This PDF file contains the front matter associated with SPIE Proceedings Volume 12306 including the Title Page, Copyright information, and Table of Contents.

In order to improve the estimation performance of the mixed uncorrelated and coherent signals at strong impulsive noise, a novel direction of arrival (DOA) estimation method for mixed independent and correlated signals is introduced. By applying the median value filtering method to the array received data, the amplitude of the impulsive part of data is normalized to a relatively small level. Since the preprocessing process cut out the strong impulsive part of noise effectively, the common secondary statistic DOA estimation algorithm can be adopted, and the DOAs of the independent and correlated signals can be estimated, respectively. Theoretical analysis and simulation results verify the effectiveness of the proposed method under strong impulsive noise as well as the excellent DOA estimation performance.

The UPF (user plane function) in 5G core network carries the user data of 5G mobile communication system, so flexible and reasonable UPF selection is of great significance. Aiming at the problem of insufficient flexibility and low efficiency of UPF selection in 5G core network, this paper proposes a UPF selection method based on knowledge graph, and implements and tests the method based on Mininet platform. The measure is the time the user sets up the PUD session. The results show that in different edge computing scenarios, the UPF selection method based on knowledge graph can meet the specific needs, improve the flexibility and efficiency of UPF selection, and reduce the signaling interaction process.

In order to solve the problem of wireless channel congestion caused by the large number of wireless collection nodes with high throughput in the shockwave-field wireless collection system, this paper proposes a scheduling algorithm suitable for shockwave-field wireless collection nodes. Firstly, the algorithm uses the shock wave propagation model to set the grouping weight for each node. Secondly, the algorithm is based on the GWO (Gray Wolf Optimization) algorithm and uses the group weight proportional fair strategy to guide the evolution of the gray wolf population, and finally achieves the goal of reducing congestion delay and improving the total throughput of the system. System simulation results show that the access strategy based on this algorithm reduces 45.71% of the nodes average waiting delay and increases 33.9% of the system throughput compared with sequential access strategy in IEEE802.11 wireless network with PCF (Point Coordination Function. It effectively reduces the delay and congestion in the shockwave-field wireless collection system and improves nodes’ ability for data responsiveness during parallel collection.

Medical images contain a large amount of private information like patient's lesions. Theft or destruction of medical images will cause irreparable losses to patients and medical institutions. How to realize the secure transmission of information in the process of telemedicine has become a research hotspot. In this paper, an encryption algorithm for hyper-chaotic systems based on DNA coding is proposed. First, a four-dimensional chaotic system is constructed by adding a dimension to the three-dimensional chaotic system, which improves the dynamic performance of the system, increases the parameters of the chaotic system, and improves the key space. Second, the algorithm complexity is increased by the added DNA encoding. The experimental results show that the key space encrypted by the method in this paper is 2469, and the key sensitivity can reach 10-16. At the same time, in the analysis of histogram, adjacent correlation and information entropy, the algorithm in this paper can achieve good security performance.

Concerning the monitoring of land use and cover changes, it is unable to achieve rapid intelligent dynamic monitoring through the traditional field survey method, though it can accurately reflect the condition of various types of ground objects. The paper uses remote sensing estimation technology to monitor change information and proposes reliable land use and cover change monitoring methods. First, the SPOT original images of two phases in the land cover change monitoring period are subjected to such preprocessing as geometric correction and image fusion to obtain remote sensing images with similar imaging conditions in different periods of the study area. Second, a change identification model is built through the spectral signature variation method, the image interpolation method, the principal component difference method and the classification and comparison method to get land cover change areas. Finally, the threshold method and the supervision and classification method are used to extract and qualitatively analyze land-use change identification information. Then a quantitative evaluation of the accuracy is performed. The adopted change information monitoring methods based on remote sensing can meet the requirements of monitoring with high accuracy, thereby achieving accurate automatic extraction and qualitative analysis of land use and cover change information.

In order to achieve the optimal balance between task execution delay and energy consumption in Mobile Edge Computing (MEC) networks. First, the Analytic Hierarchy Process (AHP) is adopted to classify the priority of all tasks, so as to establish a related model for task offloading strategy and weight allocation of resources. Then, a multi-task offloading algorithm based on DNN is introduced to generate offloading strategies using multiple DNNS. Meanwhile, training samples composed of offloading strategies and input data are stored through the experience pool. These training samples will be used to train DNN. Simulation results show that the accuracy of the proposed multi-task offloading algorithm can reach 0.99, and the total delay of task processing and system cost can be effectively reduced compared with the three comparison algorithms.

In order to meet the needs of human respiratory signal acquisition and monitoring in the field of respiratory medicine, a portable and wearable human respiratory detection system is designed. The system collects human breathing signals, and can display the breathing signals in real time to realize the visualization of breathing data and the determination of abnormal values. The experimental results show that the device can solve the problem of real-time monitoring of human breathing state and has good applicability.

This paper studies the current research status of image inpainting, and proposes the Unet network to solve the problem of image holes of objects wearing reflective balls captured by the Azure Kinect camera and other damaged images that extend to the field.

Towards solving the problem that ultrasonic ranging studied in the past was difficult to reach sub-millimeter accuracy in long-distance ranging. In this paper, after analyzing principles of ranging, a system based on the cross-correlation algorithm and an ultrasonic synchronous transceiver platform had a 192 kHz high sampling rate was designed. First, an ultrasonic transceiver platform was designed as a USB audio system with a sound card as the core. Second, a reverse design idea was introduced into the FPGA software to design the accurate clock system, to achieve 192 kHz high sampling rate and transceiver synchronization. Third, to improve the accuracy and stability of ranging, a valid method was proposed to eliminate the jitter of real-time cross-correlation peaks based on this platform. Finally, we conduct extensive experiments, the results show that the range median error is only 0.95 mm in the range of 4.2 m.

In order to solve the limitation of industrial bus application scenarios such as 1553B, the paper proposes an intelligent communication module based on Loongson 2K1000 processor as the main controller. The communication module collects the data on the 1553B separately, analyzes, filters and repackages the data packets through the Loongson 2K1000 processor, and finally uploads them to the task machine through the network. The network adopts a redundant design to ensure that the data is uploaded stability. As the same time, the communication module parses the data received by the network into the corresponding 1553B message, and completes the protocol conversion function of the network to 1553B. The localization rate of the intelligent communication module hardware proposed in the paper reaches 100%, the operating system runs a lightweight domestic embedded sylix os operating system, and the onboard mass storage unit can expand the storage capacity according to different application scenarios. Through functional and stress tests, the functional correctness and performance stability of the intelligent communication module proposed in this paper are verified.

Battlefield electromagnetic environment is an important space foundation of modern military operations, and electromagnetic environment simulation has become the focus of military combat environment simulation. The combatants can't use radar to train contraposing combat environment in areas of combat while modern military simulation training. This paper has studied the simulation of typical area, operational goal setting and electromagnetic environment simulation through the adoption of digital technology. The air electromagnetic environment simulation system has been designed and the design and implementation of the situation steering control has been discussed emphatically.

Combined with the next-generation Internet system ICN (information centric networking), data caching in satellite broadband internet services can improve the efficiency of content distribution. Aiming at the problem of wasting storage due to cache homogeneity in LRU (Least Recently Used) which is the most widely used cache scheme, the LRU hit ratio of single node and the cache degradation problem in multi node network are evaluated. Then 2qLRU-Cache is proposed, which can enhance the cache hit ratio by filtering the data hierarchically with popularity and moving hot content close to end users. The simulation results with theoretical distribution and real world trace show that compared with LCE (Leave Copy Everywhere) + LRU and LCD (Leave Copy Down) + LRU, 2qLRU-Cache improves the network cache hit ratio and reduces the transmission cost.

The video processing module is one of the most important components of the display and control computer, providing a general video processing platform for the display and control computer. The 4-channel video processing module designed in this paper supports 4-channel 4k HDMI video input, 4-channel 4k image and video simultaneous encoding and decoding, and 4-channel 4k HDMI video signal output. Each HDMI video channel supports 9-channel picture segmentation and splicing, and each picture supports compressed video decoding with a display resolution of 1920×1080p@30fps, and the code algorithm can be configured as H.264 and H.265. The video processing module receives and sends external video data through the internal dual redundant Ethernet, which greatly improves the reliability of the system. Dual redundant Ethernet switching only takes 40ms, and the code and decode system delay only needs 180ms at 3840×2160@30fps model.

Polar codes are the first channel coding method that has been strictly proven to reach the Shannon limit and are adopted into the 5th Generation Mobile Communication Technology (5G) standard with its excellent performance. The Successive Cancellation List (SCL) is a highly practical algorithm to decode the polar codes. In this work, it is proved that there are some unnecessary calculation processes in the traditional SCL decoding algorithm by formula derivation. New algorithms are proposed for two nodes in SCL decoding algorithm. The performance of SCL decoder is enhanced by removing some redundant parts without sacrificing the efficiency. The reduced SCL decoding algorithm is effective for codes with any length and rate, which can reduce decoding latency from 54.2% to 67.0%.

Satellite network has been used as Internet access and natural disaster warning due to its advantages of wide communication coverage, fast construction speed, and stable during geological disasters. However, due to the high cost of building a satellite network, researchers still use software satellite network simulators to simulate the propagation delay, communication rate, and packet loss rate of the satellite link. Limited by the CPU performance, software network simulators have limitations in simulation authenticity and network scalability. Firstly, it is difficult for the software network simulator to load the actual load program to simulate the influence of the satellite network on the real data interaction. Secondly, as the number of end nodes and network nodes increases, the simulation time of the software network simulator will increase significantly. In this paper, we present an FPGA-based satellite network simulator, namely, SanSim, which has two advantages. First, SanSim utilizes FPGA's good programmability and high processing performance, and can process real data traffic in real time by connecting the actual satellite load, and the simulation reality is higher. Second, SanSim adopts the Match-Simulation Action abstraction. It is easy to expand the number of network simulator nodes and network links. Finally, we implemented the SanSim prototype on the FPGA platform, and then deployed two simple satellite network topologies. The experimental results verified the feasibility and efficiency of SanSim.

In order to meet the processing requirements of network digital audio and analog audio, as well as the needs of multisource and multi-standard audio signal amplification, volume adjustment, and sound and light alarms in important scenarios such as shipborne, vehicle-mounted, and airborne, it closely tracks sonar, navigation, etc. Century application requirements of other scenarios. The audio filtering, digital volume control algorithm, etc. are studied, and a multi-source multi-standard audio processing system is designed and implemented, which realizes multi-channel network real-time digital audio reception, signal amplification, and volume adjustment. The design follows generalization and modularization. , Serialized design ideas to ensure better applicability and higher reliability. After the design is completed, the analog audio filter circuit is simulated. The amplification factor is 31dB, the attenuation at 100Hz is 32.9dB, and the in-band fluctuation is 1.59dB, all of which meet the design requirements. The multi-source and multi-standard audio processing system supports the reception of 20 channels of real-time digital audio in the mono network, real-time processing, and independent selection of the audio signal to be broadcast. The delay jitter does not exceed 2 ms, and the system functions and performance reach the expected goals.

With the widespread application of deep learning methods, multimodal techniques have also achieved rapid development. Since single-modal speech recognition may affect the accuracy of recognition results in noisy environments, multimodal fusion recognition gradually replaces the traditional single-modal recognition methods. In this paper, we mainly strengthen and pre-process audio and video data first, and use LSTM recurrent neural network for deep feature extraction of audio and video streams, which effectively solves the problem of long-term forgetting of general neural networks. The audio and video feature vectors are then fused by a fully connected neural network with linear connections. Compared with the speech recognition technique alone, this audiovisual fusion recognition method has a better recognition effect in the case of noise interference. Compared with the traditional audiovisual recognition method, the model simplifies the recognition work. Recognition experiments on the LRS2-BBC dataset show that the recognition accuracy of this method improves to a certain extent over that of other methods in a clean environment and greatly improves in noisy conditions.

With the development of microelectronics technology and the improvement of its basic design, intelligent integration technology based on STM32 is extremely important in the research and development of new medical devices. In treating urinary system-related diseases, it is necessary to carry out intelligent dynamic monitoring, safety monitoring, and remote data management of primary physiological characteristic data such as bladder pressure, intra-abdominal pressure, and urinary catheter pressure. However, there is currently no dedicated real-time monitoring equipment in clinical practice. To solve this clinical need, we designed an intelligent bladder monitoring system based on STM32. The lower computer of the system can monitor the bladder physiological information of the patient in real-time and then transmit the detected data to the upper computer through the LoRa (Long Range Radio) so that doctors and nurses can deal with the abnormal situation of the patient in time, to realize the purpose of comprehensive and effective bladder monitoring. The final experimental results show that the system can effectively combine the conditions of individual patients and, at the same time, consider the bladder pressure and the volume of fluid in the bladder to formulate a personalized discharge plan, minimize the impact of the catheter, and benefit the bladder function training of patients and postoperative recovery.

In recent years, the requirements of fire prevention and theft prevention in various industries have been further improved. Therefore, how to study a new intelligent fire-proof anti-theft security system has become a research hotspot, and the advantages of optimizing and improving this kind of system based on image recognition technology are more prominent. Based on this, this paper discusses the design of intelligent fire-proof anti-theft security system based on image recognition, expounds the design of hardware and software, and discusses in detail the principle and realization process of image recognition function in the system for reference.

The previous researches on signal compensation technology based on feedforward neural network (FNN) are all in underwater channel and fiber channel. In this paper, the signal compensation technology based on FNN is applied to 8Gb/s 4-PAM indoor free space optical communication (FSOC) system for the first time. Under 7% forward error correction (FEC) threshold, the FNN algorithm is compared with direct detection and traditional LMS filtering algorithm. The FNN-based method can significantly improve the receiver sensitivity and improve the performance of the communication system.

This paper is devoted to the study of the EEP problem, and the main contributions are summarised as follows: A WSN-based hybrid virus-patch propagation model is constructed using node-level propagation modelling techniques to quantify the energy efficiency of patching strategies. Based on this, we reduce the EEP problem to an optimal control problem, called the EEP model, where each control represents a patching strategy and the target generalisation to be optimised represents the energy efficiency of a patching strategy. It is theoretically proven that there is optimal control in the EEP model, i.e. the EEP problem is solvable. We then present an optimality system for the EEP model, from which we obtain that the optimal patching strategy is a bang-bang type of control, indicating that the patching strategy is easy to implement. A numerical solution algorithm is given based on the optimality system, and several numerical examples show that an optimal patching strategy can be obtained by solving the optimality system. The influence of the model parameters on the optimal patching strategy is revealed through extensive computer experiments.

In the process of real-time generation of scene echo signals in the synthetic aperture radar real-time closed-loop echo simulator, a convolution is required every time an echo calculation parameter data packet is received. The number of points for convolution of the radar excitation signal and the target response function is relatively large, and the amount of calculation is large, which poses a challenge to the design of the system. The traditional convolution operation consumes a lot of resources and time when implemented in hardware, and cannot meet the requirements of the design. This design adopts a segmented parallel convolution algorithm to give full play to the advantages of FPGA parallel processing. Multiple pipelines are performed at the same time, which saves the time required for convolution operations. In order to solve the problem of the uncertain length of the radar excitation signal, this design proposes a time-sharing loop superposition algorithm, which successfully completes the convolution operation of the arbitrary-length radar excitation signal and the fixed-length target response function. The Xilinx FPGA development tool System Generator on the Simulink platform is used to build the model, and the circuit design is directly generated in the FPGA. Comparing the hardware test results with the theoretical results of Matlab simulation, it is found that they are basically the same. The parallel convolution algorithm proposed in this paper can accurately obtain the convolution result. At the same time, by evaluating the calculation time, the algorithm has good real-time performance and can meet the real-time requirements in radar echo simulation.

With the rapid deployment of 5G base stations across the country, there is a phenomenon that C-band (3.7 ~ 4.2ghz) radio and television satellite receiving stations are interfered by 5G signals in many places in China. How to formulate effective protective measures suitable for the whole country against 5G interference is a new challenge faced by the radio and television industry. This thesis studies the interference coupling principle of 5G signal to the satellite receiving front end, analyzes and demonstrates the interference factors, constructs the calculation model of the worst interference scenario in theory, calculates and analyzes the suppression indexes required for different geographical locations in the north, South, West and east of Gansu Province, and obtains that the C-band filter with the suppression degree not lower than the theoretical calculation value is used in the actual project, so as to provide the basis for the industry to effectively protect 5G interference.

Aiming at the problems that the extraction of lane lines at home and abroad is greatly affected by light, the extraction speed is slow, and the robustness is poor, an automatic extraction method based on lidar point cloud data is proposed. Due to the large amount of road point cloud data, this method adopts the strategy of divide and conquer. First, the point cloud is divided into blocks, and then analyzed block by block. The point cloud clustering filtering algorithm is used to perform preliminary processing on each piece of data, and then highly filter it to extract the ground point cloud; then based on the point cloud brightness gradient to extract the marker line; finally, the vectorization and extraction of road dashed lines and road shoulders was realized using improved Hough linear detection methods. In order to verify the robustness and feasibility of the algorithm, we selected the data of different road segment in the university campus, and then accurately evaluate the results. The experimental results show that this method takes 5.2s to process 9812062 points, the average lateral offset rate is 7.2%, and the overall detection accuracy is 96.1%, which is better than other methods. It can accurately extract the shoulder line and the dashed line of the road motor lane. It has certain practical application value in autonomous driving and unmanned driving based on lidar point cloud data.

In view of the impact of closed management on the tourism industry under the epidemic situation, this paper first puts forward differential flow-limiting schemes for scenic spots in different risk levels with the help of the cellular automata model and Mason rotation algorithm. It is carried on the scenic spot tourist flow simulation. Then based on the gray wolf multi-objective dynamic programming algorithm and NSGA-II multi-objective optimization model, the quantitative model is analyzed, and the BP (Back Propagation) convolution neural network algorithm is used to predict and test the multi-objective programming and optimization model. Finally, it is concluded that the lower the potential risk of the epidemic situation in the area, the smaller the limited flow of the scenic spot, and the greater the maximum and instantaneous carrying capacity of the scenic spot. In turn, it leads to the increase of the income of the scenic spot, and the improvement of the tourism experience of tourists is studied. Finally, it is combined with the above analysis. The paper puts forward a differentiated management scheme for the scenic spots at different risk levels.

Massive user interaction information provides more data choices for recommender systems. With the increase of user choices, the sparsity of user interaction data has become an important challenge. For traditional recommendation algorithms, how to extract effective features to solve the problem of data sparsity has always been a hot research direction in the field of recommendation systems. In view of the above-mentioned problems, this paper proposes a recommender based on enhanced collaborative filtering algorithm for 5G marketing system. Specifically, by broadening the input vector field and fusing the user's social information and personal interaction information, the problem of multisource information collaborative modeling is effectively solved. Further, in order to achieve the enhancement of vector structure, we introduce a social regularization term and an internal regularization term to alleviate the overfitting problem of most recommendation algorithms. Experiments on a large number of real-world datasets show that our proposed method effectively improves the performance of the model on sparse data.

In the current big data environment, aiming at the problems that traditional machine learning needs manual intervention and time-consuming to detect DDoS attacks, a DDoS attack detection method based on a double-stacked long short-term memory network is presented. The preprocessed data stream is sorted by recursive feature elimination algorithm, and the features with the most DDoS attack characteristics are selected as high-quality features, forming a double-stacked long short-term memory network data input format. The Center Loss is introduced into the Softmax Loss to reduce the intraclass distance, further improve the classification accuracy. Finally, the information containing DDoS attack characteristics can be quickly extracted from the complex characteristics of traffic. The CIC-IDS2017 dataset is used to train the model. Experimental research shows that the proposed model has an accuracy rate of 99.48% compared with other neural network models, and the detection effect is better than the compared algorithms, which can effectively achieve DDoS attack detection.

UAV technology has developed rapidly in recent years, Images extracted by UAV are widely used in urban division, crop classification, land monitoring etc. However, there are problems in UAV image segmentation such as image category imbalance, object scale variation, and insufficient utilization of contextual information, etc. To address the above problems, this paper uses optimized deeplabv3+ network model, and cross-entropy loss function for balancing the dataset samples in the experimental process for image semantic segmentation research. The results show that the algorithm of this paper has a high accuracy rate for semantic segmentation of UAV images, and can recognize each category of UAV images better, and the segmentation effect is better.

Arming at the problems of large-scale remote sensing image, complex surface background and sparse aggregation of aircraft objects in remote sensing detection, we propose an efficient remote sensing aircraft object detection algorithm based on Faster R-CNN. Firstly, we embed a lightweight aircraft object recognition network in Faster R-CNN network to filter invalid remote sensing image slices. Then, we fuse the feature map of lightweight aircraft object recognition network (LAORNet) into Faster R-CNN to enhance the semantics of aircraft objects. In addition, we add an rotation parameter to the Faster R-CNN bordering regression, so that the Faster R-CNN can predict the rotation angle of the aircraft in the image. The experimental results show that our algorithm has 86.94% AP50 and 34 fps on UCAS-AOD dataset, which has achieved competitive results.

Most networks mainly use firewall and other devices to isolate from external networks. However, with the application of new technologies such as cloud computing and Internet of things, the degree of interconnection between networks is deepening, and the difficulty of security protection is greatly improved. How to effectively detect network intrusion has become very important. Compared with traditional intrusion detection technology, convolutional neural network has better ability to extract intrusion features. This paper proposes a network intrusion detection method based on one-dimensional convolutional neural network and grey wolf optimization algorithm to optimize support vector machine. Firstly, one-dimensional convolutional neural network is used to extract high-level features from intrusion detection data, and then support vector machine is used to classify and detect the extracted high-level features, in which the parameters of support vector are optimized by grey wolf optimization algorithm. Through simulation experiments, the proposed method can effectively improve the detection accuracy and model balance.

PoW (Proof of Work) consensus algorithm has the advantages of simple algorithm, easy to use, allowing 50% of the nodes in the whole network to fail, but it still has some problems, such as low efficiency, mining waste a lot of resources. PoUW (Proof of Useful Work) based on PoW uses the wasted resources of PoW to carry out meaningful calculation, while PoLe (Proof of Learning) uses the wasted resources to train the machine Learning model. However, it has some problems, such as unreasonable reward mechanism, only one model training, failed to exert all the computing power of nodes. Therefore, the PoLe consensus algorithm is implemented and improved on the consortium blockchain in this paper. Based on the limited access characteristics of the consortium blockchain, some security is sacrificed in exchange for the increase of block speed.

In order to solve the problem of device heterogeneity in federated learning, FedAsync proposes asynchronous federated learning, that is, the server and the client interact in an asynchronous manner, that is, the server updates the global model immediately after receiving the local model. Communication between server and client is non-blocking. Therefore, the server and client can update the model at any time without synchronization, which is advantageous when the devices have heterogeneous conditions. However, when the gap between clients increases, the time difference of the uploaded model parameters becomes larger, which has a great impact on the accuracy. In response to the above problems, this paper first introduces the concept of training efficiency, evaluates the training ability of customers, and determines the training frequency of customers according to the training ability, thereby narrowing the gap between customers. Second, change the semi-asynchronous strategy, that is, the uploaded model is not updated at any time, but aggregated and updated after a certain amount or a certain time is satisfied. Finally, the effect is verified by simulation experiments.

Action recognition methods based on human skeleton can explicitly represent human actions, and have gradually become one of the important research directions in the field of computer vision. To address the problems that the skeleton graph in graph convolutional networks is fixed to represent only the physical structure of the human body and the lack of adaptive ability to the skeleton topology graph structure, this paper proposes a dual-stream non-local graph convolutional network based on the attention mechanism. First, the temporal convolution layer is extended to a parallel structure with multiple kernels, and different temporal convolution kernel modules are adaptively selected to collect features by channel weights; second, an attention model consisting of an attention pooling layer is proposed to capture the correlation and temporal continuity among joints; finally, the nonlocal graph convolution network is used as the basic framework with joint information, skeletal information and respective motion information A dual-stream fusion model is constructed. The proposed method is compared with the mainstream methods in recent years on the action recognition dataset NTU RGB+D, and the experimental results show that the proposed method achieves high accuracy in action recognition.

Compared with the classical object detection algorithms with horizontal bounding box, such as Yolo and Faster-R-CNN, the oriented object detector in remote sensing images can be more robust. Considering that the challenge of oriented object detection in remote sensing images, we propose a cross-layer feature fusion improvement strategy. Specifically, to obtain multiple layer feature maps for subsequent object detection, the shallow feature map with texture information merged with the deep feature map with semantic features. The spatial attention mechanism is introduced to enhance our algorithm's attention to the non-local information in these feature maps. Extensive experiments on a public dataset, DOTA, demonstrate the effectiveness of our proposed method. Under the large difference of object scale, arbitrary orientation of object, objects with dense arrangement and complex background, the experimental results show that our method has better performance.

Time-triggered traffic (TT) is the key traffic responsible for real-time data in time-sensitive network (TSN). Following the transmission rules of gated list, selecting different transmission paths will affect the schedulability and running time of TT traffic to different degrees. Therefore, in order to reduce the running time of TT traffic scheduling, a path selection strategy (PSS) based on collision detection was proposed, which aimed at reducing the number of the worst collision detection between TT traffic, and constructed a cost function to measure the collision between different paths, providing a basis for TT traffic path selection. A critical link heuristic scheduling algorithm (CTS) was proposed to improve the success rate of TT traffic scheduling by using the selected path of TT traffic and minimizing the end-to-end delay of TT traffic.

With the rapid development of wireless technology, indoor positioning using WiFi has become a hot spot in the current positioning field. At present, WiFi signals are ubiquitous, resulting in uneven AP deployment and redundancy, which affects positioning accuracy. In this paper, an AP selection algorithm based on information entropy and mutual information is proposed by analyzing the characteristics of AP signals to optimize the AP. Its positioning accuracy is significantly better than the other two AP selection algorithms.

The rapid identification of key nodes in large-scale social networks is important for the guidance and control of public opinion. The model proposes a key node identification method that considers user attributes, behavioural attributes, historical text attributes, user expertise and rationality values. Compared with methods that only consider user attributes, behavioral attributes and historical text attributes, this method can identify key nodes more quickly. Limitations: This method cannot identify key nodes in the network in real time.

In order to effectively solve the problem of classification and identification of human peripheral blood leukocytes, this paper proposes a classification and identification framework for leukocytes based on an improved YOLOv5 network. YOLOv5s is selected as the network model, and the samples are preprocessed by constructing a priori frame through k-means clustering. Secondly, the original feature map is obtained by the Focus slicing operation, and the original feature map is sent to the Neck network, and the information is transferred and fused through FPN and PAN to obtain the optimal weight. Finally, CIOU_Loss is selected as the loss function of frame regression to achieve high-precision positioning. The experimental result shows the average accuracy of white blood cell detection of the improved YOLOv5s algorithm is 94.8%, which is 10.7% higher than the previous one. It shows that this method has high detection accuracy and can effectively improve the accuracy of human peripheral blood leukocyte identification.

Video tracking uses the semantic information between video image sequences to process, analyze and study the target to achieve target tracking. In response to the problems of sharp changes in target position, large deformation, and occlusion due to similar background interference, changes in lighting conditions, and different shapes of targets in the target tracking process, the tracking algorithm is less accurate and less robust in target appearance. An improved target tracking algorithm based on multi-domain network (MDNet) is proposed to solve this problem. By adding the Image-Align layer to the video target tracking task, a more accurate target value is obtained; applying the Directed Acyclic Graph-Recurrent Neural Network (DAG-RNN) by combining it with a convolutional neural network and modeling image neighborhood context dependencies for the target region to be tracked, we improve the problem that conventional networks only perform multi-layer extraction for the appearance features of the target, thus resulting in poor robustness to changes in target appearance; ROI Align layer is added after the convolution layer to speed up target feature extraction.

Insulated Gate Bipolar Transistor (IGBT) is the most innovative power device, widely used in the field of wind power converters and aerospace control systems. Its reliability is directly related to the safe operation of the whole system. On the basis of studying the statistical and physical characteristics of the low-frequency noise of the IGBT, a low-frequency noise detection system is designed to realize the non-destructive detection of low-frequency noise of IGBT in this paper. Taking IGBT single-tube normal device and faulty device as detection objects, the low-frequency noise time series and spectrum of the two types of devices are detected and analyzed. The experimental results show that the detection system has good performance. It can accurately obtain the low-frequency noise data of IGBT devices and complete the function of spectrum analysis. Through the exploration and research of this topic, it will provide accurate and feasible methods and techniques for the detection of low-frequency noise of IGBT devices, and provide strong support for the reliability characterization and fault diagnosis of IGBT devices.

In order to improve the problem of inaccurate results in non-contact heart rate detection due to a series of movements of the subject such as breathing, blinking, facial expressions and noise generated by changes in ambient light, the signal is processed in advance using normalisation and wavelet denoising, and then an extreme gradient boosting (XGBoost) algorithm based on a Gaussian process (GP)-based Bayesian optimization method is introduced. The GP-XGBoost machine learning model was introduced to estimate the heart rate. The results show that the estimation error of heart rate by the GP-XGBoost model is significantly reduced compared to that obtained by the conventional method, promoting the practical application of contactless heart rate measurement.

With the development of big data, information collection and analysis play a decisive role in marketing decisions. The marketing decision system support factor is one of the most effective tools for mobile phone information and correct analysis of information. The use of these systems is widely, with the continuous development of business technology, marketing means and research, domestic research in this aspect, this paper discusses the mass of WeChat marketing data mining system design, introduces the marketing can only support system architecture and function, aims to improve the use of its process, speed up the progress of marketing data mining system design[1].

Aiming at the problems of insufficient therapists in traditional rehabilitation training for autistic children, difficult to persist in training for a long time and limited training venues, a rehabilitation training system for autistic children based on virtual reality technology was designed. The system uses the Unity3D engine combined with VR equipment to build a rehabilitation training scene, and controls the interaction between the VR equipment and the virtual scene through the C# programming language. Children with autism only need to wear 3D glasses and a tracking locator to perform rehabilitation training in the virtual scene. The results show that the training method in the system can effectively improve the autism degree of autistic children, and has a positive effect on their rehabilitation.

Knowledge atlas can be effectively applied to the construction of information intelligent analysis system, especially in intelligent semantic search. Through the knowledge map, the modeling needs for user needs and the understanding of natural semantics can be realized. With the emergence and development of artificial intelligence technology, knowledge driven has become the most eye-catching technical means in the era of big data. Through the above technical means, the establishment of judicial knowledge base and intelligent analysis system for case judgment can ensure the accuracy of system analysis results while ensuring the safety of the system.

Synthetic Aperture Radar (SAR) plays an important role in ocean monitoring tasks due to its strong adaptability to weather and time conditions. Implementing ship monitoring is of great significance to maintaining sea area safety and ship management. This paper proposes a ship target detection method based on target enhancement in SAR images. The algorithm is improved from three aspects: data enhancement, transformer visual model and spatial attention to meet the needs of ship detection in complex sea conditions. The experimental data set is the SSDD ship image data set. The performance of the model under the 30% and 100% data sets is evaluated respectively. The experimental verification is that the improved YOLO v5 model based on target enhancement in this paper achieves an average accuracy of 90.5%, which can meet the Requirements for short-term lightweight ship inspection tasks.

A fingerprint image classification method based on improved Res Net was proposed through experimental research. We take the existing fingerprint data set as object, expand the number of sample images by combining data enhancement technology, finally four kinds of labels are classified according to the global characteristics of fingerprints. The new net structure was built based on convolutional neural network, and iterative optimization was carried out by SGD. Classification accuracy and training rounds were evaluated by comparing the experimental results of different networks. To verify the applicability of the proposed method, 11824 sample images were composed of four kinds of data for training and testing. The results show that after 100 times of iterative training, the accuracy of fingerprint image classification for four kinds of data is above 95.7%, and the highest is 97.4%. This method supports high precision classification of mixed fingerprint images and has good practical.

The rapid development and application of the Internet in various fields has brought great convenience to both the society and individuals, and has gradually become an indispensable part of people's life and work. The development of computer technology has changed human life, and the risk and opportunity of virus invasion will also increase. With the increasing scale and automation of network attacks, the traditional detection methods have been unable to meet the needs of intrusion detection in the current network environment. This paper uses the data format and description conditions to define the data items, and completes the network intrusion data processing based on AL-SVM algorithm through feature extraction, so as to inhibit and protect the network intrusion.

With the rapid construction of 5G low-carbon power grids in China, compared with the traditional passive distribution network, the current loss of the distribution network including renewable energy has changed. The loss of the traditional energy grid is superimposed from the line loss to the transmission network loss. , which brings new problems to the planning and construction of the power grid. This paper proposes a layered and partitioned loss reduction method for the energy Internet. First, the loss of the communication network is included in the category of grid energy saving; The layer model reduces the loss of the distribution network, and uses the optimization algorithm to find the optimal solution of the model to obtain the optimal network configuration to reduce the loss of the network and provide certain ideas for the construction of 5G low-carbon power grids.

Optical early warning and detection is an important means of early warning, the photoelectric detector of which mainly uses infrared and visible light band. The targets imaging in space-based optical detector can be seen as IR small targets, thus IR (Infrared Radiation) small target detection and tracking has become an important field of optical warning and detection study. For the fluctuation of target energy, heterogeneity of imaging detector, and noise in signal processing, the target SNR (Signal Noise Ratio) in IR picture also fluctuates. The main task of this paper is to analyze the factors affecting the SNR in IR staring imaging and realize a method of multi-frame detection.

In the research of automatic classification of epilepsy EEG (Electroencephalogram), the detection model parameters are often set based on artificial experience, and the structure lacks adaptability. The epilepsy EEG signal is used as the research object. After preprocessing the EEG signal, Use CNN (convolutional neural network) for feature extraction, and use the PSO (Particle Swarm Optimization) algorithm to adaptively optimize the CNN model parameters to form a PSO-CNN epilepsy classification model. The algorithm proposed has an accuracy of 94.8% on the epilepsy dataset of the University of Bonn. Compared with traditional detection methods and other deep learning methods, the proposed algorithm achieves a higher accuracy.

As the most vital part of the external message acquired by human beings, visual message plays a crucial role in the process of human understanding and transformation of the world. However, myopia is affecting humans' ability to perceive the outside world. Prevention and treatment of myopia is a common concern of the society. In view of the current use of electronic products for a long time, close range or use of electronic products under the condition of poor light intensity, a vision protection means is proposed. As an vital part of the visually impaired people, people with low vision have the defect of difficulty in recognizing picture message, which affects their daily work and rest. The means used in this article is apparatus vision analysis. The traditional means of analyzing videos and other related processes through simple human labor is no longer applicable in today's era. Now we often use computers to automatically classify videos to assist people in making corresponding videos. analyze. Its application scenarios involve video surveillance, human-computer interaction, virtual reality, video retrieval, etc.

Recently, the high-payment transfer of Chinese A-shares in stock investment has been highly sought after by small and medium-sized investors and has gradually become a spotlight. Listed companies make ex-rights treatment of their stocks when making high-send and transfer decisions. Investors who buy at this stage can make profits through stock appreciation in a short period. Many companies will immediately increase the daily limit when trading at a high price. Therefore, predicting the decision and buying in advance is of significance to investors. This paper uses the income statement and "high delivery" data of the vaccine cold chain, battery, steel, and planting and forestry sectors from 2010 to 2021 ,using Random Forest and XGBoost to screen out six high contributions factors with significant influence. We also use mean values to fill indicators with a missing ratio below 30% and remove the mean value of high delivery data. Then we establish a prediction model based on different Neural Network models. Some networks show high performance, in which the CNN network with regularization methods, weight decay, and dropout is the best, reaching 93 % of accuracy.

At present, the population of patients with physical movement disorders is gradually expanding. In clinical practice, lower limb rehabilitation training machine is widely used in auxiliary patient rehabilitation training, and lower limb auxiliary rehabilitation control system is an important priority in auxiliary rehabilitation training for patients with limb movement disorders. The mathematical modeling based on Petri network is widely used in discrete event systems with synchronization, concurrency, conflict and other characteristics. First, this paper expounds the lower limb rehabilitation, four different types of Petri network models, and then uses the basic Petri network to build a model for this control strategy. Finally, according to the structural characteristics of the established Petri net model, the method of establishing a reachability tree is selected to analyze the accessibility of the model, which clearly and intuitively reflects all the states that the model can reach, and provides a theoretical basis for the on-demand auxiliary control of lower limb rehabilitation.

Because of the interaction of Douyin, it has become a new generation of social tools. There is a huge browse every day, which will leave various access data. Internet criminals take advantage of the characteristics of low cost, high fraud rate and strong concealment, and the interests of users are easily violated on social platforms. Therefore, it is of far-reaching significance to protect the interests of users by forensic analysis of social platforms. In this paper, we conduct forensic analysis of Douyin. The effective data displayed by Douyin lack other relevant information, such as various id information, hardware and software information, and dynamic details information. By building a proxy server to obtain the user information data of Douyin, the knowledge map of key data information is drawn, and the extracted information is transformed into the relationship map for visual display, which improves the efficiency of tracking analysis for analysts to fix relevant evidence.

With the rapid development of the personal credit market, it is accompanied by financial risks brought about by dishonesty. Among the existing personal credit prediction models, the traditional machine learning model is ineffective in dealing with multiple categories of users due to its poor generalization ability. In contrast, the neural network model requires many samples to cover multiple modal users. Its high training cost also makes this method limited. This paper proposes an adaptive iterative multimodal clustering method to predict the default of credit users. According to the multicategory characteristics of credit users, the method first divides users into multiple sets from coarse to fine-grained and then establishes decision tree models for prediction. This method not only solves the problem of the poor generalization ability of traditional machine learning models but also ensures that the overall training cost of the model is low. Comparative experiments are conducted based on representative credit datasets to verify the superiority and effectiveness of the proposed model. Experimental results show that the composite model outperforms other machine learning models and is suitable for personal credit loan prediction problems.

Electromagnetic coding metasurface has become a research hotspot with its modulation capability brought by coding sequence in recent years. In this paper, we achieve the reduction of monostatic Radar Cross section (RCS) by the phasemodulation silicon all-dielectric random coding metasurface. The silicon all-dielectric units are designed and used to compose the 2-bit coding metasurface. To research the influence on scattering by coding sequences, a super unit composed of M×M (M=1,2,3,5) unit cells is taken as the coding element of realizing 2-D coding sequences. The number of -10dB lobes is proposed to measure the communication capacity of metasurface and the scattering entropy is similar index. According to them, it is proved that the coding metasurface consisted by 2×2 super units with random sequence is the best simulation result among four kinds of super unit cells, which can achieve -10dB RCS reduction in a frequency from 3GHz to 5GHz. We envision that our research is expected to be the potential candidates in RCS reduction and information communication.

To address the problems of inadequate feature interaction and lack of targeting in feature combination in the clickthrough rate prediction model. We propose a click-through prediction model called SELFM. It based on attention mechanism and logarithmic transformation structure. The model first incorporates the attention mechanism in the feature embedding stage to distinguish the importance of different features and avoids the effects of invalid features. Then the field-aware factorization machine is used to learn low-order feature interactions. The logarithmic transformation structure is used to convert the power of each feature in the feature combination into the coefficients to be learned and combined with the hidden layer for higher-order nonlinear feature interactions. The final output layer is processed with the Sigmoid function to get the click-through rate prediction results. The experimental results show that the AUC and Logloss of this paper's model are better than the existing click-through rate prediction models, which effectively improves the prediction accuracy and enhances the ability of the recommendation system to process data.

In order to meet the requirements of high efficiency and security of data processing in wireless distributed transient pressure testing system, a new data compression and encryption acquisition method combining two-dimensional coupling cascade chaos map and semi-tensor compression sensing is proposed in this paper. Firstly, the cascade chaotic system based on sine function enhancement has better chaos complexity and larger chaos range than seed mapping. In this paper, a sinusoidal chaotic coupling cascade model is proposed. Secondly, the chaotic system is mapped to a compressed sensing observation matrix, and the dimension of the observation matrix in the sampling process is reduced by using semi-tensor theory. Finally, an experimental analysis of the 50psi measured shock wave combined compression encryption shows that the encrypted key space is 1064 with a compression ratio of 1∕4, the key sensitivity is 10-16, and the reconstruction error is less than 0.006 under the noise interference of 10dB, which meets the transmission requirements of wireless sensor networks.

Due to the small differences in driver distraction actions and the high similarity of some actions, in this paper we propose a distracted driver behavior recognition method (BACNN) based on convolutional neural network (CNN) using bilinear fusion network and combining attention mechanism with current mainstream algorithms of deep learning. In this paper, we use a driver dataset from State Farm for testing, and use 75% of this dataset for training and 25% for testing. The driver behavior pictures in the dataset are extracted using our specific convolutional neural network model for feature extraction and classified using a fully connected layer. Experiments demonstrate that this method has better recognition results compared to single-model network extracted features.

The medium voltage distribution network three-core buried armored cable can be regarded as a mutually coupled 5- conductor transmission line model in actual operation. For the problems of strong coupling between each conductor and difficult calculation of parameter matrix, a calculation method of 5-conductor three-core cable parameters is proposed. First, based on the loop analysis method, the loop impedance matrix of the three-core cable is derived, the calculation method of each loop impedance matrix is introduced, and the 7-conductor cable impedance parameter matrix is obtained through the loop transformation matrix; then, based on the principle of electrostatic induction, the 7-conductor cable conductance parameter matrix is obtained by inverting the potential coefficient matrix; on this basis, the 5-conductor cable parameter model is derived using the Gaussian elimination method; Finally, the to verify the correctness of the proposed method, the parameters are calculated using the mathematical model in this paper and the three-core cable model in PSCAD software, respectively. The simulation results show that the proposed method can accurately calculate the parameters of the 5-conductor three-core cable.

With the explosive growth of information, traditional data processing technologies have been unable to process information quickly and efficiently. As an emerging technology, big data technology can collect, store, and process massive business data information, achieve accurate data analysis, and improve the efficiency of information management. Based on this, this paper uses big data technology to design an information management system. In the hardware part of the system, this paper adopts IBM database server and configures network controller to realize non-polar connection of communication. In the software part, this paper designs a module for collecting and uploading data, an expansion window for information management, and a data integration module is written. Finally, this paper selects the traditional BP network information management system for comparative testing. Observe the performance of the two systems in terms of access speed, performance and reliability, and take the response time, access delay and playback delay as the evaluation indicators of these three aspects in turn. The comparison results show that the information management system designed in this paper has more advantages. The efficiency improvement in these three aspects can reach 65.4%, 57.5% and 49.2% respectively, which has a good application prospect.

Firstly, this paper introduces the background and significance of browser automatic forensics, and analyzes the current situation and development trend of browser forensics at home and abroad. Cutting into the automatic analysis of browser history, bag-of-words model and TF-IDF model are established as feature extraction, and naive Bayes, support vector machine and convolutional neural network are used as classification algorithm models. A simulation experiment is carried out in the browser history file data, and through the analysis and comparison of the results, a better combination result is obtained from multiple feature extraction and classification algorithms.

A signed dominating function of a graph G = (V, E) is a function f : V → {1,-1} satisfying the condition that for every v ∈ V with f(N[v]) ≥ 1.The weight of a signed dominating function on G is the sum f (V) = ∑_v∈V f (v) and the signed dominating number, γ_s (G) is the minimum weight of a signed dominating function, and satisfying the condition γ_s (G) = f(V) said that makes the signed dominating function f is the graph of a minimum signed dominating function. In this paper, the methods of mathematical induction and classification discussion are mainly used to deeply investigate, we obtained the signed dominating number of the generalized Sierpiński networks, where G is any special class of graphs like Path P_n、Cycle C_n、Star K_1,n、Complete graph K_n .

For 325 data samples collected from a catalytic cracking gasoline refining unit, first, a selection method of octane number modeling variables is proposed by using a combination of recursive elimination method, regression analysis, and correlation analysis. Second, the main features are evaluated by support vector regression (SVR) and ridge regression, respectively. Error comparison selects 26 main variables as modeling operation variables, Then, a learning device based on Boosting method was proposed to build an octane number (RON) loss prediction model, and finally, uses the model interpretability variance, the root means square error (RMSE) and the number of Bad Cases are three indicators to evaluate the learning ability and overall performance of the octane number (RON) loss prediction model. The sensitivity analysis of the model is carried out to verify the octane number loss. Reasonableness of the prediction model.

Accurate identification of parameters is critical to the epidemiological utility of the results obtained from the COVID-19 transmission model. In order to optimize the model parameters, we propose an adaptive Cauchy quantum particle swarm optimization (QPSO) algorithm. We introduce a piecewise Cauchy mutation operator and the mutation probability is adjusted adaptively according to the fitness to enhance the global search ability of QPSO. The experimental results show that the improved QPSO algorithm has higher accuracy than original QPSO and PSO algorithms.

With the rapid development of information technology, the integration of computer application technology and information management has become an inevitable development trend in the new era. The application of computer information management system can not only effectively improve the work efficiency and quality of all walks of life, but also have a strong influence on the further development and progress of society. Computer information management system combines advanced information management methods with efficient technology, which fully demonstrates the importance of computer information management system. This paper deeply analyzes and discusses the application, main defects and related cases of computer information management and application management.

With the arrival of THE 5G era and the further growth of computing volume, how to rapidly process a large number of computing tasks is a problem that needs to be solved in the cloud computing field. In order to make the cloud computing task scheduling process with better performance, in this paper, the traditional particle swarm optimization (pso) algorithm was improved and applied to the task scheduling in cloud computing, in view of the task completion time as the optimization goal, main of population is initialized by the method of reverse learning, and the algorithm of the inertia weight and learning factor for dynamic adjustment. Through setting contrast experiments, it is confirmed that the improved algorithm has better performance and reduces the task completion time in cloud computing task scheduling.

In order to improve the water solubility of perhydroxycalix[4]arenes, computer simulation prediction was used to explore the gradual substitution of methylene groups(-CH2-) by imino groups(-NH-) in perhydroxycalix[4]arenes. The structural characteristics and spectroscopic properties of the complexes obtained by forming new ligands combined with uranyl ions were studied. And the recognition ability and binding force of the uranyl ion of the new complex formed after being replaced. The chemical stability of each complex was explored, and it was finally found that after all “-CH2-” were replaced by imino groups “-NH-”, the formed complexes had obvious advantages in chemical stability and binding force to uranyl ions.

The use of X-Ray rays to detect BGA solder joint bubbles and accurately segment the bubble area has always been a hot topic of research. In this paper, a dynamic enhancement algorithm is used to preprocess the background interference image to reduce the interference of complex background on the bubble segmentation result. The bubble is segmented by the threshold segmentation algorithm, and the segmentation accuracy of the enhanced data graph is about 23.6% higher than that of the original graph, and the area of the mis-segmented area is reduced by about 18.2%, and the segmentation accuracy is increased by 11%. It can be proved that the algorithm has better adaptability and segmentation accuracy in the context of interference.

In the ultra-precise motion control system of the lithography workpiece table. Existing encoder acquisition systems support only one specific type of optical encoder, i.e., absolute or incremental encoder. In order to achieve the requirements of accurate positioning and high transmission rate. In this paper, we use FPGA to process the encoder signal by BiSS communication protocol module, frequency counter Counter module and data splicing module. Finally, it communicates with the host computer via PCIe. The combined encoder acquisition system has the high accuracy position positioning of absolute encoder and the high transmission rate of incremental encoder. Simulation results show that the data acquisition system is able to process the data quickly and accurately and display the displacement data correctly in the simulation. The design is suitable for the field of lithography where high resolution and fast data transmission are required.

Aiming at the problem that the image edge is blurred and the middle hidden layer features are lost in image semantic segmentation, the attention mechanism of joint training in channel domain and spatial domain is proposed, and the multiobjective joint training loss function and residual connection module are used to learn the semantic features in semantic segmentation, and the hidden layer features in the process of network training are added to the calculation of loss value. The experimental results show that the introduction of multi-channel attention mechanism and residual connection in semantic segmentation network is helpful to improve the effect of semantic segmentation. Pascal voc2012 data set was used in the experiment.

With the further construction and development of smart cities, the disadvantages of existing environmental information collection solutions in terms of cost, power consumption and signal coverage have become more and more obvious. In order to realize the basic functions of the environment information acquisition system of the Internet of Things (IoT), this paper develops the terminal node of the system based on embedded system hardware, and uses the method of combining Wireless Local Area Network (WLAN) technology and Low Power Wide Area (LPWA) technology to transmit data. IoT cloud platform adopts Ali Cloud City IoT platform, and makes Web interface for data visualization. The system has high reliability and stability, and can realize the basic function of long-term continuous environmental information collection. In addition, its cost-effective performance and strong scalability make it of certain promotion value, which can provide a new idea for the combination of environmental information collection and IoT.