the detection speed and accuracy greatly.

Plaintext transmission is easily tampered with in distributed databases. To address the security risk, a multi-level control algorithm for secure access is proposed to distributed databases based on searchable encryption technology. The algorithm groups the authorized users according to the security level, and uses TF-IDF( Tem Frequency-Inverse Document Frequency) algorithm to calculate the weight of plaintext keywords, then uses AES (Advanced Encryption Standard) algorithm and round function to generate the key of the ciphertext, uses matrix function and inverse matrix function to encrypt the plaintext, and uploads the encryption results to the main server. And the Build Index algorithm is used to generate an index of ciphertext, and whether the user has access to ciphertext is reviewed based on the relevant attribute information of the user’s security level. After the review is passed, the user can issue a request for the number of ciphertext and keyword search. The server sends the ciphertext back to the user and decrypts it using a symmetric key method, achieving multi-level access control. The experimental results show that this method takes a short time in the encryption and decryption processes, and has good security access control performance.

So far, the international community has only proposed an assessment metric for the extension conciseness of knowledge graph, but has not provided a standardized assessment method and process. To address this issue, the assessment method of the extension conciseness of knowledge graph is studied and a new method to assess the extension conciseness of the Chinese English mixed knowledge graph is proposed. The formulas for grouping at the overall level and assessing the head entities, relations, and tail entities are proposed and defined. To enhance the accuracy of the evaluation, the sentence level assessment formula is also defined. Finally, the four formulas are combined to create an algorithm for assessing the extension conciseness of the knowledge graph. To verify the accuracy and performance of the proposed algorithm, the open data set OPEN KG( Knowledge Graph) is used to assess and compare the proposed algorithm with related algorithms. The results confirm that the proposed algorithm provides a certain guarantee for the accuracy and time efficiency of the conciseness assessment of the Chinese English mixed knowledge graph, and the overall performance of the proposed algorithm is better than that of the related algorithm. 

A dual-mode address hopping method is proposed to address security issues faced by the IoT(Internet of Things), such as resource scarcity and low obfuscation of traffic data. Address hopping diversity and unpredictability are enhanced through a dual-mode address selection algorithm, thereby solving the problem of limited address pool resources. Additionally, a dual-virtual address hopping method is introduced to enhance the obfuscation of data packets and reduce the correlation of network data. This method is demonstrated to be effective in reducing network data correlation, conserving IoT resources, increasing network address pool capacity, preventing data theft by attackers, and ensuring IoT security through simulation experiments conducted in an SD-IoT(Software Defined Internet of Things) environment.

With the gradual development of oil and gas exploration and development into deep and complex formations, the risks and rewards faced in the drilling process are increasing. The market increasingly needs software that can integrate monitoring and control to ensure safe and efficient drilling. Using managed pressure drilling technology can effectively control the pressure in the gas production process and make it more convenient to deal with equipment failure. It can realize real-time monitoring of fluid pressure and density changes in the production process, effectively reducing potential safety hazards, preventing the occurrence of downhole accidents, and providing a guarantee for the safe and stable production of oil and gas wells. The Managed Pressure Drilling Simulation and Control Software aims to obtain drilling-related information from logging, PWD ( Pressure While Drilling ), MWD ( Measure While Drilling ), pressure control, and other equipment. It establishes hydraulic models to calculate wellbore pressure, flow, and other parameters. This software adopts client / server architecture, which allows multiple clients to connect to a server simultaneously and synchronize data. The client data synchronization effect has been verified on-site, meeting the needs of single machine use and facilitating network connection. The results indicate that this software can accurately simulate and calculate various drilling parameters, ensuring safe and efficient drilling. The centralized analysis, processing, and remote control have created a good foundation for pressure control drilling, transitioning from on-site engineer processing mode to a data platform-based approach. This lays the foundation for interconnectivity between multiple on-site pressure control drilling equipment on one platform, effectively promoting the development of intelligent pressure control drilling.

To address the issue of source code plagiarism detection and the limitations of existing methods that require a large amount of training data and are restricted to specific languages, we propose a source code plagiarism detection method based on pre-trained Transformer language models, in combination with word embedding, similarity and classification models. The proposed method supports multiple programming languages and does not require any training samples labeled as plagiarism to achieve good detection performance. Experimental results show that the proposed method achieves state-of-the-art detection performance on multiple public datasets. In addition, for scenarios where only a few labeled plagiarism training samples can be obtained, this paper also proposes a method that combines supervised learning classification models to further improve detection performance. The method can be widely used in source code plagiarism detection scenarios where training data is scarce, computational resources are limited, and the programming languages are diverse.

A modified receiving coil derived from the BP( Bipolar Pad) coil structure is proposed to address the problem of mutual inductance dips and increased mutual inductance fluctuations in the process of wireless charging of dynamic electric vehicles using magnetically coupled resonant radio energy transmission technology during straight-line driving and turning. The modified coil solves the problem of reduced effective area of positive pair coupling caused by decoupling of conventional BP coils by means of double coils fitted to the inner and outer diameter of the track respectively, and Ansys / Maxwell software is used to carry out simulation to find out the reasonable design position and relative size of the compensation coil. The experimental data shows that the new receiver coil can suppress the mutual inductance fluctuation and enhance the mutual inductance value to a certain extent, in the process of turning and driving in a straight line. The peak mutual inductance fluctuation rate is 5. 1% , and the maximum mutual inductance reception is 28. 7% higher than that of the traditional BP coil.

Efficient and accurate water injection flow prediction can help oilfield departments formulate reasonable production plans, reduce the waste of resources, and improve the injection-production rate of the oilfield. RNN( Recurrent Neural Networks) in deep learning is often used for time series prediction, but it is difficult to extract features from historical series and can not highlight the impact of key information. Early information is also easy to lose when the time series is too long. A method of oilfield water injection flow prediction based on dual attention mechanism CNN ( Convolutional Neural Networks)-BiLSTM ( Bi-directional Long Short-Term Memory) is proposed. Taking the historical water injection data of the oilfield as the input, the CNN layer extracts the characteristics of the historical water injection data, and then enters the feature attention mechanism layer. The corresponding weights are given to the features by calculating the weight value. The key features are easier to get large weights, and then have an impact on the prediction results. The BiLSTM layer models the time series of data and introduces the time step attention mechanism. By selecting the key time step and highlighting the hidden state expression of the time step, the early hidden state will not disappear with time,

which can improve the prediction effect of the model for long time series, and finally complete the flow prediction. Taking public datasets and oilfield water injection data from a certain region in southern China as examples, and comparing them with MLP ( Multilayer Perceptron), GRU ( Gate Recurrent Unit), LSTM ( Long Short Term Memory), BiLSTM, CNN, it is proven that this method has higher accuracy in oilfield water injection flow prediction, can help oilfield formulate production plans, reduce resource waste, and improve injection recovery rate, and has certain practical engineering application value.

To ensure network security and timely control the security situation, a security threat quantification assessment algorithm is proposed for wireless networks based on susceptible, infected, and susceptible infected recovered models. Asset value, system vulnerability, and threat are selected as quantitative evaluation indicators. Value and vulnerability quantification values are obtained based on the security attributes of information assets and the agent detection values of host weaknesses, respectively. Based on the propagation characteristics of the virus, the SIR ( Susceptible Infected Recovered) model is improved, the propagation characteristics of the virus are analyzed. A quantitative evaluation algorithm for wireless network security threat situation is established based on the quantification of three indicators, and the obtained situation values is used to evaluate the network security situation. The test results show that the security threat situation values of the host and the entire wireless network evaluated by this method are highly fitted with the expected values, and the evaluation time is shorter. It can be seen that the proposed algorithm has good evaluation accuracy and real-time performance, which can provide effective data basis for network security analysis and provide reliable decision- making support to administrators in a timely manner.

To improve the recognition accuracy of occluded faces, an intelligent recognition method for occluded faces based on the improved Gabor algorithm is proposed. Firstly, the dynamic range of facial images is compressed and the anti sharpening mask filtering algorithm is selected for image enhancement processing. Secondly, Gabor filters are used to extract features from half faces with relatively complete information preservation and high brightness. Finally, the extracted Gabor features are input into an extreme learning machine to achieve intelligent recognition of occluded faces. The experimental results show that the proposed method has good processing performance for occluded facial images, and the processed facial image recognition has high accuracy and short recognition time.

To help readers quickly find the books they need and avoid overloading digital information, an intelligent recommendation algorithm for digital book resources based on tag similarity is proposed. Firstly, based on the entered user information in the digital library system, the user feature similarity and user interest similarity are obtained and regarded as comprehensive similarity indicators. Then, combined with the tag similarity index, the similarity nearest neighbors of the target user’s book resources are obtained. Finally, the tags of the book resources browsed by the user are put into a tag set, and the digital book resources that the target user likes are formed into a recommendation list through a hybrid recommendation method of user implicit behavior scoring and linear weighted fusion, and recommended to the target user. Experimental results show that the proposed algorithm performs better than traditional recommendation algorithms.

In response to the demand for simulation effects consistent with real hardware, a smooth transition method is proposed to enhance the imaging effects of the entire simulation system. By analyzing visual persistence effects and system imaging delays, a two-point third-order Hermite interpolation is used to handle smooth transition time and imaging color respectively. Through comparative experiments, the results demonstrate that this method can adaptively smooth the imaging of the entire simulation system, thereby having solved issues such as real-time dynamic imaging flicker and instability. The significance of this method lies in enhancing the imaging quality of the virtual 3D simulation experimental system for embedded microcontrollers, improving the visual effects of embedded microcontroller 3D simulation, and mitigating the impact of problems such as abrupt changes and artifacts. The significant application value of a virtual 3D simulation experimental system for embedded microcontrollers in the fields of education and design is presented.

Traditional manual repair of fishing nets is inefficient and costly. To solve the problem of traditional manual repair of fishing nets, a robotic arm that can automatically sew fishing nets is designed. The robotic arm is mainly controlled by STM32F103C8T6, and its functions are achieved through infrared sensor module, motor actuator module, and communication module. The infrared sensor module is responsible for detecting the position of damaged fishing nets. The motor actuator module drives the motors in six axis to flexibly send the sewing device to the preset node, and the communication module is responsible for transmitting the detected information of damaged fishing nets to the controller for processing. After practical testing, the designed robotic arm can achieve high-precision, fast, and effective automatic sewing of fishing nets. Compared with traditional manual repair methods, this automatic repair technology improves the efficiency and quality of fishing net repair, reduces labor costs, and has obvious advantages.

To extend the lifetime of oil and gas IoT( Internet of Things), a dual cluster head based oil and gas IoT routing algorithm is proposed. The algorithm fully considers the current residual energy of sensor nodes, historical average energy, distance between nodes and base stations, density of neighboring nodes and distance between nodes and energy harvesting sources in the cluster head election process, and elects dual cluster heads in the same cluster, while proposes a novel routing method to balance the energy consumption of cluster heads in the data transmission phase. A novel node working mode switching strategy is adopted along with the introduction of energy harvesting techniques. Simulation experiments show that the algorithm can balance the network energy consumption more effectively and extend the network lifetime compared with the traditional algorithm.

A football competition risk intelligent warning method based on an improved Copula model is proposed to address the issues of large errors between warning values and actual values, and multiple false alarms in football matches. Based on the fuzzy comprehensive evaluation matrix, the evaluation system for football competition risk indicators is determined. The indicator level status is classified, the Copula function is selected, and an improved Copula football competition risk intelligent warning method is constructed to accurately judge football competition risks and reduce risk losses. The experimental results show that the interference suppression of this method is high, maintained above 20 dB, and have high anti-interference ability. It can effectively suppress interference. This method also reduces the error between the warning value and the actual value, reduces the number of false alarms in the warning, and verifies the practicality and feasibility of this method.

In order to have a clear understanding of the exclusive access principle and mechanism of mutex on shared resources, based on the brief analysis of the meaning, application occasions, scheduling mechanism and key elements of mutex in real-time operating systems, the mbedOS mutex scheduling mechanism are theoretically analyzed. Takes the KL36 chip as an example the mbedOS mutex is realized and the scheduling process information of thread response mutex is output spontaneously based on the sequence diagram and printf method. And the real-time performance of mutex scheduling mechanism is analyzed. The analysis of mutex scheduling mechanism is helpful to further analyze other synchronization and communication methods of mbedOS, and can also provide reference for in-depth understanding of other real-time operating system synchronization and communication methods. 

Due to the large scale and variety of information data in the process of internet information security warning, the warning accuracy is low and the time is long. To improve the efficiency of early warning, a security warning for unknown access sources in mobile network privacy information databases is proposed. Principal component analysis method is used to reduce the dimensionality of information base data to reduce the difficulty of detection. The IMAP( Iterative Multivariate AutoRegressive Modelling and Prediction) algorithm is used to carry out data clustering processing, to extract discrete isolated data points, and complete the screening of unknown access source data in the information base. Unknown access source data is inputted into a support vector machine, a time window is used to transform the construction problem of the information base security warning model into a convex optimization problem of support vector machine learning. Security warning results are outputted, and globally optimize the construction parameters of the warning model are optimized to improve the warning output ability of the security warning model. The experimental results show that the proposed method has high security detection efficiency for information databases, and can achieve stable and accurate warning output in the face of multiple types of information database intrusion attacks.

In order to improve the quality and accuracy of residual oil edge detection, a method of residual oil edge detection with improved multi-direction Sobel operator is proposed. The improved bilateral filtering method is used to remove the image noise of the microscopic residual oil distribution, so as to achieve the purpose of edge preserving and denoising. Combined with Otsu algorithm, the optimal threshold of the residual oil image can be obtained adaptively. The amplitude and direction of the gradient of the remaining oil image are calculated by using the Sobel operator with the amplitude of 3×3 in the four improved directions. Non-maximum suppression algorithm is used to filter out the pseudo-edge pixels to obtain the final remaining oil edge detection image. The experimental results show that the proposed method can accurately detect the information of residual oil edge in the microscopic residual oil distribution image while removing the noise of the image.

We present an improved target detection algorithm based on Yolov3-Tiny for gas station scene because of the low accuracy of target detection algorithm in gas station scenes. This algorithm takes Yolov3-Tiny model as the basic network, innovates Mosaic image enhancement method proposed in Yolov4 algorithm for data preprocessing, uses dense connection modules to reconstruct the feature extraction network, and adds CBAM (Convolutional Block Attention Module) attention mechanism and Pyramid Pooling Module into the network, finally target detection in the gas station scene is realized. The experimental results show that the improved algorithm improves the overall mAP by 8. 2% compared with the original algorithm, and can be more effectively applied to gas station target detection.

In order to reduce the access risk of medical information shared by multiple end users, the compulsory access control system for medical information sharing is optimized and designed from two aspects of database and software functions. The database tables of medical shared information and users are built, connecting the database tables according to the logical relationship, and completing the design of the system database. The process of medical information sharing is simulated and the sensitivity level of medical information sharing is determined. Multi-terminal user roles and permissions are allocated, abnormal access behaviors of access users are detected in real time, and authorization and behavior detection results are combined to realize the mandatory access control function of medical information sharing of the system. The test results show that the access control error rate of the design system is reduced by about 24. 4% , and the access risk of medical shared information is significantly reduced under the control of the design system.

To ensure the security of cross domain data flow in the Internet of Things, a security detection algorithm for cross domain data flow sharing the same frequency in the Internet of Things is proposed. This method calculates the information entropy of the data set based on the data outlier characteristics, takes the data points with larger information entropy calculation results as the cluster center, and analyzes the data distribution characteristics through the calculation of the cluster center distance. The data distribution features are inputted into the BP( Back Propagation) neural network and combined with genetic learning algorithms to achieve deep mining of shared cross domain data on the same frequency. Wavelet analysis is used to segment effective signals and noise signals in the same frequency shared data, and introduce the wrcoef function achieving the reconstruction output of noise free signals. Based on the Markov chain state transition probability matrix, a detection model of Markov cross domain data flow security is established. By calculating the relative entropy difference value between the test sample and the standard sample, the security detection of cross domain data flow for the same frequency sharing of the Internet of Things is completed. The simulation results show that this method can effectively improve the efficiency of data flow security detection and achieve accurate perception of data flow trends across domains.

Fundus images often contain a large number of small blood vessels with significant noise interference and blurred boundaries, making segmentation challenging. To address these characteristics, a fundus image segmentation method called MDF _Net&CD ( Multi-Directional Features neural Network and Connectivity Detection) is proposed, based on multidirectional features and connectivity detection. A deep neural network model, MDF_Net( Multi-Directional Features neural Network), is designed to take different directional feature vectors of pixels as input. MDF_Net is used for the initial segmentation of the fundus images. A connectivity detection algorithm is proposed to revise the preliminary segmentation results of MDF _ Net, according to the geometric characteristics of blood vessels. In the public fundus image dataset, MDF_Net&CD is compared with recent representative segmentation methods. The experimental results show that MDF_Net&CD can effectively capture the detailed characteristics of pixels, and has a good segmentation effect on irregular, severely noisy, and blurred boundaries of small blood vessels. The evaluation indices are balanced, and the sensitivity, F1 score, and accuracy are better than other methods participating in the comparison.

During the retrieval process of imbalanced stream data, the performance of data retrieval decreases due to the presence of imbalance in the data stream and the susceptibility to differential and edge data. In order to reduce the impact of the above factors, an imbalanced stream data retrieval method based on multi similarity fuzzy C-means clustering is proposed. This method calculates the multiple similarities between imbalanced flow data, and uses fuzzy C-means algorithm to cluster data with different similarities. By constructing a octree retrieval model, the data after clustering is stored, encoded and judged to complete the retrieval of unbalanced stream data. The experimental results show that the retrieval time of the proposed method is less than 20 seconds, and the recall and precision rates remain above 80% , with high NDCG( Normalized Discounted Cumulative Gain) values.

In order to enhance the performance of FSS(Frequency Selective Surfaces) and precisely control the propagation characteristics of electromagnetic waves in the microwave frequency range to achieve reflection, transmission, or absorption of electromagnetic waves, a miniaturized FSS for the microwave frequency band is proposed. The unit cell size of the FSS is 0.024λ x 0.024λ, demonstrating excellent miniaturization performance. Within the range of 1 ~10 GHz, the FSS exhibits three passbands with exceptional polarization stability and angle stability, maintaining consistent operating frequencies and bandwidth, while exhibiting good transmission performance. This study on the miniaturized FSS serves as a basis for FSS analysis and provides insights for the design of miniaturized frequency selective surfaces.

The training actions of skiing athletes can directly reflect their level, but traditional methods for identifying and evaluating actions have shortcomings such as subjectivity and low accuracy. To achieve accurate analysis of skiing posture, a motion analysis algorithm based on improved OpenPose and YOLOv5(You Only Look Once version 5) is proposed to analyze athletes爷 movements. There are two main improvements. First, CSP-Darknet53(Cross Stage Paritial-Network 53) is used as the external network for OpenPose to reduce the dimension of the input image and extract the feature map. Then, the YOLOv5 algorithm is fused to optimize it. The key points of the human skeleton are extracted to form the human skeleton and compared with the standard action. According to the angle information, the loss function is added to the model to quantify the error between the actual detected action and the standard action. This model achieves accurate and real-time monitoring of athlete action evaluation in training scenarios and can complete preliminary action evaluation. The experimental results show that the detection and recognition accuracy reaches 95%, which can meet the needs of daily skiing training. 

Bridge sensor anomaly detection is a method based on sensor technology to monitor the status of bridge structure in real time. Its purpose is to discover the anomalies of the bridge structure in time and recognize them to prevent and avoid accidents. The author proposes an abnormal signal detection and identification method for bridge sensors based on deep learning technology, and by designing an abnormal data detection algorithm for bridge sensors based on the LSTM (Long Short-Term Memoy) network model, it can realize the effective detection of the abnormal data location of the bridge cable sensor, and the precision rate and recall rate of the abnormal data detection can reach 99. 8% and 95. 3%, respectively. By combining the deep learning network and the actual working situation of bridge sensors, we design the abnormal classification algorithm of bridge cable-stayed force sensor based on CNN(Convolution Neural Networks) network model to realize the intelligent identification of 7 types of signals of bridge cable-stayed force sensor data, and the precision rate of identification of multiple abnormal data types and the recall rate can reach more than 90%. Compared with the current bridge sensor anomaly data detection and classification methods, the author's proposed method can realize the accurate detection of bridge sensor anomaly data and intelligent identification of anomaly types, which can provide a guarantee for the accuracy of bridge sensor monitoring data and the effectiveness of later performance index identification. 

 Due to the dense traffic flow and stochastic uncertainty of vehicle behaviors, the scenario of unsignalized intersection poses significant challenges for autonomous driving. An innovative approach for autonomous driving decision-making at unsignalized intersections is proposed based on the MAPPO(Multi-Agent Proximal Policy Optimization) algorithm. Applying the MetaDrive simulation platform to construct a multi-agent simulation environment, we design a reward function that comprehensively considers traffic regulations, safety including arriving safely and occurring collisions, and traffic efficiency considering the maximum and minimum speeds of vehicles at intersections, aiming to achieve safe and efficient autonomous driving decisions. Simulation experiments demonstrate that the proposed decision-making approach exhibits superior stability and convergence during training compared to other algorithms, showcasing higher success rates and safety levels across varying traffic densities. These findings underscore the significant potential of the autonomous driving decision-making solution for addressing challenges in unsignalized intersection environments, thereby advancing research in autonomous driving decision-making under complex road conditions.自动驾驶;智能决策;无信号灯交叉口;MAPPO算法 

In order to clearly understand the working principle and mechanism of events, to analyze the role, response principle, and response process of events in real-time operating systems, based on the KL36 microcontroller, a PC-like printf output method is adopted to analyze the event response mechanism of mbedOS from the aspects of scheduling process timing, response time performance, etc. The experimental results show that the printf function can intuitively output information such as thread address, queue address, queue content, thread in and out queue status, and event bits during the event response process. This provides convenience for readers to understand the event response principle and process of mbedOS from the bottom layer, and also provides a method reference for analyzing the context structure of other synchronization and communication methods of mbedOS.

 Under local shading conditions, the power-voltage characteristic curves of photovoltaic arrays show multiple peaks, and traditional population intelligence optimization suffers from slow convergence, large oscillation amplitude and the tendency to fall into local optimality. To address the above problems, an MPPT (Maximum Power Point Tracking)control method based on the PSO-GWO(Particle Swarm Optimization-Grey Wolf Optimization) algorithm is proposed. The algorithm introduces a convergence factor that varies with the cosine law to balance the global search and local search ability of the GWO algorithm; the PSO algorithm is introduced to improve the information exchange between individual grey wolves and their own experience. Simulation results show that the proposed PSO-GWO algorithm not only converges quickly under local shading conditions, but also has a smaller power output oscillation amplitude, effectively improving the maximum power tracking efficiency and accuracy of the PV(Photovoltaic) array under local shading conditions. 

To address issues such as excessive turning points and suboptimal paths in traditional JPS(Jump Point Search) algorithms, an improved jump point search algorithm is proposed. First, based on the feasibility of the map, the obstacles are adaptively expanded to ensure a safe distance. Then, an improved heuristic function based on directional factor is integrated. And a key point extraction strategy is proposed to optimize the initial planned path, significantly reducing the number of expanded nodes and turning points while ensuring the shortest path. The experimental results show that compared to traditional JPS algorithms, the proposed ensures a shorter path length and fewer corners, while reducing the number of extended nodes by an average of 19% and improving search speed by an average of 21. 8%. 

Due to the complex structure and features of current massive data, big data exhibits unstructured and small sample characteristics, making it difficult to ensure high accuracy and efficiency in its classification. Therefore, a big data classification optimization method is proposed to improve the decision tree algorithm. A fuzzy decision function is constructed to detect sequence features of big data, and these features inputted into a decision tree model to mine and train rules. The decision tree model is improved using grey wolf optimization algorithm. The big data is classified using the improved model, and then a classifier accuracy objective function is established to achieve accurate classification of big data. The experimental results show that the proposed method achieves the highest accuracy in classification results and the lowest false positive case rate, ensuring the overall high throughput of the algorithm and improving its classification efficiency.

Medical image classification is an important method to determine the illness of patients and give corresponding treatment advice. As medical image labeling requires relevant professional knowledge, it is difficult to obtain large-scale medical image classification labels. And the development of medical image classification based on deep learning method is limited to some extent. To solve this problem, self-supervised contrast learning is applied to medical image classification tasks in this paper. Contrast learning method is used in pre-training of medical image classification. The features are learned from unlabeled medical images in the pre-training stage to provide prior knowledge for subsequent medical image classification. Experimental results show that the proposed improved method of medical image classification based on self-supervised contrast learning enhances the classification performance of the ResNet. 

A fusion model based on material balance equation and Informer is proposed to solve the prediction problem of oilfield development indicators. Firstly, the mechanism model before and after the decline of oil field development production is established through the knowledge of the material balance equation field. Secondly, the established mechanism model is fused with the loss function of the Informer model as a constraint to establish an indicator prediction model that conforms to the physical laws of oil field development. Finally, the actual production data of the oil field is used for experimental analysis. The results indicate that compared to several purely data -driven cyclic structure prediction models, this fusion model has better prediction performance under the same data conditions. The mechanism constraints of this model can guide the training process of the model, so that its rate of convergence is faster, and the prediction at the peak and trough is more accurate. This fusion model has better predictive and generalization abilities, and has a certain degree of physical interpretability. 

This study aims to improve the intelligence and accuracy of fault diagnosis in oilfield wastewater systems. A composite neural network is constructed using convolutional neural networks and long short-term memory networks, and the structure is optimized using Adam and random gradient descent method to improve the convergence speed and fault diagnosis accuracy of the model. The study is validated through relevant experiments, and the experimental results show that the optimization algorithm used in the study improves the accuracy of the model to around 0. 87 and reduces the diagnostic loss rate of the model to around 0. 032. The average detection accuracy of the composite neural network structure reaches 0. 888, with an accuracy value of 0. 883 and a recall rate of 0. 789. The composite neural networks is applied to fault diagnosis of oilfield wastewater systems, can achieve intelligent fault detection, reduce economic costs, and build smart oilfield.

 With the advancement of information technology, the digital transformation of libraries has become a key avenue for enhancing service efficiency. During this transformation process, information security issues have become increasingly prominent, posing threats to the protection of library resources and the security of user data. This study is to address the information security challenges in the digital transformation of libraries by proposing a comprehensive information security protection system. By analyzing the main information security risks faced by libraries currently, including cyber attacks, copyright disputes, insufficient security awareness among management personnel, and low levels of resource sharing, a four-tier information security protection system is constructed consisting of the user application layer, service platform layer, data center layer, and infrastructure layer. This system can effectively enhance the security and access control of library information resources, and strengthen the security of digital resource access. In the process of digital transformation, libraries must consider information security as a core factor, and build a comprehensive information security protection system through the collaborative work of technology, management, and organization to ensure the security and efficient use of digital resources.

 There is a phenomenon of domain drift in webpage search results, where the returned webpage is not related to the search keyword domain, resulting in that users are unable to search for demand information. Therefore, a web search sorting algorithm based on Softmax regression classification model is proposed. Through the Feature selection of web search text, the corresponding feature items are obtained. Using the vector representation model, the selected web search text feature items are converted into formatted data, and the web search text data is balanced to obtain the web search text data set. Using the Softmax regression classification model, the web search text dataset is classified and processed, the types of web search texts is predicted. And the OkapiBM25 algorithm is used to sort web search texts, achieving web search sorting. The experimental results show that the proposed algorithm performs well in web search sorting, effectively improving the accuracy of web search sorting and avoiding domain drift during the process of web search sorting.

The dung beetle optimization algorithm suffers from the problems of easily falling into local optimum, imbalance between global exploration and local exploitation ability. In order to improve the searching ability of the dung beetle optimization algorithm, a mixed-strategy dung beetle optimization algorithm is proposed. The Sobol sequence is used to initialize the population in order to make the dung beetle population better traverse the whole solution space. The golden sine algorithm is added to the ball-rolling dung beetle position updating stage to improve the convergence speed and searching accuracy. And the hybrid variation operator is introduced for perturbation to improve the algorithm’s ability to jump out of the local optimum. The improved algorithms are tested on eight benchmark functions and compared with the gray wolf optimization algorithm, the whale optimization algorithm and the dung beetle optimization algorithm to verify the effectiveness of the three improved strategies. The results show that the dung beetle optimization algorithm with mixed strategies has significant enhancement in convergence speed, robustness and optimization search accuracy.

To improve the employment matching and human resource utilization efficiency of college students, many researchers are dedicated to developing effective job recommendation algorithms. However, existing recommendation algorithms often rely solely on a single information source or simple user classification, which can not fully capture the multidimensional features and personalized needs of college students, resulting in poor recommendation performance. Therefore, a job recommendation algorithm for college students based on user profiles and bipartite graphs is proposed. With the aid of the conditional random field model based on the integration of long and short-term memory neural networks, the basic user information is extracted from the archives management system of the university library, based on which the user portrait of university students is generated. The distance between different user profile features is calculated, and the k-means clustering algorithm is used to complete the user profile clustering. The bipartite graph network is used to build the basic job recommendation structure for college students and a preliminary recommendation scheme is designed based on energy distribution. Finally, based on the weighted random forest model, the classification of college students’ employment positions is realized by considering users’ preferences for project features, and the score of the initial recommendation list is revised to obtain accurate recommendation results for college students’ full employment positions. The experimental results show that after the proposed method is applied, a recommendation list of 120 full employment positions for college students is given, and the hit rate of the recommendation result reaches 0. 94. This shows that the research method can accurately obtain the results of college students’ employment position recommendation, so as to improve the employment matching degree and human resource utilization efficiency. 

To detect transactional fraud in financial services industry and maintain financial security, an end-to- end modeling framework, methodology, and model architecture are proposed for financial transactional data with imbalanced and discrete classes. The framework covers data preprocessing, model training, and model prediction. The performance and efficiency of different models with different numbers of features are compared and validated on a real-world dataset. The results demonstrate that the proposed approach can effectively improve the accuracy and efficiency of financial transactional fraud detection, providing a reference for financial institutions to select models with different types and numbers of features according to their own optimization goals and resource constraints. Tree-based models excel with over 200 features in resource-rich settings, while neural networks are optimal for medium-sized feature sets (100 ~200). Decision trees or logistic regression are suitable for small feature sets in resource-constrained, long-tail scenarios. 

 In view of the low accuracy and efficiency of Internet of Things authentication due to the influence of noise, a new optimization method of Internet of Things authentication based on improved RSA(Rivest-Shamir- Adleman) algorithm was proposed. In this method, a transmission channel model is constructed to obtain user identity information and a noise reduction model is constructed to preprocess user identity data. Based on the processed data, the user identity characteristic information is extracted to build the Internet of Things identity authentication algorithm. On this basis, RSA algorithm is introduced to encrypt and process user identity information data to realize the optimization of Internet of Things identity authentication. In addition, the proposed method is not easily affected by noise environment. Under the condition of noise, the maximum error between the authentication rate and the ideal authentication rate is only 3.7%. Therefore, the proposed method is feasible and effective. 

In order to solve the problem of heterogeneous data from multiple sources encountered by electric power construction companies in the process of transitioning to whole-process consulting business, as well as the challenge of design managers needing to work together remotely due to unforeseen circumstances, it is proposed to construct a knowledge graph by using the enterprise’s private cloud as the basic environment and combining the technology of fusion of heterogeneous data from multiple sources. The system optimises the data management process and ensures data consistency and availability by integrating diverse data sources from various provinces and cities across the country. It ultimately realises distributed collaborative management of the whole process of consulting business and significantly improves the core competitiveness of the enterprise. At the same time, it effectively solves the problems of data variety, wide range of sources and diverse and inconsistent protocols, improves data quality and accuracy, and unifies the storage architecture to enhance the overall data management efficiency and decision-making support capabilities.

 The current learner model lacks exploration of this dimension of cognitive load, which, as a load generated by the cognitive system during the learning process, has a significant impact on the learning state of learners. Based on the CELTS-11(China E-Learning Technology Standardization-11) proposed by the China E-Learning Technology Standardization Committee, cognitive load is integrated into the learner model as a dimension, and an LMICL ( Learner Model Incorporating Cognitive Load) combining static and dynamic information is constructed. Afterwards, relying on an adaptive learning system, the data of the unmixed cognitive load learner model and the LMICL data were used as the basis for recommending learning resources, resulting in two different learning resource recommendation results. Two classes of learners were randomly selected to learn system, and then their academic performance. The results of cognitive load and satisfaction were used to validate the effectiveness of LMICL , and it was found that the recommendation learning effect based on LMICL was better than that of the learner model without integrating cognitive load.

 Camera is utilized to capture target LED (Light Emitting Diode) image in OCC(Optical Camera Communication) system, the performance degradation of OCC system occurs due to outdoor ambient light interference. The strong sunlight causes great difficulty in decoding at the receiving end of OCC system, in order to solve the problem a Gradient-Harris decoding algorithm based on piecewise linear gray transformation is proposed. A set of OCC experimental system is built, original images are captured by camera at the receiving end of OCC experimental system, and the target LED array region is extracted by standard correlation coefficient matching method. The image of target LED array region is enhanced by segmented linear gray transformation, a Gradient-Harris decoding algorithm is used for shape extraction and state recognition of target LED array. The experimental results show that the proposed Gradient-Harris decoding algorithm based on piecewise linear gray transform is effective for OCC experimental system in strong sunlight environment, the average decoding rate is 128. 08 bit/ s, the average bit error rate is 4. 38 x 10-4, and the maximum communication distance is 55 m. 

 To solve the problem of low accuracy in identifying abnormal data in existing methods. An abnormal data recognition algorithm for multi-dimensional feature-based communication network is proposed. The current speed and position of particles in particle swarm optimization algorithm is adjusted to obtain multi-dimensional data samples of communication network. Data features are extracted through clustering analysis in data mining, determining density indicators, and obtaining multidimensional features of the data. The extracted multidimensional features are Introduced into the deep belief network for recognition, and anomaly recognition of communication network data is achieved based on changes in feature spectrum amplitude. The experimental results show that the algorithm can effectively identify abnormal data features in communication networks and has high recognition accuracy. 

In the process of 3D reconstruction of digital images, problems such as noise and distortion in the original data lead to low efficiency and accuracy of feature matching. To address this issue, a 3D digital image reconstruction method based on SIFT(Scale-Invariant Feature Transform) feature point extraction algorithm is proposed. The Bilateral filter algorithm is used to eliminate the environmental noise in the digital image, retain the edge information of the digital image, and improve the accuracy of feature point extraction. The SIFT algorithm is used to obtain feature point pairs. Using this feature point pair as the initial patch, a dense matching method for spatial object multi view images is used to achieve 3D reconstruction of digital images. The experimental results show that the proposed method has high feature matching efficiency and accuracy and strong noise reduction ability. The average time required for generating 3D reconstructed images is 26. 74 ms. 

In order to ensure the security management of the hospital information network and avoid medical information leakage, an intrusion detection algorithm for abnormal information in the hospital network based on deep generative model was proposed. Using binary wavelet transform method, multi-scale decomposition of hospital network operation data, combined with adaptive soft threshold denoising coefficient to extract effective data. The Wasserstein distance algorithm and MMD(Maximun Mean Discrepancy) distance algorithm in the optimal transportation theory are used to reduce the dimension of the hospital network data in the depth generative model, input the reduced dimension network normal operation data samples into the anomaly detection model, and extract the sample characteristics. Using the Adam algorithm in deep learning strategy, generate an anomaly information discrimination function, and compare the characteristics of the tested network operation data with the normal network operation data to achieve hospital network anomaly information intrusion detection. The experimental results show that the algorithm can achieve efficient detection of abnormal information intrusion in hospital networks, accurately detect multiple types of network intrusion behaviors, and provide security guarantees for the network operation of medical institutions. 

During the classification process of unstructured big data, due to the large amount of redundant data in the data, if the redundant data cannot be cleaned in a timely manner, it will reduce the classification accuracy of the data. In order to effectively improve the effectiveness of data classification, a non structured big data classification optimization method based on the improved ID3(Iterative Dichotomiser 3) algorithm is proposed. This method addresses the problem of excessive redundant data and complex data dimensions in unstructured big data sets. It cleans the data and combines supervised identification matrices to achieve data dimensionality reduction; Based on the results of data dimensionality reduction, an improved ID3 algorithm is used to establish a decision tree classification model for data classification. Through this model, unstructured big data is classified and processed to achieve accurate data classification. The experimental results show that when using this method to classify unstructured big data, the classification effect is good and the accuracy is high. 

 To effectively solve the problem of uneven brightness in 3D virtual images, a dynamic feature enhancement algorithm for 3D virtual images is proposed. Median filtering algorithm and wavelet soft thresholding algorithm are combined effectively for denoising of 3D virtual images. By setting new structural elements based on visual selection characteristics, connected particle attributes are constructed, and using hierarchical statistical models to perform color conversion and structural element matching on the image, corresponding mapping subgraphs are obtained, and dynamic features are extracted. The 3D virtual image is inputted into an improved U-net++network, dense connections are used at different layers to enhance the correlation of image features at different levels, and all dynamic features are fused for detail reconstruction to achieve dynamic feature enhancement of the 3D virtual image. According to the experimental results, the proposed algorithm can achieve satisfactory dynamic feature enhancement effects in 3D virtual images. 

Digital media deep interactive images are affected by noise, resulting in poor edge detection performance and affecting segmentation accuracy. Therefore, a digital media deep interactive image segmentation algorithm based on edge detection is proposed. Firstly, the wavelet transform method is used to denoise images in digital media to improve the accuracy of image segmentation. Secondly, Gaussian function and low-pass filter are used to enhance the denoised image, improve the image definition, and facilitate image segmentation. Finally, based on the adaptive threshold algorithm, edge detection is performed on digital media images. There are two thresholds in the pixel collection, the upper threshold and the lower threshold. The high and low thresholds in the pixel set are calculated based on the calculation of their upper and lower thresholds, and edge connections between the two thresholds are implemented to achieve digital media image segmentation. The experimental results show that the proposed method has good denoising effect, high segmentation accuracy, and high segmentation efficiency for segmented digital media images. 

In response to the difficulty in determining the characteristics of users and data for mobile terminal access, which leads to high difficulty in access control, an access control technology based on EVM(Error Vector Magnitud) measurement algorithm is proposed to effectively solve the problem. Considering the impact of noise and other interference factors in the environment, the Qos(Quality of Srvice) condition is used as the initial access condition for users. The characteristics of users or data that meet this condition are calculated, and the feature values are converted into weight factors as reference for access control. The EVM measurement algorithm is used to calculate the difference between the internal and external signals of the terminal channel, and the user weight factor is used to derive the access threshold of the mobile terminal. The increasing and decreasing functions between different user threshold values and control values are solved, and precise control of mobile terminal access is achieved according to the priority order of the functions. The experimental data shows that the proposed method has high access control accuracy, and after control, the terminal transmission delay and blocking rate have been significantly improved, and the data arrival rate has also been significantly improved. 

The use of distributed deployment of sensors will lead to the edge of the server data distribution single imbalance phenomenon, the model training under edge computing can also result in serious privacy leakage problem due to the data set pollution caused by gradient anomaly. RDSEC(Strongly Robust Data Security Algorithms for Edge Computing) is proposed, encryption algorithm is used to encrypt the parameters of the edge server to protect privacy. If an anomaly is found in the gradient anomaly detection of the edge node, the edge node uploads the gradient with a signal to tell the cloud center if the current parameters uploaded by the edge node are available. The experimental results on CIFAR10 and Fashion data sets show that the algorithm can efficiently aggregate the parameters of edge servers and improve the computing power and accuracy of edge nodes. Under the condition of ensuring data privacy, the robustness, accuracy and training speed of the model are greatly improved, and the high accuracy of edge node is achieved. 

 The traditional desktop application library integration method has some limitations in practical applications, such as the expansion of the standard OS directory, the complexity of distribution package making, the need to modify the middle layer library when multiple level library calls are included, and the inconsistency between development and deployment environments. To solve the problems, an integration framework is proposed. The cores of it are managing libraries in the way of managing resources such as images and implement the dynamic deployment of libraries at runtime based on the detection results of constraints and dependencies between libraries. Through the design of the four components, Library resource management, runtime dynamic deployment, runtime dynamic loading and resource manager, and their collaboration, the integration framework for the first time implements the combination of the above two cores. The practical application of CIFLog Integrated Logging Platform method module integration shows that the integration framework can solve the problems existing in the traditional library integration. The applicability of this framework can be applied to the library integration of all desktop applications, providing a new idea for the library integration of desktop applications. 

In order to overcome the oversimplified treatment of the defect discovery rate in the traditional G-O (Goel-Okumoto) software reliability model, an improved model that more accurately describes the actual change of the defect discovery rate over time is proposed. Unlike the conventional assumption that it is treated as a constant or monotonic function, the improved model considers the progress of testers’ learning and debugging capabilities and the inherent tendency of the software’s defect discovery rate to decrease over time. Therefore, it assumes that the defect discovery rate first increases before showing a dynamic trend of decline. The model’s effectiveness is verified by applying it to two sets of public software defect detectionda tasets and comparing it with a variety of classic models. Experimental results confirm that the improved G-O model demonstrates excellent performance in both fitting and prediction capabilities, proving its applicability and superiority in software reliability assessment.