To explore the feasibility of using femtosecond lasers for doping and modification of hBN( hexagonal Boron Nitride), for hBN films growing on sapphire substrates low-pressure chemical vapor deposition is used. The films are then irradiated with femtosecond lasers in an atmospheric environment. X-ray photoelectron spectroscopy results indicate that carbon and oxygen impurities from the atmosphere are incorporated into the hBN. As the laser energy density increased, the contents of carbon and oxygen impurities in hBN monotonically increased. The oxygen and some of the carbon impurities mainly occupy nitrogen vacancies in the hBN, acting as donors and acceptorsing respectively, resulting in impurity compensation. With the increase in carbon concentration, some of the carbon impurities exist in the form of clusters or interstitial atoms. The resistivity of doped hBN samples significantly decreased, reaching up to 1 / 800 of that of undoped samples. These results demonstrate that femtosecond laser technology can be used for doping and controlling the electrical properties of hBN. 

At present, the majority of neural network-based radar modulation signal recognition algorithms predominantly rely on a single source of information, overlooking the potential benefits that arise from the synergistic use of multi-modal information features. To tackle this limitation, a novel multi-modal parallel feature fusion model has been proposed, which leverages both one-dimensional signal sequences and two-dimensional time-frequency representations. Initially, the temporal feature extraction module incorporates a two-dimensional temporal change modeling approach to capture temporal dynamics, while the frequency domain feature extraction module employs an inverse residual structure with a band linear bottleneck layer to extract spectral features. Subsequently, the integration of two distinct attention mechanisms, along with residual connections, facilitates an effective fusion of multi-modal features, enhancing their complementary nature. Empirical evaluations conducted on DeepRadar2022 and self-built datasets demonstrate that this model provides a more comprehensive feature representation achieves higher classification accuracy and exhibits noise resilience, making it a promising solution for advanced radar signal processing applications.

Electromagnetic wave is highly theoretical and difficult to understand, therefore an electromagnetic wave underground target detection experimental system is designed based on LabVIEW. The functions of precise antenna positioning, automatic real-time data acquisition and imaging are realized. In the operation of the system, high-frequency electromagnetic waves are transmitted to the ground via a transmitting antenna, with the echoes reflected from underground received by a receiving antenna subsequently. On the basis of the differences in the electrical properties of underground media, the collected echo signals are processed and analyzed systematically. By analyzing the results of numerical simulation and physical simulation, the electromagnetic response characteristics of the underground target can be mastered, and the spatial location, structure and burial depth of the underground medium can be inferred. This experimental system is convenient for students to understand and master abstract concepts, stimulate students’ enthusiasm for learning and interest in scientific research and exploration. 

In order to improve the control scheme of PSFB ( Phase-Shifting Full Bridge) converter and obtain better control indicators, a fuzzy PID( Proportional Integral Derivative) control method is proposed based on IGWO ( Improved Grey Wolf Algorithm) optimization. The fractional impedance components in the PSFB converter circuit have non integer order characteristics. Using fractional calculus theory and state space averaging method, the fractional order phase-shifting full bridge converter is modeled and optimized for control analysis. Adaptive inertia weight and Euclidean distance method are used to optimize the global characteristics and convergence of GWO, and then the quantization factor of the fuzzy controller calculated to optimize the control effect. The IGWO optimized fuzzy PID control scheme is used in PSFB converters to compare with two traditional schemes, and a 50 W prototype is designed to verify the effectiveness of the control strategy. The simulation results show that the fuzzy PID control optimized by IGWO reduces the time savings, overshoot, and steady-state error 98. 18% , 99. 63% and 44. 44% , and 66. 67% , 33. 33% , and 89. 13% , respectively, compared to the traditional two PID control methods. The experimental results of fuzzy PI ( Proportional Integral ) control simulation verification based on hardware IGWO optimization of TMS320F28034 show that it has better control accuracy and stronger anti-interference ability compared to fuzzy PID.

In view of the problems of speed sensor dependence and parameter drift in permanent magnet semi- direct drive pumping unit, a vector control method combining speed sensorless and parameter identification is proposed. A sliding mode observer is designed to replace speed sensors such as rotary transformers, and a model reference adaptive system model is constructed based on Popov hyperstability theory. The closed-loop system architecture of stator resistance, inductance and flux parameter identification is established synchronously. The experimental results show that the proposed method can accurately estimate the motor speed and dynamically track the parameter change through the model reference adaptive algorithm. Compared to the control scheme that relies on the rotating transformer, the new system eliminates the hardware cost of the sensor and significantly improves the system parameter estimation ability through the double closed-loop coordination of the sliding mode observer and parameter identification. The design provides a new solution for the stable operation of the oil pumping unit under harsh working conditions. 

To simultaneously meet the multi-objective requirements of reducing the comprehensive cost of a microgrid integrating wind, solar, hydrogen, and storage, improving the renewable energy consumption rate, and enhancing the utilization rate of sources and loads, a multi-objective optimization model is established for microgrid based on its output model, with economy, renewable energy consumption rate, and source-load utilization rate as optimization objectives. A multi-strategy improved HBA ( Honey Badger Algorithm ) is proposed. Finally, Using real off-grid data from Northwest China tests are conducted, and results show the improved HBA converges are faster and more stable. Unlike single-objective approaches yielding one solution, this model provides 36 Pareto-optimal options for flexible decision-making.

To address the challenges of slow tracking speed and low accuracy in MPPT(Maximum Power Point Tracking) under PSC( Partial Shading Conditions) where PV( photovoltaic) arrays exhibit multi-peak power characteristics. A hybrid algorithm combining an improved GWO(Grey Wolf Optimizer) with a variable-step INC( Incremental Conductance) method is proposed. First, a peak voltage initialization strategy is incorporated into the GWO by analyzing the voltage positions corresponding to power peaks. Second, a nonlinear convergence factor is introduced to enhance the GWO’ s global search capability. The hybrid approach employs the modified GWO for global exploration and then switches to a variable-step INC method ( adjusted by dP / dU) for precise local refinement. MATLAB / Simulink simulations demonstrate that the proposed algorithm significantly improves tracking speed and accuracy under both static and dynamic PSC while reducing output power oscillations. 

Insulator faults in transmission lines directly threaten power grid stability. To accurately identify insulator faults, an insulator fault diagnosis method based on an improved YOLOv3(You Only Look Once v3)- Tiny framework is proposed. Firstly, a channel-spatial attention mechanism is integrated into the feature extraction network to enhance critical feature capture capabilities. Subsequently, a CSP-RFB ( Cross Stage Partial-Receptive Field Block) module is designed to improve small-target detection performance while reducing computational complexity. Finally, a novel loss function is adopted to optimize localization accuracy. Experimental results demonstrate that the enhanced YOLOv3-Tiny algorithm achieves up to 97. 4% MAP(Mean Average Precision) in insulator fault detection, significantly outperforming the original YOLOv3-Tiny model.

With the accelerated promotion of China’s new energy power system, large-scale grid connected operation of new energy units will induce voltage instability risks in the power system. To address this technical challenge, a SVG( Static Var Generator) can be used to output reactive power to enhance the voltage support capability of the power grid. However, its optimization configuration method still needs further research. To solve this technical problem, the grid type SVG with autonomous voltage regulation capability enhances the voltage support capability of the power grid by outputting reactive power, but its optimization configuration method still needs further research. In response to this demand, the MRSCR( Multiple New Energy Station Short Circuit Ratio) is uesd as an indicator for assessing voltage stability and develops a quantitative model to analyze the voltage support capability of the power grid, and focuses on revealing the mechanism of the improvement of MRSCR index by grid connected SVG. Based on the constraint of voltage stability critical threshold, a configuration model with device capacity optimization as the objective is constructed, and a multi-objective optimization configuration method for grid type SVG considering voltage stability is proposed. The simulation results show that the proposed configuration method can significantly improve the voltage support characteristics of the system nodes in the transient process.

Traditional methods like molecular docking often face high time costs or infeasibility in large-scale virtual screening tasks. To address this problem, a parallel graph Bayesian optimization framework incorporating incremental learning is proposed to efficiently handle such tasks. The method utilizes a deep graph Bayesian optimization framework for screening and employs parallelization to enable flexible deployment across multiple computational nodes on various servers, significantly improving computational efficiency. To tackle the issue of long surrogate model training times, an incremental learning strategy is introduced, along with an exponential moving average mechanism and a replay mechanism to mitigate catastrophic forgetting in incremental learning. Experimental results demonstrate that the framework can identify over 96% of the optimal molecules by docking only 6% of the molecular library. When deployed on four computational nodes, the parallel framework reduces time costs by 71% compared to the serial framework. With the incremental learning strategy, the total runtime is further reduced by 13. 8% , while still identifying 93. 7% of the optimal molecules. The proposed method significantly reduces the time cost of virtual screening while maintaining high screening performance. 

Aiming at the issue of reduced efficiency caused by separate use of first-principles methods and machine learning models in crystal structure prediction, which hinders the effective utilization of information provided by machine learning models, a stable crystal structure prediction algorithm based on multi-fidelity Bayesian optimization is proposed. The algorithm models the potential energy surface of crystal structures through a surrogate model, and the acquisition function selects sampling points along with their corresponding evaluation fidelity based on the modeling results of the potential energy surface. The evaluation function then assesses the selected sampling points, and the evaluation results are used to update the surrogate model. Upon meeting the termination criteria, the algorithm ceases iteration and outputs the final predicted stable crystal structure. Experimental results demonstrate that the proposed algorithm effectively leverages the information from machine learning models, ensuring both the accuracy and quality of the final prediction results while achieving higher efficiency.

Existing Bayesian optimization algorithms often suffer from high computational complexity and suboptimal efficiency in locating the global optimum. To address these limitations, an IBO( Improved Bayesian Optimization) algorithm is proposed. First, the MCMC(Monte Carlo Markov Chains) algorithm is improved by using a Gaussian proxy function to improve the proposal function, while adding a hyper-prior to the prior function, which simplifies the complexity of the internal hyper-parameter optimization computation of the Gaussian process, and improves computational efficiency. Second, Gaussian modeling of the loss function with respect to the dataset size in the hyperparameter optimization process of CNN(Convolutional Neural Networks) is proposed to enable the model to adaptively select the dataset size to optimize the hyperparameters, so that the combination of hyperparameters that minimizes the loss function can be found using fewer datasets. The improved Bayesian optimization algorithm is tested using the Branin-Hoo function, which proves that the IBO algorithm is able to find the optimal value in the shortest time. Using IBO-CNN for fault diagnosis on PU( University of Paderborn) dataset and comparing with other hyper-parametric optimization algorithms, the results prove that the IBO algorithm can find the minimum value of the loss function more quickly, so that the training process converges quickly, the diagnostic accuracy is higher than the other algorithms by about 0. 5% to 3% , and it exhibits a good fault diagnostic performance on dataset under different working conditions. Thus it is proved that the algorithm has higher computational efficiency than other optimization algorithms. 

To address the issue of low system integration and resource sharing levels in existing shipbuilding enterprises, which leads to difficulties in querying and tracing process and component information during actual production, a shipbuilding identification information integration and blockchain traceability system is proposed, based on the practical traceability requirements of shipbuilding. A comprehensive information identification system based on the production process is designed for the shipbuilding scenario to achieve unified identification of construction information. Blockchain technology is used for centralized management of construction data, and a fast blockchain-based keyword query traceability mechanism is introduced to meet the traceability needs of shipbuilding. This mechanism locates the last relevant block through an auxiliary database and utilizes the category keyword field in the block header to enable jump queries between blocks. Experimental results demonstrate that the system accurately traces shipbuilding information and significantly improves query traceability performance, with an average performance improvement of 2. 07 times. 

To solve the overfitting problem caused by leveraging pedestrian detail features to reduce the modality gap between visible and infrared images in the visible-infrared person re-identification task, an end-to-end improved CNN ( Convolutional Neural Networks) based on data augmentation techniques and detail feature extraction methods is proposed. With the goal of reducing the modality gap by generating diverse embedding features while simultaneously matching pedestrian details, an efficient dual-branch attention module is designed to learn more informative feature representations, and a triplet loss function with data augmentation is proposed to alleviate overfitting. Extensive experiments on the public datasets SYSU-MM01 and RegDB demonstrate that the proposed method outperforms other methods, effectively mitigating the overfitting problem caused by attention mechanisms and improving the accuracy of person re-identification.

Due to the explosion of online diagnosis and treatment information, the processing and application of diagnosis and treatment information are greatly increased, which greatly hinders the development of online diagnosis and treatment system. Therefore, the research on multi-dimensional feature extraction method of online diagnosis and treatment information based on data mining technology is proposed. The pretreatment online diagnosis and treatment information is filled by cleaning, integration and missing values. The data mining technology-ant colony algorithm is introduced to multi-dimensionally cluster online diagnosis and treatment information. Based on the data mining technology-convolutional neural network, the multi-dimensional feature extraction framework of online diagnosis and treatment information is formulated, and the multi-dimensional features of online diagnosis and treatment information are effectively extracted through the synergy of convolution layer, pooling layer and full connection layer. The experimental results show that there are significant boundaries between the dimensions of patient basic information, patient symptom information, patient treatment process record information and patient follow-up information in the multidimensional clustering results of online diagnosis and treatment information obtained by the proposed method. The characteristics of patient treatment process record information and patient treatment time are consistent with the actual results, which can effectively improve the efficiency and accuracy of diagnosis and treatment information processing and provide practical technical support for the development of online diagnosis and treatment system.

The topology of sensor network routing changes dynamically, the boundary between normal and abnormal is blurred, the sources of error information are complex, and traditional intrusion detection is difficult to distinguish between real intrusion and temporary faults. To address this challenge, a novel intrusion detection method for anomalous nodes in sensor networks based on roughness improved bee colony algorithm is proposed. This method aims to accurately identify and distinguish abnormal behavior, reducing the false alarm rate caused by system dynamics. Roughness function is introduced to extract and screen intrusion data feature values of sensor network nodes, intrusion detection objective function is constructed using coefficient matrix, bee colony algorithm is improved with information entropy, new fitness function is constructed, important features are distinguished, real intrusion and temporary system faults are identified, and intrusion detection of abnormal nodes in the network is achieved. The test results show that the average accuracy of this method for detecting abnormal node intrusion is 96. 0% , the average missed detection rate of node intrusion is 1. 3% , and the average detection delay is 0. 25 s. The proposed method can effectively cope with the detection difficulties caused by the dynamic changes in the routing topology of sensor networks, accurately distinguish between real intrusion and temporary system failures, and demonstrate the advantages of high accuracy, low missed detection rate, and short detection delay in detecting abnormal node intrusion, providing reliable guarantee for the safe and stable operation of sensor networks. 

 In the field of laparoscopic surgical instrument detection, aiming at the insufficient utilization of multi- scale features in the existing object detection networks and the limitations of the existing detection models in adapting to downstream tasks through fine-tuning methods, a laparoscopic surgical instrument detection algorithm based on the improved Retinanet is proposed. The algorithm includes an adaptive prompt fine-tuning module that promotes the efficient adaptation of the model to downstream tasks, the DR-UNet(Deep Residual UNet) module and the RFA(Residual Feature Augmentation) module that enhance the model’s ability to capture information of targets of different sizes. Experiments on the laparoscopic surgical instrument detection dataset show that the proposed algorithm achieves 1. 1% improvement over the baseline model at an IOU(Intersection over Union) of 0. 5, with an mAP @ 0. 5 of 97. 3% . The detection performance is superior to many other state-of-the-art methods, demonstrating significant value and importance in practical applications such as surgical instrument detection. 

Traditional GNNs(Graph Neural Networks), while capable of representing interactions between users and items, are typically limited to pairwise binary relationships, making it challenging to fully model high-order associations between users and items. To address this issue, a GR-HGAN(Global Relational-aware Hypergraph Attention Network for Recommender Systems) is proposed. First, GR-HGAN adopts a hypergraph structure that allows a single hyperedge to simultaneously connect multiple users and items, naturally capturing high-order associations in multi-party interactions. Then, GR-HGAN uses a hypergraph attention mechanism to focus on the relevance of different neighboring nodes to the target node, enhancing the influence of significant nodes in feature aggregation. Finally, by integrating a message-passing mechanism from both the global hypergraph and local subgraphs, it effectively learns high-order associations between nodes from a global perspective while incorporating critical information from local subgraphs, further improving the representation capability of node embeddings. Extensive experiments conducted on widely used datasets such as Yelp, MovieLens, and Amazon demonstrate that GR-HGAN outperforms state-of-the-art recommendation methods. 

The current access permission control methods for e-commerce websites are not secure. Illegal users can directly access e-commerce websites, resulting in a relatively small number of transactions successfully processed by the system within a unit of time and an abnormally large number of users accessing e-commerce websites. To solve this problem, an algorithm for the access permission control mechanism of e-commerce websites under the optimization of smart contracts is designed. The optimized architecture of the smart contract for controlling the access rights of e-commerce websites is designed, and the smart contract in the initialization stage, encryption stage, decryption stage, signature stage, verification stage and access stage is deployed. The ciphertext of e-commerce user data is decrypted using a temporary symmetric key to obtain the personal data of e-commerce users, and the information sent by e-commerce users is obtained by combining the authentication method based on non-interactive zero-knowledge proof. Based on the first-level security authentication password of the e-commerce website, an Ethereum address is restored and the access rights of the e-commerce website is controlled through the credit fuzzy layer analysis results. It can be known from the experimental results that the maximum number of transactions successfully processed by this algorithm is 12 x 10? bit / s and the minimum is 5x 10? bit / s, and it can effectively control the user access rights. 

To address the low efficiency of manual summarization caused by lengthy texts, complex topics, and excessive redundant information in ideological and political case texts amid the information explosion, and the limitations of existing methods-purely extractive models overlook the logical connections between sentences, while purely generative models tend to deviate from policy guidance, a CLUSTER2SENT(CLUSTER to SENTence) algorithm is proposed. This algorithm realizes the automatic generation of summaries for ideological and political cases through a three step process: 1) extracting important utterances relevant to each part of the summary, 2) performing clustering analysis on the extracted important utterances, 3) generating summary text for each cluster. Experimental results show that the CLUSTER2SENT algorithm outperforms its purely extractive counterpart by 8 percentage points in the ROUGE-1 metric, which verifies the effectiveness of the algorithm. The study indicates that constructing a section structure when building a summary corpus can significantly improve model performance.

 In response to the inefficiency caused by the reliance on manual measurement in existing road crack classification methods, a road crack detection model is proposed. This model improves detection accuracy by enhancing the detection algorithm and employs a bisection method to precisely measure the actual width of cracks, thereby enabling automatic classification and grading of crack damage levels. Specifically, a COTECANet ( Contextual Transformer Efficient Channel Attention Network ) model based on the ResNet50 architecture is first introduced, which outperforms other compared deep learning models. For the detection results of this model, the maximum inscribed circle radius of the crack contours in the image is calculated using the bisection method for roads with pavement cracks, thereby obtaining the maximum pixel width of the road cracks. The actual width of the pavement cracks can be derived by converting the measurements according to the corresponding scale, and the damage level of the road cracks is classified and graded based on national standards. Experimental results demonstrate that the COTECANet model can effectively detect pavement cracks, achieving an accuracy rate of 99. 8% in road crack identification. The above method provides more scientific and efficient technical support for road maintenance, with significant theoretical and engineering application prospects.

Considering the characteristics and complexity of multidimensional data streams, in order to fully utilize parallel computing resources and ensure the scalability of the algorithm, a parallel mining algorithm for frequent patterns of multidimensional data streams in Hadoop environment is proposed. Design a Hadoop data stream processing platform based on HDFS ( Hadoop Distributed File System) and MapReduce, propose an HpFitStream clustering algorithm based on feature projection and fitting, using the polynomial fitting algorithm to handle abnormal data streams, and reducing the dimensionality of the processed data streams through feature projection to reduce computational costs. Implement frequent pattern parallel mining of multidimensional data streams in Hadoop environment using PFPonCanTree algorithm. The experimental results show that the proposed method can effectively reduce computational complexity while improving the scalability and load balancing ability of the algorithm.

During holidays, the highway traffic pressure increases, and the traffic data collected directly is prone to the defects of data loss and error, which further affects the accuracy of prediction. Therefore, the prediction algorithm of holiday highway traffic congestion under the Internet of Things technology is proposed. The algorithm collects holiday road traffic data in real time through the Internet of Things technology, and carries out correlation analysis and sets up relevant intersection sets. Combining the traffic-occupancy model and radial basis function neural network model, the traffic flow is repaired, and each road section is divided by K-means clustering analysis. By constructing the compression matrix of road network data, the mapping relationship between the original data matrices of road network is realized, and the traffic prediction results of all sections of the whole road network during holidays are obtained. The experimental results show that the CE(Coefficient of Efficiency) of the proposed algorithm is significantly close to 1, which is closer to the actual observation value. It proves that the proposed algorithm has good performance in highway traffic congestion prediction and can effectively realize highway traffic congestion prediction on holidays. 

Due to the complex density overlap and texture interference between prohibited items and background materials in subway security X-ray images, the feature representation ability is insufficient, making it difficult to effectively distinguish prohibited items from normal items. The traditional methods are prone to losing key spatial information of small prohibited objects during feature extraction, ultimately leading to serious missed and false detections in detection systems. To address this issue, a subway security X-ray prohibited object intelligent recognition algorithm based on feature extraction and enhancement is proposed. A multi-scale feature extraction framework is constructed based on improved SSD-VGG16(Single Shot MultiBox Detector-Visual Geometry Group 16). The ability to extract microscopic features of prohibited objects is enhanced by adding Conv3 _3 detail capture layer and Conv5_3 small object sensitivity layer, and integrating semantic information from Conv4_3 and other basic network layers using feature fusion technology, significantly improving the completeness of feature representation; On this basis, a spatial attention mechanism is introduced to obtain X-Y bidirectional attention vectors by decomposing and aggregating features, effectively focusing on key areas of prohibited items. At the same time, an ECA( Efficient Channel Attention) channel attention module is embedded to implement cross channel interactive learning, achieving dynamic enhancement of discriminative features of prohibited items; By using the DIoU-NMS (Distance-Intersection over Union Non-Maximum Suppression)algorithm to comprehensively consider the target box overlap rate and center distance for optimization screening, the missed detection rate in dense scenes is significantly reduced; By using adaptive threshold segmentation method and combining Wiener filtering and median filtering preprocessing techniques to eliminate image noise interference, accurate area segmentation of prohibited objects is achieved based on grayscale or pseudo color distribution characteristics, thereby realizing X-ray prohibited object recognition. According to the experimental results, the pixel brightness corresponding to the metal knife, fire machine, and glass bottle recognized by the algorithm is 255, 153, and 51, respectively, which is consistent with the experimental indicators and can accurately identify various prohibited items.

In the process of low light digital photography image processing, a guided filtering based image enhancement algorithm is proposed to avoid image distortion and detail loss. Firstly, the brightness component area of low light digital photography images is divided, the classic gamma correction algorithm is optimized, and then a sliding window is used to convert the RGB(Red, Green, Blue) mode of the original image to HSV(Hue, Saturation, Value) mode. The position and radius range of the sliding window is adjusted in real time to ensure that the gamma value is within the optimal range and retain the local information of low light digital photography images. By guiding filtering, the image is divided into a base layer and a detail layer, and processing methods such as segmentation clustering, compression mapping, and contrast function adjustment, and fractional order differential masks are used to enhance the base layer and detail layer. Finally, the basic layer and detail layer are fused through fusion factors to obtain the enhanced image. The experimental results show that the algorithm can better highlight the detailed edges and enhance the details of the image, and can achieve good image enhancement effects. 

 In order to improve the stitching accuracy of images while ensuring image resolution, a new stitching algorithm is designed for digital photographic images when establishing image scenes in virtual reality environments. In a virtual reality environment, the overlapping area of the digital photography image to be spliced is determined, and the feature points are detected within it. Then, by determining the extreme points in the image scale space, unstable feature points in the overlapping areas are filtered out, and feature points are matched. Finally, based on the matching results, precise stitching of digital photographic images is achieved. According to the experiment, compared to traditional algorithms, using this algorithm to concatenate images results in better information entropy and better stitching effect. 

 In order to enable enterprises to quickly obtain the required digital media resources from the massive data resources in the multi service value chain, a digital media resource recommendation algorithm for the multi service value chain is designed. A dataset of digital media resources is established, the concept of meta learning is integrated, optimal clustering method is used to partition the attributes of digital media resources in different time periods, the probability of users being divided into the same cluster through consistency matrix is calculated, and a multi service value chain digital resource recommendation objective function is constructed under phase space reconstruction. The potential Dirichlet allocation method is introduced to solve the objective function, and penalty factors are used to handle the long tail problem of digital media resource recommendation. Resources are arranged according to user interest levels to achieve high-precision digital media resource recommendation. The experimental results show that the proposed algorithm has high accuracy in user preference evaluation and recommendation coverage, effectively improving the quality of digital media resource recommendations and bringing new opportunities for the healthy development of enterprises. 

The current research on mechanical fault detection of ultrasonic instrument trajectory balls requires dividing multiple categories in feature classification to complete fault diagnosis. Although the multi class pattern can improve diagnostic efficiency, this multi class fault recognition pattern is prone to cyclic iterative comparisons of similar faults between classes, resulting in a decrease in recognition accuracy. Utilizing the feature of fusible spectral features of mechanical faults in trackballs, a diagnostic method using ultrasonic instruments under unified spectral features is designed. By using spectrum analysis technology to perform spectrum analysis on the trajectory ball signal of ultrasonic instruments, high-precision amplitude and phase spectra are obtained. By fusing signal amplitude and phase information, a two-dimensional holographic spectrum is constructed on various monitoring cross-sections of the ultrasonic instrument trajectory ball to extract fault signal features. By randomly assigning labels to each sample based on the extracted features, a sample library is constructed to train a deep neural network (DNN: Deep Neural Network). After DNN training and testing iterations, fault samples with similar features are aggregated into the same class, ultimately achieving fault diagnosis. The experimental results show that the proposed method can accurately diagnose mechanical faults in the trajectory ball of the ultrasound instrument, ensuring the stable operation of the ultrasound instrument. 

The existing electronic scanning archive information denoising methods are difficult to cope with complex noise such as ink stains, greases, and paper textures mixed in the scanned archives. During the denoising process, it can cause blurry edges of the text. An adaptive unscented Kalman filter denoising algorithm is designed for electronic scanning of archival information in libraries. is divided The electronic scanned archive information of the library is devided into multiple sub regions and the information entropy of each sub region is calculated. Linear transformation is performed on sub regions with low information entropy, pixel diversity is increased by adjusting the range and offset of pixel values, and the information entropy of the electronic scanning archive area in the library improved. Based on information entropy optimization, the library electronic scanned archive information is denoised using an adaptive unscented Kalman filter denoising algorithm. The test results show that the method preserves more details in the processed images while effectively removing noise, making the overall image clearer. The TLI(Text Legibility Index) value of the design is generally higher than 0. 96, and the HPF(Historical Feature Preservation) value of the design method reaches a maximum of 0. 97, indicating that the algorithm can provide effective technical support for the digital protection and utilization of electronic scanned archives in libraries.