Wireless intelligent multifunctional grounding monitoring method, device, equipment and storage medium

Abstract

The present application relates to the field of wireless monitoring technology, and discloses a wireless intelligent multifunctional grounding monitoring method, device, equipment and storage medium. The method includes: integrating multiple sensors on security inspection equipment to build a multimodal perception system; collecting the original data set of passing personnel and transmitting it wirelessly to obtain a layered encrypted data stream; performing distributed edge computing preprocessing to obtain multimodal feature data; performing multimodal fusion and feature encoding processing on the multimodal feature data to obtain a deep feature vector; inputting the deep feature vector into a multi-task deep neural network model to analyze the grounding status of personnel and obtain a comprehensive grounding assessment result; performing centralized multi-dimensional analysis and associated visualization processing on the comprehensive grounding assessment result through an intelligent monitoring management platform, and generating a security inspection equipment execution instruction sequence corresponding to the passing personnel. The present application improves the detection accuracy of metal security inspection equipment.

Description

Technical Field

[0001] The present application relates to the field of wireless monitoring technology, and in particular to a wireless intelligent multifunctional grounding monitoring method, device, equipment and storage medium. Background Art

[0002] With the widespread application of electronic equipment and the rapid development of industrial production, the problem of electrostatic discharge has become increasingly prominent, posing a serious threat to personnel safety and equipment stability. Traditional grounding monitoring methods often rely on a single sensor and simple threshold judgment, which is difficult to cope with complex and changing actual environments. In addition, existing grounding monitoring systems generally have problems such as incomplete data collection, limited analysis capabilities, and poor real-time performance, which cannot meet the needs of high-precision and multi-functional grounding monitoring.

[0003] In the field of security inspection, static electricity is particularly critical, which not only affects the normal operation of the equipment, but also may cause false alarms or missed inspections, reducing the efficiency and accuracy of security inspections. However, most of the current security inspection equipment fails to fully consider the importance of grounding monitoring and lacks comprehensive assessment and intelligent management of the static electricity status of personnel. Summary of the invention

[0004] The present application provides a wireless intelligent multifunctional grounding monitoring method, device, equipment and storage medium for improving the detection accuracy of metal security inspection equipment.

[0005] In a first aspect, the present application provides a wireless intelligent multifunctional grounding monitoring method, the wireless intelligent multifunctional grounding monitoring method comprising:

[0006] Integrate multiple sensors into security inspection equipment to build a multimodal perception system with electric field, pressure, infrared and image acquisition functions;

[0007] The original data set of the passing personnel is collected by the multimodal sensing system, and the original data set is wirelessly transmitted to obtain a layered encrypted data stream supporting multi-protocol adaptive switching;

[0008] Performing distributed edge computing preprocessing on the layered encrypted data stream to obtain multimodal feature data;

[0009] Performing multimodal fusion and feature encoding processing on the multimodal feature data to obtain a deep feature vector fused with an attention mechanism;

[0010] Inputting the deep feature vector into a preset multi-task deep neural network model to analyze the grounding status of personnel, and obtaining a comprehensive grounding assessment result including static electricity, posture characteristics and environmental factors;

[0011] The comprehensive grounding assessment results are subjected to centralized multi-dimensional analysis and associated visual processing through an intelligent monitoring and management platform, and a security inspection equipment execution instruction sequence corresponding to the passing personnel is generated.

[0012] In a second aspect, the present application provides a wireless intelligent multifunctional grounding monitoring device, the wireless intelligent multifunctional grounding monitoring device comprising:

[0013] Building modules for multi-sensor integration of security inspection equipment to build a multi-modal perception system with electric field, pressure, infrared and image acquisition functions;

[0014] A transmission module, used to collect an original data set of a person passing through the multimodal perception system, and wirelessly transmit the original data set to obtain a layered encrypted data stream supporting multi-protocol adaptive switching;

[0015] A processing module, used for performing distributed edge computing preprocessing on the layered encrypted data stream to obtain multimodal feature data;

[0016] An encoding module, used for performing multimodal fusion and feature encoding processing on the multimodal feature data to obtain a deep feature vector fused with an attention mechanism;

[0017] An analysis module, used for inputting the deep feature vector into a preset multi-task deep neural network model to analyze the grounding status of personnel, and obtaining a comprehensive grounding assessment result including static electricity, posture characteristics and environmental factors;

[0018] A generation module is used to perform centralized multi-dimensional analysis and associated visualization processing on the comprehensive grounding assessment results through an intelligent monitoring management platform, and generate a security inspection equipment execution instruction sequence corresponding to the passing personnel.

[0019] The third aspect of the present application provides a wireless intelligent multifunctional grounding monitoring device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the wireless intelligent multifunctional grounding monitoring device executes the above-mentioned wireless intelligent multifunctional grounding monitoring method.

[0020] The fourth aspect of the present application provides a computer-readable storage medium, in which instructions are stored. When the computer-readable storage medium is run on a computer, the computer executes the above-mentioned wireless intelligent multifunctional grounding monitoring method.

[0021] In the technical solution provided by this application, by integrating multiple sensors such as electric field, pressure, infrared and image acquisition, all-round and multi-angle monitoring of personnel grounding status is realized, which greatly improves the comprehensiveness and accuracy of data collection. Distributed edge computing technology is used to pre-process the encrypted data stream, which significantly reduces the amount of data transmission and the burden of the central server, and improves the real-time response capability of the system. Multimodal fusion and attention mechanism are used to realize the deep feature extraction of personnel grounding status in complex environments, and enhance the system's ability to identify minor grounding anomalies. Through the multi-task deep neural network model, static electricity, posture characteristics and environmental factors are analyzed simultaneously, which improves the comprehensiveness and accuracy of grounding status assessment. The introduction of non-dominated genetic algorithm to dynamically optimize the model enables the system to continuously adapt to new grounding monitoring scenarios and requirements, and improves the scalability and long-term effectiveness of the system. Through the intelligent monitoring management platform, multi-dimensional analysis and correlation visualization are provided to provide intuitive and effective decision support for managers, and improve the efficiency of grounding risk management. Based on the comprehensive evaluation results, the execution instruction sequence of security inspection equipment is automatically generated, which realizes the intelligent linkage between grounding monitoring and security inspection equipment, and improves the detection accuracy of metal security inspection equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0023] Figure 1 A schematic diagram of an embodiment of a wireless intelligent multifunctional grounding monitoring method in an embodiment of the present application;

[0024] Figure 2 This is a schematic diagram of an embodiment of a wireless intelligent multifunctional grounding monitoring device in an embodiment of the present application. DETAILED DESCRIPTION

[0025] Embodiments of the present application provide a wireless intelligent multifunctional grounding monitoring method, device, equipment and storage medium. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , an embodiment of the wireless intelligent multifunctional grounding monitoring method in the embodiment of the present application includes:

[0027] Step S101: integrate multiple sensors on security inspection equipment to build a multi-modal perception system with electric field, pressure, infrared and image acquisition functions;

[0028] It is understandable that the execution subject of the present application may be a wireless intelligent multifunctional grounding monitoring device, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0029] Specifically, a multi-sensor integration platform is designed. On this platform, electric field sensors, pressure sensors, infrared sensors and image sensors are arranged reasonably to ensure that they can work together. Electric field sensors are used to detect static electricity and electric field changes, which helps to identify the static electricity state of personnel and possible charge accumulation problems; pressure sensors are responsible for monitoring the force applied by personnel on security inspection equipment and analyzing the posture and behavior characteristics of personnel through pressure data; infrared sensors can capture heat distribution and infrared radiation, which helps to accurately detect human thermal imaging information under various lighting conditions; image sensors can provide high-resolution visual data to record the appearance and movement details of passing personnel. At the hardware level, the multi-sensor integration platform needs to ensure that the layout of each sensor is reasonable and does not interfere with each other. Through efficient sensor interface and data bus design, the synchronous acquisition and transmission of multiple sensor data can be realized. At the software level, the multi-sensor data fusion algorithm is used to process the data streams from different sensors in real time, synchronize and pre-process the electric field, pressure, infrared and image data, eliminate noise and interference, and extract useful feature information. Through the fusion algorithm, sensor data of different modes can be integrated into a unified multi-modal perception data stream.

[0030] Step S102: collecting an original data set of a person passing by through a multimodal perception system, and wirelessly transmitting the original data set to obtain a layered encrypted data stream supporting multi-protocol adaptive switching;

[0031] Specifically, the electric field strength data of the passing personnel are collected by the electric field sensor in the multimodal sensing system, and the electric field strength data reflects the change and distribution of the static electricity of the personnel. The electric field strength data is subjected to fast Fourier transform to obtain the spectrum characteristics of human static electricity, which reflects the distribution of static electricity at different frequencies. The electric field strength data is adaptively filtered according to the spectrum characteristics of human static electricity to remove the noise components and obtain the static charge data after noise reduction. At the same time, the original pressure data of the passing personnel are collected by the pressure sensor in the multimodal sensing system. The pressure data reflects the force applied by the personnel when passing through the security inspection equipment, which is used to analyze the gait and posture of the personnel. The original pressure data is subjected to wavelet decomposition to obtain a multi-scale pressure feature matrix, which reflects the change of pressure at different scales. Based on the multi-scale pressure feature matrix, the original pressure data is analyzed to extract the gait feature data of the personnel, which is used to identify and judge the behavior and posture of the personnel, and helps to improve the safety and accuracy of security inspection. The temperature distribution data of the passing personnel is collected by the infrared sensor in the multimodal sensing system to provide important information about the heat distribution of the human body. Infrared data can reflect the temperature changes of different parts of the human body. Gaussian mixture clustering is performed on the temperature distribution data to obtain the coordinate set of the hot spot area of ​​the human body. These coordinate sets can indicate the area with higher temperature of the human body, which is helpful to identify the key parts of the human body. The temperature distribution data is segmented by regional growth according to the coordinate set of the hot spot area of ​​the human body, and the contour feature data of the human body is extracted. The original visible light image is obtained by the image acquisition device in the multimodal perception system to obtain important information about the appearance and clothing of the person, which is used to identify and analyze the appearance characteristics of the person. The original visible light image is subjected to multi-scale convolution operation to obtain a deep image feature map, which reflects the details and structure of the image at different scales. Based on the deep image feature map, the original visible light image is semantically segmented to extract the classification data of the material of the person's clothing. The timestamp alignment and data format unification of the denoised electrostatic charge data, the gait feature data of the person, the human contour feature data, and the classification data of the material of the person's clothing are performed to form a multimodal synchronous data packet. The multimodal synchronous data packet is dynamically encrypted and layered encapsulated to obtain a layered encrypted data stream that supports multi-protocol adaptive switching, ensuring the security of the data during transmission, and can adaptively switch the communication protocol according to different network environments to maintain the continuity and reliability of data transmission.

[0032] Step S103: performing distributed edge computing preprocessing on the layered encrypted data stream to obtain multimodal feature data;

[0033] Specifically, the layered encrypted data stream is decrypted and the protocol is parsed. The decryption process can use the preset key and decryption algorithm to ensure that the data maintains integrity and security during the decryption process. The protocol parsing restores the original multimodal data stream to its original state by identifying the protocol header information in the data stream. According to the timestamp information in the original multimodal data stream, the time synchronization of each modal data is performed to form an aligned multimodal data matrix. Outlier detection and interpolation are performed on the aligned multimodal data matrix. Outlier detection identifies abnormal points in the data through statistical and algorithmic models, and interpolation processing is performed to complete the missing parts of the data to obtain a cleaned multimodal data set. Based on the cleaned data set, each modal data is normalized and standardized. By performing a unified scale transformation on the data, the dimensional differences between different modal data are eliminated to obtain a standardized feature matrix. Feature dimensionality reduction is performed on the standardized feature matrix to reduce the dimension of the data while retaining the main features of the data. Feature dimensionality reduction can use dimensionality reduction algorithms such as principal component analysis to map high-dimensional data to a low-dimensional space to obtain a reduced-dimensional feature vector set. This process can reduce data redundancy, reduce computational complexity, and retain the main information of the data. The data of each modality is analyzed according to the feature vector set after dimensionality reduction to obtain separated independent feature components, which can reflect the key features of different modal data. The separated independent feature components are subjected to wavelet transform to obtain multi-scale time-frequency feature maps. Wavelet transform is an effective time-frequency analysis method that can decompose data into different scales and extract the features of data at different frequencies and times. According to the multi-scale time-frequency feature map, adaptive feature extraction is performed on each modal data to extract the feature descriptor of each modality. The adaptive feature extraction process can dynamically adjust the feature extraction strategy according to the characteristics of the data to obtain more accurate and effective feature descriptors. Nonlinear manifold learning is performed on the feature descriptors of the modality to map the high-dimensional feature descriptors to a low-dimensional embedded feature space. Nonlinear manifold learning can reveal the intrinsic structure of the data so that the data still maintains its main features and relationships in the low-dimensional space. Sparse encoding is performed on the features of each modality according to the low-dimensional embedded feature space to extract the sparse feature vector of each modality. Sparse coding can improve the interpretability and distinguishability of features, making different modal features clearer and more effective in subsequent analysis and processing, and ultimately obtaining multimodal feature data.

[0034] Step S104, performing multimodal fusion and feature encoding processing on the multimodal feature data to obtain a deep feature vector fused with an attention mechanism;

[0035] Specifically, the multimodal feature data is normalized to eliminate the scale differences between different features and obtain a standardized feature matrix. The mutual information between modes is calculated based on the standardized feature matrix to obtain the modal correlation measure. The mutual information between modes can reveal the dependency relationship and information sharing degree between different modal features. The modal correlation measure is threshold screened to select those modal pairs with high correlation to form a set of high-correlation modal pairs. By screening the high-correlation modal pairs, the feature combinations that have a greater impact on the system performance are focused on. Based on the set of high-correlation modal pairs, the joint probability distribution of features is estimated to obtain the multimodal joint feature representation. The multimodal joint feature representation is analyzed to obtain the feature space after dimensionality reduction. Dimensionality reduction can simplify the data structure, reduce redundant information, and retain the main features of the data. In the feature space after dimensionality reduction, the feature importance weight is calculated to obtain the weighted feature vector. The feature importance weight reflects the contribution of each feature in the overall data. Through weighted processing, the influence of important features is enhanced and the interference of unimportant features is suppressed. The weighted feature vector is nonlinearly transformed to obtain a high-order feature representation. Nonlinear transformation can capture the complex relationship and high-order interaction information between features. Based on the high-order feature representation, a feature similarity graph is constructed to obtain the feature topology structure, revealing the mutual relationship and structural information between features. Spectral clustering is performed on the feature topology structure to obtain the feature subspace division. Spectral clustering can divide the features into different subspaces according to the similarity between features, thereby simplifying the feature structure and highlighting the differences between features. According to the feature subspace division, the high-order feature representation is weighted and aggregated to obtain the deep feature vector fused with the attention mechanism. The attention mechanism can dynamically adjust the weight of the feature, highlight the key features, suppress the secondary features, and improve the accuracy and effectiveness of the feature representation.

[0036] Step S105, inputting the deep feature vector into a preset multi-task deep neural network model to analyze the grounding status of personnel, and obtaining a comprehensive grounding assessment result including static electricity, posture characteristics and environmental factors;

[0037] Specifically, the deep feature vector is input into a preset multi-task deep neural network model, which performs feature mapping processing on the deep feature vector through the input layer to obtain a target feature mapping vector. The target feature mapping vector is calculated by a multi-head self-attention mechanism to obtain a context-enhanced feature representation. The multi-head self-attention mechanism can capture the complex relationship and dependency between features by simultaneously focusing on different parts of the input features, and enhance the context information of the feature representation. The context-enhanced feature representation is subjected to a task-shared convolution operation to obtain a multi-scale feature map. Through the convolution operation, feature information at different scales is extracted. According to the multi-scale feature map, a feature pyramid network preset for the task is constructed to obtain a preliminary feature representation of static electricity, posture features and environmental factors. The feature pyramid network can effectively fuse and hierarchically process multi-scale features to provide a preliminary feature representation for each task. The preliminary feature representation is subjected to cross-attention processing to obtain a fusion feature of information interaction between tasks. The cross-attention processing can interact and share information between different tasks, so that the feature representation of each task can integrate the information of other tasks and improve the overall feature representation capability. Based on the fusion feature of information interaction between tasks, a fully connected layer preset for the task is constructed to obtain multiple independent prediction results. The multiple independent prediction results are respectively the static electricity quantity prediction results, the posture feature prediction results and the environmental factor prediction results. Through the calculation of the fully connected layer, the fusion features are converted into specific task prediction values, providing independent prediction results for each task. Soft parameter sharing analysis is performed on multiple independent prediction results to obtain corrected prediction values. Soft parameter sharing analysis can improve the accuracy and consistency of predictions by sharing parameters between multiple tasks. According to the corrected prediction values, the importance weights of multiple independent prediction results are constructed. The importance weights can reflect the contribution of each prediction result in the overall prediction, and through weighted processing, the information of different prediction results is integrated. The importance weights and the corrected prediction values ​​are weighted and summed, and different prediction results are effectively integrated to obtain a comprehensive prediction vector. The comprehensive prediction vector is Softmax normalized to obtain a comprehensive grounding assessment result including static electricity quantity, posture features and environmental factors. Softmax normalization can convert the comprehensive prediction vector into a probability distribution.

[0038] Step S106: The comprehensive grounding assessment results are centrally analyzed in multiple dimensions and correlated and visualized through the intelligent monitoring management platform, and a security inspection equipment execution instruction sequence corresponding to the passing personnel is generated.

[0039] Specifically, a multi-dimensional cluster analysis is performed on the comprehensive grounding assessment results, and the grounding status is divided into different categories to identify the characteristics and patterns of various grounding statuses. According to the classification of grounding status categories, a time series analysis of historical data is performed, and the probability distribution of future grounding status is predicted using past data trends and patterns, revealing the law of grounding status changes over time. Multi-level grounding risk assessment information is generated based on the probability distribution of future grounding status, and grounding risks at different levels are analyzed and evaluated to form a detailed risk assessment report. Association rule mining is performed on multi-level grounding risk assessment information to mine potential risk factor association networks and reveal the mutual relationship and influence between different risk factors. A force-directed graph layout is generated based on the potential risk factor association network, and the association and network structure between risk factors are intuitively displayed through the force-directed graph layout. The force-directed graph layout is processed by heat map superposition to obtain a comprehensive risk intensity distribution map, showing the risk intensity of different areas. Threshold segmentation is performed based on the comprehensive risk intensity distribution map to identify the specific location of high-risk areas, and the areas that need to be focused on and managed are found through the positioning results. Search optimization is performed on the positioning results of high-risk areas to obtain the best security inspection equipment control strategy. These control strategies can help managers develop specific action plans and response measures to improve the management and control efficiency of security equipment. According to the security equipment control strategy, the corresponding execution instruction sequence of the security equipment is generated. These instruction sequences will be sent to specific security equipment to guide the equipment to perform corresponding operations and adjustments, thereby achieving effective management and control of the grounding status.

[0040] Through parsing, the security inspection equipment execution instruction sequence is analyzed to extract the specific target execution action set. According to the target execution action set, the security inspection equipment is instructed to execute, and the original execution result data of the equipment is recorded in real time, including all relevant information generated during the equipment execution process. Multi-dimensional feature extraction is performed on the original execution result data to extract the feature vector reflecting the execution effect. Through feature extraction, the key features and patterns in the data are captured. According to the execution effect feature vector, a multi-objective optimization function is constructed to integrate the needs of different objectives to form a target set of the optimization model. On this basis, weight allocation processing is performed to assign appropriate weights to each objective to form a weighted optimization objective function. According to the weighted optimization objective function, the multi-task deep neural network model is initialized to generate a candidate model parameter set. The initial population generation is the starting point of the optimization process. A set of possible model parameter combinations is generated through random or heuristic methods. The candidate model parameter set is non-dominated sorted to obtain the Pareto optimal solution set. Non-dominated sorting compares the advantages and disadvantages of different parameter combinations to screen out a set of solutions that are not dominated by other combinations, namely the Pareto optimal solution set. The selection, crossover and mutation operations of the genetic algorithm are performed according to the Pareto optimal solution set to form a new generation of population. The selection operation ensures the transmission of excellent characteristics, the crossover operation generates new solutions by combining the characteristics of different solutions, and the mutation operation introduces random changes to improve the diversity of the population. Through these operations, the model parameters are continuously optimized and gradually approach the optimal solution. The model reconstruction and performance evaluation are performed on the new generation of population to obtain the optimized model candidate set. Model reconstruction rebuilds the model according to the new parameters, and performance evaluation measures the performance of the model in practical applications, thereby screening out models with excellent performance. Cross-validation is performed based on the model candidate set and the original execution result data to ensure the reliability and generalization ability of the model. Cross-validation divides the data into training sets and validation sets, trains and validates the model multiple times, evaluates its performance on different data sets, and finally obtains the metal security inspection verification model corresponding to the security inspection equipment.

[0041] In the embodiment of the present application, by integrating multiple sensors such as electric field, pressure, infrared and image acquisition, all-round and multi-angle monitoring of personnel grounding status is realized, which greatly improves the comprehensiveness and accuracy of data collection. Distributed edge computing technology is used to pre-process the encrypted data stream, which significantly reduces the amount of data transmission and the burden of the central server, and improves the real-time response capability of the system. Multimodal fusion and attention mechanism are used to realize the deep feature extraction of personnel grounding status in complex environments, and enhance the system's ability to identify minor grounding anomalies. Through the multi-task deep neural network model, static electricity, posture characteristics and environmental factors are analyzed simultaneously, which improves the comprehensiveness and accuracy of grounding status assessment. The introduction of non-dominated genetic algorithm to dynamically optimize the model enables the system to continuously adapt to new grounding monitoring scenarios and requirements, and improves the scalability and long-term effectiveness of the system. Through the intelligent monitoring management platform, multi-dimensional analysis and correlation visualization are provided to provide intuitive and effective decision support for managers, and the efficiency of grounding risk management is improved. Based on the comprehensive evaluation results, the execution instruction sequence of security inspection equipment is automatically generated, which realizes the intelligent linkage between grounding monitoring and security inspection equipment, and improves the detection accuracy of metal security inspection equipment.

[0042] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0043] (1) The electric field sensor in the multimodal sensing system collects the electric field strength data of the passing personnel, and performs fast Fourier transform on the electric field strength data to obtain the human body static electricity spectrum characteristics;

[0044] (2) Adaptively filter the electric field strength data according to the human body static electricity spectrum characteristics to obtain the static electricity charge data after noise reduction;

[0045] (3) The original pressure data of the passing personnel is collected through the pressure sensor in the multimodal sensing system, and the original pressure data is decomposed by wavelet to obtain a multi-scale pressure feature matrix;

[0046] (4) Analyze the original pressure data according to the multi-scale pressure feature matrix to obtain the gait feature data of the person;

[0047] (5) The infrared sensor in the multimodal perception system is used to collect the temperature distribution data of the passing personnel, and Gaussian mixture clustering is performed on the temperature distribution data to obtain the coordinate set of the hot spot area of ​​the human body;

[0048] (6) Performing region growing segmentation on the temperature distribution data according to the coordinate set of the hot spot area of ​​the human body to obtain the human body contour feature data;

[0049] (7) Obtaining the original visible light image through the image acquisition device in the multimodal perception system, and performing a multi-scale convolution operation on the original visible light image to obtain a deep image feature map;

[0050] (8) Perform semantic segmentation on the original visible light image based on the deep image feature map to obtain the material classification data of the person’s clothing;

[0051] (9) Perform timestamp alignment and data format unification on the denoised electrostatic charge data, personnel gait feature data, human body contour feature data, and personnel clothing material classification data to obtain a multi-modal synchronous data packet;

[0052] (10) Dynamically encrypt and layered encapsulate multi-modal synchronous data packets to obtain a layered encrypted data stream that supports multi-protocol adaptive switching.

[0053] Specifically, the electric field sensor in the multimodal sensing system collects the electric field strength data of people passing by, and these data reflect the changes in the electric field around the people. The electric field strength data is fast Fourier transformed, and the formula is as follows:

[0054] ;

[0055] in, represents the frequency domain signal, represents the time domain signal, is the length of the signal, is the frequency index, is an imaginary unit. Through fast Fourier transform, the time domain signal is converted into a frequency domain signal to obtain the spectrum characteristics of human static electricity. These spectrum characteristics can reveal the distribution of static electricity at different frequencies. According to the obtained static spectrum characteristics, the electric field strength data is adaptively filtered, and a filtering algorithm is used to remove noise to obtain the static charge data after noise reduction. Assume that the transfer function of the filter is , the spectrum of the input signal is , then the spectrum of the output signal is:

[0056] ;

[0057] in, is the signal spectrum after filtering. Through adaptive filtering, the filter parameters are dynamically adjusted, and the noise is removed according to the spectral characteristics of the input signal to obtain more accurate electrostatic charge data. At the same time, the pressure sensor in the multimodal sensing system collects the original pressure data of the passing personnel. These data reflect the force exerted by the personnel on the security inspection equipment. The original pressure data is decomposed by wavelet to obtain a multi-scale pressure feature matrix. Wavelet decomposition can extract the details and rough information of the signal by decomposing the signal into different scales. Assume that the signal The continuous wavelet transform of is:

[0058] ;

[0059] in, are the wavelet coefficients, is the scale parameter, is the displacement parameter, is the mother wavelet. Through wavelet transform, a multi-scale pressure feature matrix is ​​obtained to analyze the gait characteristics of the personnel. The infrared sensor in the multimodal perception system collects the temperature distribution data of the personnel, which reflects the temperature of each part of the personnel's body. The Gaussian mixture clustering method is used to obtain the coordinate set of the hot spot area of ​​the human body. The Gaussian mixture model assumes that the data points are mixed by multiple Gaussian distributions. The parameters of each Gaussian distribution and the distribution of the data points can be obtained by the expectation maximization algorithm. According to these parameters, the area with higher human body temperature is determined, that is, the coordinate set of the hot spot area of ​​the human body. According to the coordinate set of the hot spot area of ​​the human body, the temperature distribution data is segmented by regional growth to obtain the human body contour feature data. The regional growth segmentation starts from the hot spot area and gradually expands to the adjacent area until a certain similarity criterion is met, thereby segmenting the complete human body contour. At the same time, the original visible light image is obtained by the image acquisition device in the multimodal perception system, and the original visible light image is subjected to multi-scale convolution operation to obtain the deep image feature map. The convolution operation processes the image through convolution kernels of different scales to extract the details and features of the image. Assume that the input image is , the convolution kernel is , then the output of the convolution operation is:

[0060] ;

[0061] in, and is the size of the convolution kernel. Through multi-scale convolution operations, a deep image feature map is obtained. According to the deep image feature map, the original visible light image is semantically segmented to obtain the material classification data of the person's clothing. Semantic segmentation identifies the different materials and categories of the person's clothing by segmenting the image into regions with semantic meaning. The denoised electrostatic charge data, the person's gait feature data, the human body contour feature data, and the person's clothing material classification data are timestamped and unified in data format to form a multimodal synchronous data packet.

[0062] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0063] (1) Decrypt and parse the layered encrypted data stream to obtain the original multimodal data stream, and synchronize the time of each modal data according to the timestamp information in the original multimodal data stream to obtain an aligned multimodal data matrix;

[0064] (2) Perform outlier detection and interpolation on the aligned multimodal data matrix to obtain a cleaned multimodal data set, and normalize and standardize each modality data based on the cleaned multimodal data set to obtain a standardized feature matrix;

[0065] (3) Performing feature dimensionality reduction on the standardized feature matrix to obtain a set of feature vectors after dimensionality reduction, and analyzing each modal data based on the set of feature vectors after dimensionality reduction to obtain separated independent feature components;

[0066] (4) Performing wavelet transform on the separated independent feature components to obtain a multi-scale time-frequency feature map, and adaptively extracting features from each modal data based on the multi-scale time-frequency feature map to obtain the feature descriptor of the modality;

[0067] (5) Perform nonlinear manifold learning on the feature descriptors of the modalities to obtain a low-dimensional embedded feature space, and then sparsely encode the features of each modality based on the low-dimensional embedded feature space to obtain multimodal feature data.

[0068] Specifically, the layered encrypted data stream is decrypted and the protocol is parsed. The decryption process can use the preset key and decryption algorithm to ensure that the data maintains integrity and security during the decryption process. Protocol parsing restores the original multimodal data stream to its original state by identifying the protocol header information in the data stream, and obtains the original multimodal data stream. According to the timestamp information in the original multimodal data stream, the time synchronization of each modal data is performed to form an aligned multimodal data matrix. Outlier detection and interpolation processing are performed on the aligned multimodal data matrix. Outlier detection identifies abnormal points in the data through statistical and algorithmic models. Interpolation processing is used to complete the missing parts of the data to ensure the integrity and continuity of the data, thereby obtaining a cleaned multimodal data set. Based on the cleaned multimodal data set, each modal data is normalized and standardized. Normalization and standardization eliminate the dimensional differences between different modal data by performing a unified scale transformation on the data to form a standardized feature matrix. Assume that the original value of a modal data is , the normalized value is , then the normalization formula is:

[0069] ;

[0070] in, is the mean of the data, is the standard deviation of the data. Through this formula, all data are standardized to the same scale. Feature dimensionality reduction is performed on the standardized feature matrix to reduce the dimension of the data while retaining the main features of the data. Feature dimensionality reduction can use dimensionality reduction algorithms such as principal component analysis to map high-dimensional data to a low-dimensional space and obtain a set of reduced feature vectors. The principal component analysis algorithm finds the main directions of the data and projects the data onto these directions. The formula is:

[0071] ;

[0072] in, is the set of feature vectors after dimensionality reduction, is the standardized feature matrix, is the eigenvector matrix. According to the eigenvector set after dimension reduction, each modal data is analyzed to obtain separated independent characteristic components, which reflect the key characteristics of different modal data. The separated independent characteristic components are subjected to wavelet transform to obtain multi-scale time-frequency characteristic graphs. Wavelet transform is an effective time-frequency analysis method that can decompose data into different scales and extract the characteristics of data at different frequencies and times. Assume that the signal The discrete wavelet transform of is:

[0073] ;

[0074] in, are the wavelet coefficients, is the scale parameter, is the displacement parameter, is the mother wavelet. Through wavelet transform, multi-scale time-frequency feature maps are obtained. These feature maps can reflect the changes of data at different scales. According to the multi-scale time-frequency feature maps, adaptive feature extraction is performed on each modal data to extract the feature descriptors of each modality. The adaptive feature extraction process can dynamically adjust the feature extraction strategy according to the characteristics of the data to obtain more accurate and effective feature descriptors. Nonlinear manifold learning is performed on the feature descriptors of the modality to map the high-dimensional feature descriptors to a low-dimensional embedded feature space. Nonlinear manifold learning can reveal the intrinsic structure of the data so that the data still maintains its main features and relationships in the low-dimensional space. In this process, methods such as local linear embedding are used to construct a low-dimensional embedded feature space by maintaining the linear relationship within the local neighborhood. The formula is as follows:

[0075] ;

[0076] in, is a low-dimensional embedding feature space, is the mapping matrix, is a high-dimensional feature descriptor matrix. After obtaining the low-dimensional embedded feature space, each modal feature is sparsely encoded to extract the sparse feature vector of each modality. Sparse coding improves the interpretability and distinguishability of features by limiting the number of non-zero elements in each feature vector, making different modal features clearer and more effective in subsequent analysis and processing. Finally, all sparse feature vectors are integrated to obtain multimodal feature data.

[0077] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0078] (1) Normalize the multimodal feature data to obtain a standardized feature matrix, and calculate the mutual information between modes based on the standardized feature matrix to obtain the modal correlation measure;

[0079] (2) Threshold screening is performed on the modal correlation measure to obtain a set of highly correlated modal pairs, and the joint probability distribution of features is estimated based on the set of highly correlated modal pairs to obtain a multimodal joint feature representation;

[0080] (3) Analyze the multimodal joint feature representation to obtain the feature space after dimensionality reduction, and calculate the feature importance weight based on the feature space after dimensionality reduction to obtain the weighted feature vector;

[0081] (4) Perform nonlinear transformation on the weighted feature vector to obtain a high-order feature representation, and construct a feature similarity graph based on the high-order feature representation to obtain the feature topological structure;

[0082] (5) Perform spectral clustering on the feature topology to obtain the feature subspace division, and perform attention weighting and aggregation on the high-order feature representation according to the feature subspace division to obtain the deep feature vector of the fused attention mechanism.

[0083] Specifically, the multimodal feature data is normalized. By standardizing the data to make them have the same scale, the dimension differences between different features are eliminated to form a standardized feature matrix. The mutual information between modes is calculated based on the standardized feature matrix. The mutual information is used to measure the dependence between two variables. Assume that there are two modal feature variables and , their joint probability distribution is , the marginal probability distributions are and The mutual information can be calculated by the following formula:

[0084] ;

[0085] in, Representation variables and The mutual information between them is calculated to obtain the correlation measure between the modes. The modal correlation measure is threshold screened to select those modal pairs with high correlation to form a set of high-correlation modal pairs. Through threshold screening, modal pairs with low correlation are filtered out to focus on important feature combinations. After obtaining a set of high-correlation modal pairs, the joint probability distribution of features is estimated, and the joint relationship of multiple modal features is described using the joint probability distribution to obtain a multi-modal joint feature representation. The multi-modal joint feature representation is analyzed, and the high-dimensional joint features are mapped to a low-dimensional space using a dimensionality reduction algorithm to obtain a reduced-dimensional feature space. Commonly used dimensionality reduction algorithms include principal component analysis and linear discriminant analysis. For example, principal component analysis calculates the eigenvalues ​​and eigenvectors of the data covariance matrix and projects the data into the space formed by the eigenvectors to achieve dimensionality reduction. The feature importance weights are calculated based on the reduced-dimensional feature space to form a weighted eigenvector. The feature importance weights are reasonably weighted by measuring the contribution of each feature in the overall data. The formula is as follows:

[0086] ;

[0087] in, It is The importance weight of the feature, It is The variance of each feature. The weighted feature vector is obtained by multiplying each feature with its weight. The weighted feature vector is transformed nonlinearly to obtain a high-order feature representation. Nonlinear transformation can capture the complex relationship and high-order interaction information between features, making the feature representation richer and more meaningful. An activation function such as ReLU (rectified linear unit) can be used for nonlinear transformation. The formula is:

[0088] ;

[0089] in, It is the high-order feature representation after nonlinear transformation. and are weights and biases, respectively. Based on high-order feature representation, a feature similarity graph is constructed to form a feature topology. The feature similarity graph constructs a graph structure by calculating the similarity between features, where nodes represent features and edges represent similarity relationships between features. Commonly used similarity metrics include cosine similarity and Euclidean distance. Spectral clustering is performed on the feature topology to divide the features into different subspaces. Spectral clustering finds the best partitioning of the data by eigendecomposing the Laplacian matrix of the similarity graph. The Laplacian matrix is ​​defined as:

[0090] ;

[0091] in, is the degree matrix, is a similarity matrix. By performing eigendecomposition on the Laplacian matrix, the clustering structure of the data is found. According to the feature subspace division, the high-order feature representation is weighted and aggregated to obtain the deep feature vector of the fusion attention mechanism. The attention mechanism highlights the key features and suppresses the secondary features by weighting the importance of the features. The formula is as follows:

[0092] ;

[0093] in, It is The attention weight of each feature, is the attention weight matrix. Through weighted sum aggregation, the deep feature vector of the fusion attention mechanism is obtained.

[0094] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0095] (1) Input the deep feature vector into a preset multi-task deep neural network model, and perform feature mapping on the deep feature vector through the input layer to obtain a target feature mapping vector;

[0096] (2) Perform multi-head self-attention calculations on the target feature map vector to obtain context-enhanced feature representations;

[0097] (3) Perform task-shared convolution operations on the context-enhanced feature representation to obtain a multi-scale feature map, and construct a task-preset feature pyramid network based on the multi-scale feature map to obtain preliminary feature representations of electrostatic quantity, posture features, and environmental factors;

[0098] (4) Performing cross-attention processing on the preliminary feature representation to obtain the fusion features of the information interaction between tasks, and constructing the task preset fully connected layer based on the fusion features of the information interaction between tasks to obtain multiple independent prediction results, which are the static electricity prediction results, posture feature prediction results, and environmental factor prediction results;

[0099] (5) Perform soft parameter sharing analysis on multiple independent prediction results to obtain revised prediction values, and construct importance weights of multiple independent prediction results based on the revised prediction values;

[0100] (6) The importance weights and the corrected prediction values ​​are weighted and summed to obtain a comprehensive prediction vector, which is then normalized using Softmax to obtain a comprehensive grounding assessment result that includes static electricity, posture characteristics, and environmental factors.

[0101] Specifically, the deep feature vector is input into the preset multi-task deep neural network model, and the deep feature vector is subjected to feature mapping through the input layer. The purpose of feature mapping is to map the original feature vector to a feature space that is more suitable for model processing. The formula is as follows:

[0102] ;

[0103] in, is the target feature map vector, is the weight matrix, is the bias vector, is the input deep feature vector. The multi-head self-attention mechanism is calculated on the target feature map vector. The multi-head self-attention mechanism can capture the complex relationship and dependency between features. The calculation formula of the multi-head self-attention mechanism is:

[0104] Attention softmax ;

[0105] in, (Query), (Key) and (Value) is the matrix obtained by linear transformation of the feature map vector, is the scaling factor. Through the multi-head self-attention mechanism, the context-enhanced feature representation is calculated, which can contain more context information and enhance the expressiveness of the feature. The context-enhanced feature representation is subjected to task-shared convolution operations to obtain a multi-scale feature map. The convolution operation can extract local information and patterns in the feature. Through convolution operations at multiple scales, feature representations at different scales are obtained to form a multi-scale feature map. Based on the multi-scale feature map, a feature pyramid network preset for the task is constructed. The feature pyramid network can hierarchically process the multi-scale features to obtain preliminary feature representations of static electricity, posture features and environmental factors. The preliminary feature representation is subjected to cross-attention processing, which can interact and share information between different tasks, so that the feature representation of each task can integrate the information of other tasks. Through cross-attention processing, the fusion features of information interaction between tasks are obtained, and the fusion features contain complementary information between multiple tasks. According to the fusion features of information interaction between tasks, the fully connected layer preset for the task is constructed to obtain multiple independent prediction results. The fully connected layer maps the fusion features to the prediction space through linear transformation to obtain the prediction results of static electricity, posture features and environmental factors. The formula is as follows:

[0106] ;

[0107] in, It is The prediction results of the tasks are is the weight matrix, is the fusion feature, is the bias vector. Soft parameter sharing analysis is performed on multiple independent prediction results. Soft parameter sharing analysis improves the accuracy and consistency of predictions by sharing parameters between multiple tasks. Through soft parameter sharing analysis, the corrected prediction value is obtained. According to the corrected prediction value, the importance weights of multiple independent prediction results are constructed. The importance weight reflects the contribution of each prediction result in the overall prediction. Through weighted processing, the information of different prediction results can be integrated. The formula is as follows:

[0108] ;

[0109] in, It is The importance weight of the prediction results, Var It is The variance of the prediction results is calculated. The importance weights and the modified prediction values ​​are weighted and summed to obtain a comprehensive prediction vector. The weighted summation process can effectively integrate different prediction results. The comprehensive prediction vector is Softmax normalized to obtain a comprehensive grounding assessment result including static electricity, posture characteristics and environmental factors. Softmax normalization can convert the comprehensive prediction vector into a probability distribution, making the final evaluation result more intuitive and easy to understand. The formula is as follows:

[0110] ;

[0111] Among them, Softmax It is The probability of the categories, is the first comprehensive prediction vector A quantity.

[0112] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0113] (1) Perform multi-dimensional cluster analysis on the comprehensive grounding assessment results to obtain the grounding status classification, and perform time series analysis of historical data based on the grounding status classification to obtain the future grounding status probability distribution;

[0114] (2) Generate multi-level grounding risk assessment information based on the probability distribution of future grounding states, and perform association rule mining on the multi-level grounding risk assessment information to obtain a potential risk factor association network;

[0115] (3) Generate a force-directed graph layout based on the potential risk factor association network, and perform heat map superposition processing on the force-directed graph layout to obtain a comprehensive risk intensity distribution map;

[0116] (4) Perform threshold segmentation based on the comprehensive risk intensity distribution map to obtain the high-risk area positioning results, and perform search optimization on the high-risk area positioning results to obtain the security inspection equipment control strategy;

[0117] (5) Generate a security inspection equipment execution instruction sequence corresponding to the security inspection equipment according to the security inspection equipment control strategy.

[0118] Specifically, a multi-dimensional cluster analysis is performed on the comprehensive grounding assessment results. Cluster analysis is a technique that divides data points into multiple groups so that data points in the same group are more similar to each other. Commonly used clustering methods include K-means, hierarchical clustering, and DBSCAN. Assuming that K-means clustering is used, the number of clusters is selected. , and then optimize the objective function by iteration:

[0119] ;

[0120] in, It is Clusters, It is The centroid of the cluster, is a data point. By iteratively updating the centroid and redistributing the data points, the grounding state classification is obtained and different grounding state categories are identified. According to the grounding state classification, the historical data time series analysis is performed. The time series analysis can reveal the law of data changes over time. The autoregressive moving average model is used to predict the future grounding state. The autoregressive moving average model consists of three parameters , and The ARIMA model is composed of the following:

[0121] ;

[0122] in, is the current value, is a constant term, is the autoregressive coefficient, is the moving average coefficient, is white noise. Through the analysis of historical data time series, the probability distribution of future grounding status is obtained, and the possible grounding status at different time points is predicted. Multi-level grounding risk assessment information is generated according to the probability distribution of future grounding status. Multi-level grounding risk assessment evaluates future grounding status by considering different risk levels and influencing factors. Association rule mining is performed on multi-level grounding risk assessment information. Association rule mining is a method to discover interesting relationships from large data sets. Commonly used algorithms include Apriori and FP-Growth. The basic form of association rules is ,in and is an item set, support and confidence are important indicators for evaluating rules:

[0123] Support ;

[0124] ;

[0125] in, It also contains and The number of records, is included The number of records, is the total number of records. Through association rule mining, potential risk factor association networks are obtained, which reveal the relationship between different risk factors. A force-directed graph layout is generated based on the potential risk factor association network. The force-directed graph layout simulates the force action in physics so that the distance between nodes reflects the closeness of their relationship. Each node represents a risk factor, and the edges between nodes represent their association relationship. The nodes are laid out using a force-directed algorithm, and an equilibrium state is finally reached by calculating the attraction and repulsion between each pair of nodes, so that the graph structure is clearly presented. The force-directed graph layout is processed by heat map superposition, and the heat map reflects the distribution of risk intensity through color changes. High-risk areas are represented by darker colors, and low-risk areas are represented by lighter colors, forming a comprehensive risk intensity distribution map. Threshold segmentation is performed based on the comprehensive risk intensity distribution map to identify high-risk areas. Threshold segmentation sets a risk value threshold and marks areas above the threshold as high-risk areas. The high-risk area positioning results are searched and optimized, and optimization algorithms such as genetic algorithms or particle swarm optimization are used for optimization to find the optimal security inspection equipment control strategy. The execution instruction sequences corresponding to the security inspection equipment are generated according to the optimized control strategy. These instruction sequences can guide the security inspection equipment to take specific measures in high-risk areas, such as strengthening monitoring, adjusting equipment parameters, etc., thereby improving the safety and stability of the security inspection system.

[0126] In a specific embodiment, the wireless intelligent multifunctional grounding monitoring method further includes the following steps:

[0127] (1) Parse and process the execution instruction sequence of the security inspection equipment to obtain the target execution action set, and execute the instructions on the security inspection equipment according to the target execution action set to obtain the original execution result data;

[0128] (2) Perform multi-dimensional feature extraction on the original execution result data to obtain the execution effect feature vector, and construct a multi-objective optimization function based on the execution effect feature vector to obtain the target set for model optimization;

[0129] (3) Perform weight assignment on the target set of model optimization to obtain a weighted optimization objective function;

[0130] (4) Generate an initial population of the multi-task deep neural network model based on the weighted optimization objective function to obtain a set of candidate model parameters;

[0131] (5) Perform non-dominated sorting on the candidate model parameter set to obtain the Pareto optimal solution set, and perform the selection, crossover and mutation operations of the genetic algorithm based on the Pareto optimal solution set to obtain a new generation of population;

[0132] (6) Reconstruct the model and evaluate the performance of the new generation population to obtain the optimized model candidate set;

[0133] (7) Cross-validation is performed based on the model candidate set and the original execution result data to obtain the metal security inspection verification model corresponding to the security inspection equipment.

[0134] Specifically, the security inspection equipment execution instruction sequence is parsed and processed. The parsing process decomposes the instruction sequence into specific execution actions through the parser to obtain the target execution action set. The target execution action set includes a series of specific operation instructions, which are sent to the security inspection equipment for execution. Through the execution of these instructions, the original execution result data is obtained, and all feedback information of the equipment during the execution process is recorded, including whether the execution is successful or not, execution time, equipment status, etc. Multi-dimensional feature extraction is performed on the original execution result data. Through the feature extraction algorithm, the key features of the execution effect are extracted from the original data to form an execution effect feature vector. Feature extraction can use dimensionality reduction algorithms such as principal component analysis to convert high-dimensional raw data into low-dimensional feature vectors. The formula is as follows:

[0135] ;

[0136] in, is the eigenvector, is the eigenvector matrix, is the original data. Through feature extraction, the key factors affecting the execution effect are captured. A multi-objective optimization function is constructed based on the execution effect feature vector. This function combines multiple optimization objectives to form a target set for model optimization. Assuming there are multiple optimization objectives , then the multi-objective optimization function can be expressed as:

[0137] minimize ;

[0138] in, is the eigenvector, It is a multi-objective optimization function. By optimizing these objectives, the overall performance of the system is improved. The objective set of model optimization is weighted to form a weighted optimization objective function. The weight allocation determines the contribution of each objective in the overall optimization by setting the importance weight of each objective. The formula is as follows:

[0139] minimize ;

[0140] in, is the weighted optimization objective function, It is The weight of the target, It is objective functions. By weight allocation, the conflicts between different objectives are balanced to find a comprehensive optimization solution. According to the weighted optimization objective function, the initial population of the multi-task deep neural network model is generated to form a candidate model parameter set. The initial population generation generates a set of possible model parameter combinations through random or heuristic methods. The candidate model parameter set is processed by non-dominated sorting. Non-dominated sorting compares the advantages and disadvantages of different parameter combinations to screen out a set of solutions that are not dominated by other combinations, namely the Pareto optimal solution set. Each solution in the Pareto optimal solution set performs well in some objectives and is not inferior to other solutions in other objectives. After obtaining the Pareto optimal solution set, the selection, crossover and mutation operations of the genetic algorithm are performed to form a new generation of population. The selection operation ensures the transmission of excellent characteristics, the crossover operation generates new solutions by combining the characteristics of different solutions, and the mutation operation introduces random changes to improve the diversity of the population. Through these operations, the model parameters are continuously optimized and gradually approach the optimal solution. The new generation population is reconstructed and its performance is evaluated. The reconstruction process rebuilds the model with new parameters, and the performance evaluation measures the performance of the model in practical applications, thereby screening out the candidate set of models with excellent performance. The following formula can be used for model reconstruction:

[0141] ;

[0142] in, is the output of the model, is the input data, are model parameters, and by using the new parameters Reconstruct the model and select the best performing model candidate set through performance evaluation indicators such as accuracy, precision, recall, etc. Cross-validation is performed based on the model candidate set and the original execution result data. Cross-validation is a method for evaluating model performance by dividing the data set into multiple subsets, repeatedly training and validating, and finally obtaining the average performance indicator of the model. The basic steps of cross-validation include: dividing the data set into k subsets, using each subset as a validation set and other subsets as training sets, repeating k times, and finally calculating the average result of k validations. The formula is as follows:

[0143] CV score ;

[0144] Among them, CV_score is the average score of cross validation, It is Through cross-validation, the generalization ability of each model candidate set is evaluated to select the optimal model and finally obtain the metal security inspection verification model corresponding to the security inspection equipment.

[0145] The wireless intelligent multifunctional grounding monitoring method in the embodiment of the present application is described above. The wireless intelligent multifunctional grounding monitoring device in the embodiment of the present application is described below. Figure 2 In the embodiment of the present application, an embodiment of the wireless intelligent multifunctional grounding monitoring device includes:

[0146] Building module 201, used to integrate multiple sensors on security inspection equipment and build a multi-modal perception system with electric field, pressure, infrared and image acquisition functions;

[0147] The transmission module 202 is used to collect the original data set of the passing personnel through the multimodal perception system, and wirelessly transmit the original data set to obtain a layered encrypted data stream supporting multi-protocol adaptive switching;

[0148] The processing module 203 is used to perform distributed edge computing preprocessing on the layered encrypted data stream to obtain multimodal feature data;

[0149] The encoding module 204 is used to perform multimodal fusion and feature encoding processing on the multimodal feature data to obtain a deep feature vector fused with an attention mechanism;

[0150] An analysis module 205 is used to input the deep feature vector into a preset multi-task deep neural network model to analyze the grounding status of personnel and obtain a comprehensive grounding assessment result including static electricity, posture characteristics and environmental factors;

[0151] The generation module 206 is used to perform centralized multi-dimensional analysis and associated visualization processing on the comprehensive grounding assessment results through the intelligent monitoring management platform, and generate a security inspection equipment execution instruction sequence corresponding to the passing personnel.

[0152] Through the collaboration of the above components, by integrating multiple sensors such as electric field, pressure, infrared and image acquisition, all-round and multi-angle monitoring of personnel grounding status is realized, which greatly improves the comprehensiveness and accuracy of data collection. Distributed edge computing technology is used to pre-process the encrypted data stream, which significantly reduces the amount of data transmission and the burden of the central server, and improves the real-time response capability of the system. Multimodal fusion and attention mechanism are used to realize the deep feature extraction of personnel grounding status in complex environments, and enhance the system's ability to identify minor grounding anomalies. Through the multi-task deep neural network model, static electricity, posture characteristics and environmental factors are analyzed simultaneously, which improves the comprehensiveness and accuracy of grounding status assessment. The introduction of non-dominated genetic algorithm to dynamically optimize the model enables the system to continuously adapt to new grounding monitoring scenarios and requirements, and improves the scalability and long-term effectiveness of the system. Through the intelligent monitoring management platform, multi-dimensional analysis and correlation visualization are provided to provide intuitive and effective decision-making support for managers, and improve the efficiency of grounding risk management. Based on the comprehensive evaluation results, the execution instruction sequence of security inspection equipment is automatically generated, which realizes the intelligent linkage between grounding monitoring and security inspection equipment, and improves the detection accuracy of metal security inspection equipment.

[0153] The present application also provides a wireless intelligent multifunctional grounding monitoring device, which includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the wireless intelligent multifunctional grounding monitoring method in the above-mentioned embodiments.

[0154] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the wireless intelligent multifunctional grounding monitoring method.

[0155] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0156] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0157] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A wireless intelligent multifunctional grounding monitoring method, characterized in that: The method comprises: Integrate multiple sensors into security inspection equipment to build a multimodal perception system with electric field, pressure, infrared and image acquisition functions; The multimodal sensing system collects the original data set of the passing personnel, and wirelessly transmits the original data set to obtain a layered encrypted data stream that supports multi-protocol adaptive switching; specifically includes: collecting the electric field strength data of the passing personnel through the electric field sensor in the multimodal sensing system, and performing fast Fourier transform on the electric field strength data to obtain the electrostatic spectrum characteristics of the human body; adaptively filtering the electric field strength data according to the electrostatic spectrum characteristics of the human body to obtain the electrostatic charge data after noise reduction; collecting the original pressure data of the passing personnel through the pressure sensor in the multimodal sensing system, and performing wavelet decomposition on the original pressure data to obtain a multi-scale pressure feature matrix; analyzing the original pressure data according to the multi-scale pressure feature matrix to obtain the gait feature data of the personnel; collecting the passing personnel through the infrared sensor in the multimodal sensing system temperature distribution data, and Gaussian mixture clustering is performed on the temperature distribution data to obtain a human hot spot area coordinate set; regional growth segmentation is performed on the temperature distribution data according to the human hot spot area coordinate set to obtain human contour feature data; the original visible light image is acquired through the image acquisition device in the multimodal perception system, and a multi-scale convolution operation is performed on the original visible light image to obtain a deep image feature map; semantic segmentation is performed on the original visible light image according to the deep image feature map to obtain personnel clothing material classification data; timestamp alignment and data format unification are performed on the denoised electrostatic charge data, the personnel gait feature data, the human contour feature data, and the personnel clothing material classification data to obtain a multimodal synchronization data packet; dynamic encryption and layered encapsulation are performed on the multimodal synchronization data packet to obtain a layered encrypted data stream that supports multi-protocol adaptive switching; Performing distributed edge computing preprocessing on the layered encrypted data stream to obtain multimodal feature data; The multimodal feature data is subjected to multimodal fusion and feature encoding processing to obtain a deep feature vector of a fusion attention mechanism; specifically comprising: performing feature normalization on the multimodal feature data to obtain a standardized feature matrix, and calculating the mutual information between modalities based on the standardized feature matrix to obtain a modal correlation measure; performing threshold screening on the modal correlation measure to obtain a set of high-correlation modal pairs, and performing feature joint probability distribution estimation based on the set of high-correlation modal pairs to obtain a multimodal joint feature representation; analyzing the multimodal joint feature representation to obtain a reduced-dimensional feature space, and calculating feature importance weights based on the reduced-dimensional feature space to obtain a weighted feature vector; performing nonlinear transformation on the weighted feature vector to obtain a high-order feature representation, and constructing a feature similarity graph based on the high-order feature representation to obtain a feature topological structure; performing spectral clustering on the feature topological structure to obtain a feature subspace division, and performing attention weighting and aggregation on the high-order feature representation based on the feature subspace division to obtain a deep feature vector of a fusion attention mechanism; Inputting the deep feature vector into a preset multi-task deep neural network model to analyze the grounding status of personnel, and obtaining a comprehensive grounding assessment result including static electricity, posture characteristics and environmental factors; The comprehensive grounding assessment results are subjected to centralized multi-dimensional analysis and associated visual processing through an intelligent monitoring and management platform, and a security inspection equipment execution instruction sequence corresponding to the passing personnel is generated.

2. The wireless intelligent multifunctional grounding monitoring method according to claim 1 is characterized in that: The step of performing distributed edge computing preprocessing on the layered encrypted data stream to obtain multimodal feature data includes: Decrypting and parsing the layered encrypted data stream to obtain an original multimodal data stream, and performing time synchronization on each modal data according to timestamp information in the original multimodal data stream to obtain an aligned multimodal data matrix; Performing outlier detection and interpolation on the aligned multimodal data matrix to obtain a cleaned multimodal data set, and normalizing and standardizing each modality data according to the cleaned multimodal data set to obtain a standardized feature matrix; Performing feature dimensionality reduction on the standardized feature matrix to obtain a feature vector set after dimensionality reduction, and analyzing each modal data according to the feature vector set after dimensionality reduction to obtain separated independent feature components; Performing wavelet transform on the separated independent feature components to obtain a multi-scale time-frequency feature map, and performing adaptive feature extraction on each modal data according to the multi-scale time-frequency feature map to obtain a feature descriptor of the modality; Nonlinear manifold learning is performed on the feature descriptors of the modalities to obtain a low-dimensional embedding feature space, and sparse encoding is performed on each modal feature according to the low-dimensional embedding feature space to obtain multimodal feature data.

3. The wireless intelligent multifunctional grounding monitoring method according to claim 1 is characterized in that: The deep feature vector is input into a preset multi-task deep neural network model to analyze the grounding status of personnel, and a comprehensive grounding assessment result including static electricity, posture characteristics and environmental factors is obtained, including: Inputting the deep feature vector into a preset multi-task deep neural network model, and performing feature mapping processing on the deep feature vector through an input layer to obtain a target feature mapping vector; Performing multi-head self-attention mechanism calculation on the target feature map vector to obtain a context-enhanced feature representation; Performing a task-shared convolution operation on the context-enhanced feature representation to obtain a multi-scale feature map, and constructing a feature pyramid network preset for the task based on the multi-scale feature map to obtain a preliminary feature representation of static electricity, posture features, and environmental factors; Performing cross-attention processing on the preliminary feature representation to obtain fusion features of information interaction between tasks, and constructing a fully connected layer preset for the task according to the fusion features of information interaction between tasks to obtain multiple independent prediction results, wherein the multiple independent prediction results are static electricity quantity prediction results, posture feature prediction results, and environmental factor prediction results; Performing soft parameter sharing analysis on the multiple independent prediction results to obtain a revised prediction value, and constructing importance weights of the multiple independent prediction results according to the revised prediction value; The importance weight and the modified prediction value are weighted and summed to obtain a comprehensive prediction vector, and the comprehensive prediction vector is Softmax normalized to obtain a comprehensive grounding assessment result including static electricity, posture characteristics and environmental factors.

4. The wireless intelligent multifunctional grounding monitoring method according to claim 3 is characterized in that: The intelligent monitoring management platform performs centralized multi-dimensional analysis and associated visualization processing on the comprehensive grounding assessment results, and generates a security inspection equipment execution instruction sequence corresponding to the passing personnel, including: Performing a multi-dimensional cluster analysis on the comprehensive grounding assessment results to obtain a grounding status classification, and performing a historical data time series analysis based on the grounding status classification to obtain a future grounding status probability distribution; Generate multi-level grounding risk assessment information according to the future grounding state probability distribution, and perform association rule mining on the multi-level grounding risk assessment information to obtain a potential risk factor association network; Generating a force-directed graph layout according to the potential risk factor association network, and performing heat map superposition processing on the force-directed graph layout to obtain a comprehensive risk intensity distribution map; Performing threshold segmentation according to the comprehensive risk intensity distribution map to obtain high-risk area positioning results, and performing search optimization on the high-risk area positioning results to obtain a security inspection equipment control strategy; A security inspection device execution instruction sequence corresponding to the security inspection device is generated according to the security inspection device control strategy.

5. The wireless intelligent multifunctional grounding monitoring method according to claim 1, characterized in that: The wireless intelligent multifunctional grounding monitoring method also includes: Parsing the execution instruction sequence of the security inspection device to obtain a target execution action set, and executing the instructions on the security inspection device according to the target execution action set to obtain original execution result data; Performing multi-dimensional feature extraction on the original execution result data to obtain an execution effect feature vector, and constructing a multi-objective optimization function based on the execution effect feature vector to obtain a target set for model optimization; Performing weight distribution processing on the target set optimized by the model to obtain a weighted optimization target function; Generating an initial population of the multi-task deep neural network model according to the weighted optimization objective function to obtain a candidate model parameter set; Performing non-dominated sorting on the candidate model parameter set to obtain a Pareto optimal solution set, and performing selection, crossover and mutation operations of a genetic algorithm according to the Pareto optimal solution set to obtain a new generation population; Reconstructing the model and evaluating the performance of the new generation population to obtain an optimized model candidate set; A cross-validation is performed based on the model candidate set and the original execution result data to obtain a metal security inspection verification model corresponding to the security inspection equipment.

6. A wireless intelligent multifunctional grounding monitoring device, characterized in that: The device is used to perform the wireless intelligent multifunctional grounding monitoring method according to any one of claims 1 to 5, comprising: Building modules for multi-sensor integration of security inspection equipment to build a multi-modal perception system with electric field, pressure, infrared and image acquisition functions; A transmission module, used to collect an original data set of a person passing through the multimodal perception system, and wirelessly transmit the original data set to obtain a layered encrypted data stream supporting multi-protocol adaptive switching; A processing module, used for performing distributed edge computing preprocessing on the layered encrypted data stream to obtain multimodal feature data; An encoding module, used for performing multimodal fusion and feature encoding processing on the multimodal feature data to obtain a deep feature vector fused with an attention mechanism; An analysis module, used for inputting the deep feature vector into a preset multi-task deep neural network model to analyze the grounding status of personnel, and obtaining a comprehensive grounding assessment result including static electricity, posture characteristics and environmental factors; A generation module is used to perform centralized multi-dimensional analysis and associated visualization processing on the comprehensive grounding assessment results through an intelligent monitoring management platform, and generate a security inspection equipment execution instruction sequence corresponding to the passing personnel.

7. A wireless intelligent multifunctional grounding monitoring device, characterized in that: The wireless intelligent multifunctional grounding monitoring device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the wireless intelligent multifunctional grounding monitoring device executes the wireless intelligent multifunctional grounding monitoring method as described in any one of claims 1-5.

8. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the wireless intelligent multifunctional grounding monitoring method as described in any one of claims 1 to 5 is implemented.

Images