The existing logging image interpretation work highly relies on manual experience and expert opinions, which can not quickly understand and give the gist meaning of the image, seriously affecting the depth mining and utilization of the information contained in the logging image. A logging image description method based on the ConvNext network coding-decoding architecture is proposed. The ConvNext network with mixed dilated convolution encoder is adopted to enhance model’s ability to extract low resolution image detail information. Then, the mechanism of additive attention is used to replace the original model for long attention mechanism, cooperated with LSTM(Long Short Term Memory) capable of temporal information memory, enhancing the model’s ability to capture long rely on information, which can generate descriptions of logging image more accurate and more natural. The experimental results show that the evaluation indexes of BLEU-4 (Bilingual Evaluation Understudy-4), METROR(Metric for Evaluation of Translation with Explicit ORdering) and CIDEr (Consensus-based Image Description Evaluation) are improved by 3. 8,4. 0 and 5. 3, respectively, compared with the baseline model method. This research scheme using ConvNext architecture to describe log image information is feasible. 

Existing research on MIMO ( Multiple -Input Multiple -Output) backscatter communication mostly improves reliability at the cost of rate, which fails to meet the performance requirements of high -rate scenarios. In order to balance the SER(Symbol Error Rate) and transmission rate of MIMO backscatter communication, unlike previous works where the power supplier only serves as an energy provider, an antenna selection -based power supply scheme for MIMO backscatter communication systems is proposed. The essence of the proposed scheme is to select only the power -transmitting antennas corresponding to the links with optimal communication quality to enhance MIMO backscatter communication performance. Rigorous mathematical proofs and numerous simulation results indicate that the proposed scheme can provide significant performance improvement in rate under the same diversity order with respect to the conventional approach, and this gain can also be translated into a significant improvement in SER performance at the same transmission rate. This scheme provides effective theoretical support and technical guidance for high -rate, high -performance transmission in MIMO backscatter communication.

With growing demands for energy efficiency and carbon reduction in power grid operation, addressing the shortcomings of traditional scheduling approaches has become increasingly urgent. A hierarchical carbon trading mechanism is integrated with user-side responsiveness to construct a demand-side response model under carbon constraints. To enhance the solution process, the white shark optimizer is improved through chaotic initialization, directional update, and cosine mutation strategies, mitigating issues such as limited convergence accuracy and susceptibility to local optima. An optimal scheduling model for the active distribution network is formulated, targeting the minimization of operational costs. The refined algorithm is employed to solve the model, with simulation outcomes demonstrating its effectiveness and accuracy. Experimental results indicate that the proposed method significantly improves energy scheduling performance while supporting low-carbon objectives, and holds potential for broader application in smart grids incorporating renewable energy sources.

To accurately assess the health status of charging station equipment and effectively predict the operational trends of the equipment, a health status assessment method for charging equipment is proposed. Firstly, by analyzing the technical and functional characteristics of charging piles, a comprehensive health status assessment system is constructed. Secondly, the AHP(Analytic Hierarchy Process) is used to calculate the subjective weights, and CRITIC(Criteria Importance Through Intercriteria Correlation) is adopted to calculate the objective weights. Based on the theory of cooperative game, the subjective and objective weights are linearly combined into a comprehensive weight. The normal distribution cloud model is introduced as the membership function of the assessment model. By calculating the membership function values of each index, the membership matrices corresponding to different health status levels are constructed. Finally, the real-time operation data of a certain charging equipment is analyzed to verify the feasibility of the assessment model.

Hospital operation data needs to be synchronized across multiple systems and faces frequent access and processing, leading to issues such as data loss, errors, or duplication in the storage process, which affects the accuracy of data analysis and decision-making. To ensure data consistency and accuracy while providing efficient data access and processing capabilities, a hospital operation data cloud storage method based on deep integrated networks is proposed. Build a random forest tree using filtered hospital operation data and perform spatial dimension random extraction within the forest. Using convolutional neural networks as individual sub classifiers and combining with majority voting mechanism to form a deep ensemble network, the final sample category is determined through transfer learning strategy. Effectively identifying and filtering duplicate data in hospital operation data using Bloom filter technology, and by setting an objective function and using collaborative evolution algorithm through multiple rounds of iteration and optimization, ultimately obtaining the optimal solution for cloud storage of hospital operation data. The experimental results show that the proposed method has a throughput of over 400 MByte/ s when processing data, and its memory and CPU(Central Processing Unit) usage rates are only 10. 39% and 8. 88% when the data dimension is as high as 9 000. The proposed method has better cloud storage performance for hospital operation data, which can effectively improve the efficiency ofhospital information management and the ability of collaborative work. 

In complex network environments, the increase in the number and diversity of device terminals have led to network bandwidth and latency, becoming bottlenecks that restrict fast data access. In order to effectively manage and allocate different device terminal resources, a fast data access method for hospital equipment terminals facing 5G(5th Generation Mobile Communication Technology) cellular IoT(Internet of Things) is proposed. Establish a 5G cellular IoT scenario, determine the network environment and available resources, and provide a basis for data access. Combining the K-means clustering algorithm with the maximum minimum distance algorithm to cluster and partition terminal data, establishing a sequence of data access, effectively managing the process of data access, and achieving rapid access of hospital equipment terminal data. The experimental results show that the proposed method can efficiently and stably achieve fast access to terminal data.

This study addresses the intra and inter-strip radiometric non-uniformity in UAV(Unmanned Aerial Vehicle) push-broom hyperspectral imagery through an adaptive radiometric correction model based on minimum spectral features. By constructing a black-box model to simulate imaging physics, pixel-wise nonlinear mapping functions referenced to the minimum spectral features is established, synergistically optimizing intra-image spectral consistency and inter-strip radiometric uniformity. Experiments demonstrate that the corrected imagery exhibits radiation intensity fluctuation amplitude reduced to within 5% of original data, with over 90% elimination rates for inter-strip banding artifacts and brightness gradient distortion. The absolute error of total phosphorus concentration inversion in water bodies derived from corrected data remains stably controlled within the 0. 05 mg/ L precision threshold. The proposed model enhances multi-strip radiometric correction efficiency by establishing nonlinear relationships between spectral radiation characteristics and sensor response, significantly improving engineering applicability for quantitative inversion of push-broom hyperspectral data and providing a reliable radiometric reference for regional-scale hyperspectral remote sensing monitoring. 

In order to make bionic machine simulate human sense of touch to perceive the surface information of objects, according to the piezoelectric effect and pyroelectric effect of PVDF ( Polyvinylidene Fluoride) piezoelectric film, a set of multi-tactile information acquisition system is designed and built, realizing the collection and recognition of various tactile information such as hardness, viscosity, humidity, roughness and temperature of objects. The PVDF piezoelectric film is pasted on the surface of the rubber hemisphere to make a tactile sensor, a single-chip microcomputer is used to control the lifting table and the lead screw slide table to drive the object to press and friction the tactile sensor, effectively collecting the tactile information of the object, and establisheing a mathematical model of viscosity and temperature by using the eigenvalues of the collected signal for analysis. The viscosity and temperature sensitivity are-2 730 V/ m2 and 0. 984 mV/ °C respectively, the hardness of the object is effectively characterized by variance, and the success rate of predicting humidity and roughness of more than 90% is achieved by establishing a BP(Back Propagation) neural network.

The drone-based TEM ( Transient Electromagnetic) system combines the advantages of being lightweight, flexible, efficient and safe, with broad application prospects. However, it also produces various noises that affect the detection performance in practical applications. Based on the dipole model and DE (Differential Evolution) algorithm, targets are located, noise sources and types are analyzed through field tests, and the influence of noise on system positioning performance is studied. The results show that the system is affected by static noise, low-frequency noise caused by coil rotation/ swing, and drone noise. When only affected by static noise with target depth of 80 ~100 cm, the positioning error does not exceed 9 cm, and the inversion depth is shallower than the actual depth. When coil rotation/ swing amplitude is small, low-frequency noise has minor influence. As amplitude increases, the influence grows significantly, with positioning error increasing to 29 cm and inversion depth becoming deeper than actual depth. Drone noise also increases positioning error to 12 cm, with inversion depth deeper than actual depth. The research determines the types, sources and effects of noise, and guides the subsequent system optimization. 

Design of JOM-5V Vector Overhauser Magnetometer ZHOU Runze, ZHANG Shuang, CHEN Shudong (College of Electronic Science and Engineering, Jilin University, Changchun 130012, China) Abstract: Vector Overhauser magnetometer can measure magnetic inclination I and magnetic declination D on the basis of total field F, which is widely used in the field of geomagnetic science research area. To meet the demand of simultaneous measurement of F, I and D, JOM-5V vector Overhauser magnetometer is developed. The magnetometer adopts the sensors of the land JOM-5 series portable Overhauser magnetometer, and combines the Helmholtz coil and solenoid coil to form the main structure of the vector magnetometer, which is simpler than the structure of the vector magnetometer with spherical coils, and is easy to commercialize. The experimental results show that the sensitivity of JOM-5V vector Overhauser magnetometer can reach 0. 016 8 nT at a measurement period of 5 s, and the sensitivities of magnetic inclination and magnetic declination are 0. 011 5° and 0. 013 8°, respectively. The results show that the JOM-5V vector Overhauser magnetometer is able to meet the measurement needs of F,I,D geomagnetic variables, which provides technical and equipment support for geomagnetic scientific research in China.

In response to the challenges of complex terrain and dynamic environments in deep-sea marine exploration, a split marine towed magnetic survey system architecture is proposed and the JPM-HT2 magnetometer is developed. This system decouples the electronic unit from the sensor, with the electronic module packaged in a lightweight aluminum box to enhance the system’s deployability. The sensor adopts an “8”-shaped coil structure to meet the omnidirectional detection requirements of marine magnetic surveys. A high-strength fiberglass tube is used to protect the sensor from complex mechanical impacts underwater. Experimental results show that the signal-to-noise ratio measured by the sensor at different attitudes can reach a maximum of 38. 2 and a minimum of 11. 6, which is 30. 4% of the maximum value. The sensitivity results of the sensor at different orientations indicate that the optimal sensitivity of the JPM-HT2 magnetometer in a 3-second measurement cycle is 0. 050 nT, and the worst is 0. 065 nT, better than the optimal sensitivity of 0. 1 nT of most commercial magnetometers. The dynamic measurement process is simulated in the ocean, and the dynamic sensitivity of the instrument at this time is 0. 19 nT. The magnetic resolution of the evaluation instrument is better than 0. 01 nT, which indicates that the instrument can conduct accurate magnetic detection in deep water complex terrain and dynamic environments.

To meet the requirements of 5G networks for high capacity, low interference, and high spectral efficiency, massive MU-MIMO(Multi-User Multiple-Input Multiple-Output) has become a key base-station-side enabling technology. Among linear precoding schemes, ZF(Zero-Forcing) can effectively suppress multiuser interference. However, it suffers from noise amplification in the low-to-medium SNR(Signal-to-Noise Ratio) regime. RZF(Regularized Zero-Forcing) introduces a regularization term to balance interference suppression and noise mitigation, while system performance is highly sensitive to the choice of the regularization coefficient, and a fixed coefficient is difficult to adapt to varying SNR conditions and user loads. To address these issues, an A-RZF(Adaptive RZF) precoding method is proposed, where the regularization coefficient is set adaptively according to the number of users and the noise level, thereby improving spectral efficiency and cell-edge user rates without significantly increasing computational complexity. Based on a downlink MU-MIMO system model, reproducible simulation comparisons against baseline schemes including MRT(Maximum Ratio Transmission), ZF, and fixed-coefficient RZF are established, and performance evaluation in terms of sum spectral efficiency, antenna-scaling gains, cell-edge user rates, and computational complexity is conducted. Simulation results under typical parameter settings show that A-RZF achieves more robust performance advantages over ZF and fixed RZFin the low-to-medium SNR range, while maintaining capacity gains consistent with the growth in the number of antennas.

The variability and randomness of wind energy make power prediction challenging, and traditional methods struggle to fully exploit the underlying features of data when sample sizes are limited. To address this, a TCN-LSTM(Temporal Convolutional Network-Long Short-Term Memory) short-term wind power forecasting method based on improved Generative Adversarial Network sample augmentation is proposed. First, to tackle the issue of insufficient data, a TWGAN-GP ( Temporal Dynamic Maximum Mean Discrepancy-Constrained Wasserstein Generative Adversarial Networks with Gradient Penalty) model with dual constraints of TD-MMD (Temporal Dynamic Maximum Mean Discrepancy) and Wasserstein distance is introduced to generate high- quality wind power time series data for augmenting the training set. Secondly, a TCN-LSTM forecasting model is constructed, and TCN(Temporal Convolutional Networks) is used to capture long-term temporal dependencies, combined with LSTM(Long Short-Term Memory) to extract dynamic features, and to optimize hyperparameters through the RIME(Rime Optimization Algorithm) optimization algorithm. Finally, through case analysis, theproposed method demonstrates a significantly superior prediction accuracy compared to benchmark models. This method effectively resolves the forecasting challenges caused by the randomness and volatility of wind power, overcomes the shortcomings of traditional methods in feature extraction under small sample conditions, and provides a reliable approach for precise short-term wind power prediction.

To cultivate innovative talents in information disciplines under the background of emerging engineering education, an experimental project on detection and classification methods for pine wilt disease based on high- resolution satellite remote sensing imagery is designed, taking the “Digital Image Processing’’ and “Remote Sensing Principles and Applications’’ courses as the foundation, and combining with the “College Students’ Innovation and Entrepreneurship Training Program’’ of Jilin University. Using high-resolution satellite imagery from “Jilin-1’’, a pine wilt disease data sample library is established through image preprocessing. By comparing and analyzing the detection results of pine wilt disease using YOLOv5(You Only Look Once version 5), Faster R-CNN(Faster Region-Convolutional Neural Network), SSD(Single Shot MultiBox Detector), and YOLOv7 network models, the YOLOv7 network model is selected for optimization. The improved YOLOv7 network model has achieved a disease detection accuracy of 90. 66%. By analyzing the spectral characteristics of infected trees with different disease severity levels, SVM(Support Vector Machine)method is adopted to classify the severity of pine wilt disease, achieving a classification accuracy of 95. 44%. The results demonstrate that this experimentcan effectively help students integrate professional theoretical knowledge with practical techniques, enhance students’ practical innovation capabilities, and achieve the expected teaching outcomes.

To address the issue of eyelid occlusion affecting recognition performance in iris recognition, a solution based on ViT(Vision Transformer) is proposed. Firstly, a FFM(Feature Fusion Module) is proposed to achieve feature extraction and fusion at different scales, solving the problem of information loss during feature extraction. Secondly, the local feature encoder is pre-trained by minimizing reconstruction loss to avoid forming triplets with heterogeneous irises sharing the same dominant features. This prior knowledge endows the model parameter adjustment with certain interpretability. An interactive encoding structure is constructed with ViT and residual blocks as the core, efficiently fusing information from different iris blocks to form comprehensive feature representation. Finally, the traditional triplet loss is improved by incorporating the threshold concept, providing a clearer learning direction for model training. Experimental results show that the proposed method can effectively eliminate the negative impact of occlusion on iris recognition and significantly improve recognition performance.

To address the challenges of data scarcity, high annotation costs, and limitations of traditional relation extraction models in imaging logging, a few-shot relation extraction method based on the ConceptFERE(Concept- Enhanced Few-Ehot Relation Extraction) model is proposed. Using the BERT-PAIR(BERT-Paired Sentence Encoding) framework, the ConceptFERE model is improved and the SDG-ConceptFERE(Semantic Difference Gate-ConceptFERE) model is introduced. It includes a semantic difference gate mechanism that dynamically assesses the relevance between support and query instances. By incorporating external entity concepts into the support instances, the classification errors from incorrect semantic enhancement are prevented. Experiments show SDG-ConceptFERE improves accuracy by 3.57% and 2.78% over ConceptFERE in 5-way-1-shot and 5-way-5- shot tasks, proving its effectiveness in providing better text support for logging researchers and advancing intelligent decision-making in exploration and development. 

The security of private data is affected by internal and external threats, such as data leakage, malicious tampering, hacker attacks, etc. In the process of privacy data tamper proof encryption, the key issue is to ensure the integrity and security of data while maintaining its privacy. Therefore, an encryption method of privacy data tamper proof based on longitudinal federated learning algorithm is proposed. By using vertical federated learning algorithms to construct a hierarchical structure for network malicious user detection, malicious user detection is carried out to ensure the integrity and security of private data. Using a random forest regression model to accurately classify private data, and generating chaotic random sequences of private data using composite chaotic sequences. In the sequence encoding stage, a binary optimization method is adopted to ensure the efficiency and accuracy of the encoding. Encryption keys for private data is designed using key control methods, achieving tamper proof encryption of private data through feature clustering and encoding fusion techniques. The experimental results show that the proposed method performs well in protecting the security of private data and can effectively prevent the risk of tampering and leakage of private data.

In multi-source data integration, data may be distributed in different subspaces and have a high degree of data imbalance. In order to improve the efficiency of data analysis, a multi-source data integration processing method based on the Lubang subspace clustering algorithm is proposed. Firstly, by improving the data balancing algorithm, the maximum number of class samples and the average number of class samples are calculated, and the composite minority class oversampling technique is used to obtain a relatively balanced subset, solving the problem of imbalanced data distribution. Then, by using the Dice coefficient similarity measure, the cosine similarity of multi-source data is calculated. By evaluating the similarity between data from different sources, the problem of data heterogeneity and redundancy is solved. Finally, on the basis of establishing self representativeness and establishing affinity graphs to reveal the inherent correlations of data, the Lu Bang subspace clustering algorithm is used to identify the feature subspaces of different data. By introducing a robustness mechanism which can resist the influence of noise and redundant features, the membership degree of the data is calculated, and data integration processing performed based on the membership degree. The experimental results show that this method can achieve integrated processing of multiple source data, improve data analysis efficiency, and ensure data consistency and reliability.

To address the problems of high computational cost, heavy storage requirements, and low deployment efficiency of full fine-tuning in adapting PLMs ( Pretrained Language Models) to downstream tasks, the development of PEFT(Parameter-Efficient Fine-Tuning) techniques is studied, and a classification framework covering adapter-based methods, low-rank adaptation methods, prompt-based methods, selective fine-tuning methods, and dynamic tuning methods is constructed. These methods are comparatively analyzed in terms of their basic principles, representative approaches, applicable scenarios, and application characteristics. The analysis shows that different PEFT methods exhibit distinct advantages in parameter efficiency, task generalization ability, and deployment flexibility, and can provide references for the lightweight deployment, multi-task transfer, and personalized adaptation of large models. 

When filling electronic medical record data, a fixed filling value will result in low completeness of the filled data, affecting the quality of the filling method. Therefore a dynamic filling algorithm for missing values in multimodal data of electronic medical records is proposed. The obtained multimodal data of electronic medical records is fused, an electronic medical record data model is constructed, corresponding sequence features are mined, and the missing value positions of electronic medical record data is detected under the action of forward propagation network. The missing values of electronic medical record data is estimated based on the established time series matrix. Using the estimated missing values as initial values, the electronic medical record data is preliminarily filled in, and then the fuzzy clustering algorithm is applied to perform preliminary clustering. By continuously updating the clustering centers and filling values, the optimal filling value is calculated to achieve the filling of missing values in the electronic medical record data. The experimental results show that the designed filling algorithm can accurately fill in electronic medical record data, and the completeness of the filled electronic medical record data is 0. 95, indicating high data quality.

 In order to allocate doctor resources reasonably, minimize outpatient waiting time, and improve patient experience and hospital operation efficiency, a simulated annealing based outpatient waiting time minimization scheduling algorithm is proposed. This algorithm is based on queuing theory to predict the length of outpatient waiting queues and the waiting time per unit time period, and to define a scheduling optimization decision function with the goal of minimizing waiting time. To ensure the feasibility and rationality of the decision-making objectives, multi-dimensional constraints are embedded in the algorithm, including constraints on doctors’ working time and doctor skill matching. Simulated annealing algorithm is introduced to solve the decision function and output the optimal scheduling plan through a Markov chain that iteratively generates new solutions, determines, accepts, and discards. The experimental results show that after simulating annealing algorithm scheduling, the waiting time of patients can be controlled within 16 minutes, providing practical reference for doctors’ scientific scheduling.

Meteorological business data contain temporal, spatial and multivariable dimensions. The increase in dimensions leads to an increase in data sparsity, presenting different correlation patterns at different temporal/ spatial scales. It is difficult to map the correlation between meteorological elements and data, resulting in poor structural similarity of the filling results. Therefore, a high-dimensional correlation deficiency data block-filling algorithm for meteorological services is proposed. Combined with the mutual information algorithm, the probability distribution of continuous meteorological variables is approximated based on KDE(Kernel Density Estimation). The nonlinear statistical dependence between variables is quantified through the mutual information formula to generate a symmetric mutual information matrix and capture the local correlation of meteorological elements. The normalized mutual information matrix is transformed into a similarity matrix, and the strong correlation is strengthened and the weak correlation is weakened through exponential function mapping. The Laplacian matrix is constructed, its eigenvectors are calculated, and the k-means algorithm is used to cluster the eigenvectors to achieve attribute blocking. The meteorological data is divided into strongly correlated sub-blocks through block processing, and an independent CGAN(Conditional Generative Adversarial Network) is designed for each sub-block. By designing the loss function and training the conditional generative adversarial network, the model can generate imputed values consistent with the distribution of real meteorological data. The experimental results show that when the proposed method is used for block filling of missing data, the structural similarity of the filling results is stable at 0. 92, indicating that this method has an ideal filling effect.

To address the limitations of the DCP (Deep Closest Point) point cloud registration algorithm, including its poor fine-grained feature extraction capability, low computational efficiency, and feature misalignment issues, a deep learning-based point cloud registration network that integrates Mamba and SoftBBS (Soft Best Buddies Similarity) is proposed. Firstly, high-dimensional features are extracted from raw point cloud data using a DGCNN(Dynamic Graph Convolutional Neural Network)and the Mamba network, enhancing local feature extraction and computational efficiency. Secondly, SoftBBS is employed to compute an optimal point-pair similarity matrix, reducing the impact of low-reliability matches on registration results and improving robustness. Finally, a LS(Least Squares Method) method is used to calculate the optimal rigid transformation matrix, enhancing registration accuracy. Experimental results show that, compared to DCP, the proposed registration algorithm improves accuracy by 65. 1% and also outperforms several recently popular deep-learning-based registration networks in robustness.

To address the problem of information narrowing caused by the bidirectional feedback characteristics of algorithmic recommendation systems which reveals its underlying mechanisms in the phenomenon of information cocoons in the digital age, a method of constructing a dynamic bidirectional feedback simulation model based on user algorithm is proposed. The model characterizes the positive feedback cycle mechanism of bidirectional feedback by quantifying user interest vector, openness parameter and confirmation bias strength, and combining the hybrid recommendation strategy of collaborative filtering and content filtering. The simulation results show that the degree of personalization of the algorithm, as the key driving factor, significantly affects the formation intensity and evolution rate of the information cocoon. The diversity of users’ initial interests constitutes the basic defense mechanism against the information cocoon. The higher the level of diversity, the lower the risk of individuals falling into the information cocoon. The user openness acts as the key postnatal adjustment variable of the information cocoon room, which regulates the development process of the information cocoon by affecting the reception and processing of heterogeneous information by users. The polarization degree of users’ final views is jointly determined by the personalized strength of the algorithm, the openness of users, and the diversity of users’ initial interests. The research systematically reveals the dynamic formation mechanism of information cocoon under the environment of recommendation algorithm, and provides empirical basis and theoretical support for algorithm governance.

In traditional methods for capturing human upper limb movements, sensor data is susceptible to noise interference and occlusion, resulting in data loss and affecting the accuracy of motion capture. Therefore, the cuckoo search algorithm is introduced, which utilizes its global search ability to find the optimal or approximately optimal solution in the presence of noise and occlusion, thereby improving the robustness of data processing and achieving accurate capture of human upper limb movements. Firstly, the Kinect sensor is used to collect data on human upper limb movements, and the quaternion method is used to obtain the position coordinates of human upper limb joint points. Then, based on optimization algorithms, the orientation of human upper limb movements is determined. Finally, the cuckoo search algorithm is used to optimize the results of human upper limb motion capture, improving the accuracy of joint point positioning. The coordinates of the upper limb joint points measured by the optimized sensor are connected to determine the position of the upper limb, and combined with the orientation of the upper limb, upper limb motion capture is achieved. The experimental results show that the method has high accuracy in measuring the orientation angle of the upper limbs and good motion capture performance.

The unstructured data structure of network measurement points is not clear. In order to improve the similarity of clustering, a mathematical modeling method for clustering the similarity of unstructured data of network measurement points is studied. Using the method of unstructured data network partitioning, the unstructured data of network measurement points is transformed into semi-structured data, obtaining a semi- structured data meta path. The semi-structured data is decomposed into two non negative matrices using the non negative matrix decomposition method. The non negative matrices are multiplied and fitted, and the regularization coefficient is introduced in the process to establish a comprehensive similarity rectangle on the original path of the semi-structured data. This enables the highly similar network measurement point semi- structured data to establish a similar cluster indicator vector and construct a similarity clustering mathematical model. After the model iteration, the clustering results are more reasonable and consistent. The experimental results show that this method can effectively convert unstructured data from network measurement points into semi-structured data. The clustering density of unstructured data from network measurement points in similarity clustering is high, and the NMI(Normalized Mutual Information) value is distributed in a higher area. Its clustering performance for network measurement point unstructured data is good. 

The data in the distributed network comes from different levels and devices, which makes the data have the heterogeneity of time series. Because the network state and traffic pattern change rapidly, it is difficult for the clustering results to accurately reflect the real distribution of data, affecting the accuracy of anomaly detection. Therefore, a cross-layer anomaly detection method for distributed network data based on improved extreme learning machine is proposed. Based on information entropy, the information volume of distributed network data is calculated to obtain the optimal probability distribution. Combined with sliding window technology, the weight attenuation function under the optimal probability distribution is constructed to achieve the distributed network data clustering, so as to accurately reflect the real distribution of data. The PReLU (Parametric Rectified Linear Unit) activation function is introduced to optimize the algorithm of the extreme learning machine, and the data after cluster processing is used as the input of the improved extreme learning machine to complete the cross-layer anomaly detection and improve the accuracy of anomaly detection. The experimental results show that the proposed method can effectively improve the cross-layer anomaly detection of distributed network data.

To accurately classify materials and improve industrial production capacity, a visual image enhanced automatic target recognition algorithm for material conveyor belts is proposed. Using a single scale algorithm to process the collected visual images. Image parameters are adjusted through Gaussian wrapping function, and restoring the optimal color of the visual image based on color constancy theory, ensuring the original color characteristics, completing color channel indexing based on image feature conditions, and enhancing the target visual image of the material conveyor belt. The Laplace operator is used to improve the scale space of the image, determine the extreme value points of feature pixels, establish a function to fit and recognize the strong edge response of the target, describe the distribution of key points of the target pixel through a histogram, obtain image gradient features and angle features, and determine the contour response value of the transmitted material through a sliding window. The features are trained to complete automatic target recognition. The experimental results show that the proposed method can effectively identify material targets on the conveyor belt, with high recognition accuracy and strong practicality. 

In order to solve the problem of difficulty in anomaly detection caused by multidimensional dynamic changes in wireless communication network data, a wireless communication network abnormal data elimination algorithm based on VAE-WGAN ( Variational Autoencoder-Wasserstein Generative Adversarial Network) is proposed. Principal component analysis is used to reduce the dimension of wireless communication network data, and wavelet transform is used to denoise the reduced dimension data. The VAE-WGAN model is built by using VAE module and WGAN module, and the data after dimensionality reduction and denoising are input into the model, and the model will output an anomaly score. When the anomaly score is greater than the anomaly detection threshold, the data will be identified as abnormal data and eliminated, so as to achieve the goal of eliminating abnormal data. The experimental results show that the data processing effect of the proposed algorithm is good, and the abnormal data can be effectively detected and accurately eliminated.

In order to solve the problem of user identity authentication in multi-application systems and multi- application scenarios, an extensible authentication center architecture with dynamic feedback is proposed. Combined with operation definition, intelligent evaluation and feedback mechanism, authentication gateway and authentication equipment are used to establish the reliable, complete and effective unified identity authentication capability between terminal application and information system resources. An example is given to verify the security, reliability and effectiveness of the design method of the system in practical application. It has certain reference value for the application of unified identity authentication, comprehensive audit and access control in the future.

To accurately identify abnormal data on the vehicle instrument panel during electromagnetic pulse effect test, a remote intelligent monitoring system for vehicle instrument with resistance to electromagnetic pulse interference is designed and implemented. The instrument design and dynamic simulation software SimHub is utilized to simulate the vehicle electromagnetic pulse effect phenomenon, corresponding to pointer deflection angle changes and abnormal alarm icon displays on the instrument panel. A video acquisition front-end based on embedded system development, optical network transmission and electromagnetic protection technology is designed. The electric field shielding effectiveness of the shell is analyzed using the three-dimensional electromagnetic simulation software CST ( Computer Simulation Technology Studio Suite). Real-time video acquisition of instrument panel and remote fiber-optic transmission under electromagnetic pulse environment are realized. An intelligent monitoring terminal software integrating dynamic tracking and feature enhancement is developed. The pointer deflection angle changes and abnormal alarm icon displays in the monitoring video streams of instrument panel are automatically identified. System testing results indicate that the average error in calculating pointer deflection angles is within 2°, and the recognition accuracy for abnormal alarm icon displays reaches 96. 3%. Remote intelligent monitoring of vehicle instrument electromagnetic pulse effect phenomenon is realized.

A portable grating spectrometer based on the Internet of Things is designed to visually demonstrate the working principle of the grating spectrometer in the undergraduate practical teaching process. The STM32 is used as the main controller. The off-chip circuits include the array photodiode light intensity signal acquisition module, the analog-to-digital conversion module, the stepper motor drive module and the Wi-Fi communication module. The optical system of the spectrometer is simulated, analyzed and optimized by using the optical design software Zemax. The software part is developed using JAVA and C/ C + + languages. Relying on the Keil development environment, the underlying program of the hardware system is written, and a spectral analysis platform based on wechat mini-program was developed as a supporting tool. To verify the system performance, a standard mercury lamp light source is selected for spectral testing, and the obtained spectral data is compared with the reference spectra of commercial spectrometers. The experimental results show that the measurement of this system are reliable and its performance meets the requirements of teaching demonstrations. This research provides a feasible solution for the development of a low-cost and highly demonstrable grating spectrometer for teaching purposes, which has certain value for promotion and application. 

A collaborative brain-computer interface integrated with an agent-based framework is presented to optimize work efficiency, with key challenges in work-state management for knowledge workers-specifically, inadequate real-time capability, lack of personalization, and poor integration between neuroscience tools and management workflows-being tackled. The system acquires EEG(Electroencephalogram) signals via an 8-channel OpenBCI(Open Brain-Computer Interface) device. Time-frequency images are constructed using short-time Fourier transform and differential entropy for emotion recognition, and then processed by an enhanced Pyramid SR-CNN ( Super-Resolution-Convolutional Neural Network) model for three-class classification ( negative, neutral, or positive), achieving 94. 01% accuracy on the SEED(SJTU Emotion EEG Dataset) dataset. For fatigue monitoring, a three-class classification ( normal, mild, or severe fatigue) is performed based on a weighted θ/β power ratio, with multi-channel spatial weighting strategies incorporated to improve robustness. A low-latency data pathway is established using the LSL(Lab Streaming Layer) and FastAPI WebSocket. The front end visualizes subject states via ECharts and invokes a COZE agent to generate periodic feedback reports. Six subjects performed cognitive tasks of varying intensity to validate system feasibility. Testing results show that the system effectively detects emotional fluctuations and fatigue states, significantly outperforming conventional behavioral monitoring methods. By generating intervention protocols adapted to individual EEG characteristics, the system enhances work efficiency and offers a scientifically grounded technological approach for managing the working states of knowledge workers.