A comprehensive safety supervision system and method for the postal industry

Abstract

The present invention relates to a comprehensive security supervision system and method for the postal industry, specifically in the field of information monitoring. It collects video data of relevant scenarios in the postal industry, constructs an object detection network based on a feature extraction network structure of convolution and pooling and combined with residual blocks, and outputs the final recognition result of the scenario category. It collects picture data of relevant scenarios in the postal industry, selects LSTM units to build a recurrent neural network model, and outputs the predicted classification result of the picture data with category labels. It collects the packet image data of relevant scenarios of passing through security inspection, uses the SVM classification algorithm to separate the packet image data with different labels, effectively distinguishes the image data with different labels, improves the recognition accuracy of the packet image data, determines the label type of the packet image data through the positive and negative class relationship of the prediction results of the packet image data, provides the management function of early warning rules and supports the management functions of adding, modifying, and deleting multiple types of business rules, so as to improve the operation efficiency and the quality of express delivery services.

Description

Technical Field

[0001] The present invention relates to the field of information monitoring, and more specifically, to a comprehensive security supervision system and method for the postal industry. Background Art

[0002] With the rapid development of e-commerce, express delivery has gradually become an indispensable part of daily life. While the express delivery industry has made great contributions to the development of the national economy, many chaotic phenomena have emerged. However, the current security supervision of the express delivery industry is weak, the supervision methods are single, and the informatization support is insufficient, which cannot adapt to the rapid development speed of the express delivery industry and cannot effectively guarantee the safe and healthy development of the express delivery industry.

[0003] The production process of the postal industry is decentralized and the data generation frequency is high, which belongs to a typical data-intensive industry. At present, the informatization support can monitor the market operation situation at the city and district / county levels, which can meet the relatively macro supervision needs, but it is difficult to penetrate deeper levels of the network such as streets, outlets and terminals, lacking the ability to deeply collect data, which will inevitably lead to insufficient or lack of risk prediction ability. Without the necessary risk prediction ability, the support for security supervision basically stays at the level of presenting the current situation and history, with insufficient support, and even becomes a disguised burden under certain conditions.

[0004] Therefore, there is an urgent need for a comprehensive security supervision system and method for the postal industry. Summary of the Invention

[0005] In view of the technical problems existing in the prior art, the present invention provides a comprehensive security supervision system and method for the postal industry to solve the problems raised in the above background art.

[0006] The technical solution of the present invention to solve the above technical problems is as follows: A comprehensive security supervision system for the postal industry, comprising:

[0007] In a preferred embodiment, it includes a video security inspection and supervision module, a receiving and inspection system supervision module, a screening security inspection system supervision module, and a business pressure risk warning module;

[0008] Video security inspection and supervision module: Using a camera to collect video data of relevant scenes in the postal industry and using the VOTT annotation tool for regional annotation, constructing a feature extraction network structure based on convolution and pooling and combining a target detection network with residual blocks;

[0009] Receiving and inspection system supervision module: Connecting to the picture database of relevant scenes in the postal industry to collect picture data of relevant scenes in the postal industry, using the Label Me annotation tool for bounding box selection and specifying category labels, selecting LSTM units to build a recurrent neural network model, and constructing an autoencoder model;

[0010] Cross - machine security inspection system supervision module: Around the security inspection machine equipment, cameras are used to collect the image data of the package collection related scenes for cross - machine security inspection. The SVM classification algorithm is used to separate the image data of different tags, and intelligent security inspection alarm information is automatically generated.

[0011] Business pressure risk warning module: Provides the management function of warning rules and supports the management functions of adding, modifying, and deleting multiple types of business rules. The historical data is used to test the rule matching degree, and the formal rules are submitted as the basis for automatic warning according to the rules.

[0012] In a preferred embodiment, the video security patrol supervision module is divided into a video acquisition and annotation sub - unit and a video scene recognition sub - unit, and specifically includes the following:

[0013] Video acquisition and annotation sub - unit: Cameras are used to collect the video data of the scenes related to the postal industry and transmit it back to the server for storage in the form of files. The VOTT annotation tool is used to select the rectangular annotation shape to annotate the regions of the video data, and corresponding task tags are added to each annotated region.

[0014] Video scene recognition sub - unit: A target detection network based on the feature extraction network structure of convolution and pooling and combined with residual blocks is constructed. The video data of the scenes related to the postal industry is input. Convolution layers are added to extract the basic features of the video data through multiple convolutional kernels. Multiple convolutional layers and skip connections are combined to form residual units. After each residual unit, ReLU is used as the activation function layer to introduce non - linearity and convolutional operations are performed on the video data using convolutional kernels. The specific formula for the convolutional operation is:

[0015]

[0016] Among them, A(i,j) represents an element of the basic features of the video data, I(m,n) represents the pixel value of the basic features of the video data, K(i - m,j - n) represents the weight value of the basic features of the video data, m and n respectively represent the indexes of the rows and columns inside the convolutional kernel, i and j respectively represent the indexes of the rows and columns of the basic features of the video data. Through the global average pooling layer of the classification head and the down - sampling operation on the basic features of the video data of the convolutional layer, the first convolution and pooling operations are repeated iteratively. The high - level features of the video data are flattened into a one - dimensional vector through the fully - connected layers of the classification head and the regression head and input into the fully - connected layer. The fully - connected layer uses the weight matrix to map the high - level features of the video data to different scene categories of the video data, and the final scene category recognition result is output.

[0017] In a preferred embodiment, the receiving and inspection system supervision module is divided into a picture acquisition and annotation sub - unit and a picture scene recognition sub - unit, and specifically includes the following:

[0018] Image acquisition and annotation subunit: Connect to the image database of the postal industry-related scenarios to collect image data of the postal industry-related scenarios and transmit it back to the server for storage in the form of files. Use the LabelMe annotation tool to box each object that appears in the image data of the postal industry-related scenarios and assign category labels.

[0019] Image scene recognition subunit: Select the LSTM unit to build a recurrent neural network model. According to the current timestamp, input the image data with category labels and the hidden state of the previous timestamp, and calculate the activation value of the input layer through the Sigmoid activation function. The specific formula is:

[0020] T l = S(w l *[I t-1 ,x t +b l )

[0021] Among them, T l represents the activation value of the input layer, S() represents the Sigmoid activation function, w l represents the weight matrix, I t-1 represents the hidden state of the previous timestamp, x t represents the image data with category labels input at the current timestamp, b l represents the bias term. The output layer selects a fully connected layer and calculates the activation value of the output layer through the Sigmoid activation function according to the image data with category labels input at the current timestamp, the hidden state of the previous timestamp, and the cell state of the image data with category labels input at the current timestamp. The specific formula is:

[0022] T o = S(w o *[I t-1 ,x t +b o )

[0023] Among them, T o represents the activation value of the forgetting layer, S() represents the Sigmoid activation function, w o represents the weight matrix, I t-1 represents the hidden state of the previous timestamp, x t represents the image data with category labels input at the current timestamp, b o represents the bias term. Output the predicted classification result of the image data with category labels through the fully connected layer, construct an autoencoder model to input the predicted classification result of the image data with category labels and map it to the latent space, and perform supervised training on the autoencoder through the predicted classification result of the image data with category labels.

[0024] In a preferred embodiment, the in-transit security inspection system supervision module is divided into a security inspection collection and annotation sub-unit and a security inspection scenario recognition sub-unit, and specifically includes the following:

[0025] Security inspection collection and annotation sub-unit: Use a camera around the security inspection machine equipment to collect packet image data of the relevant scenarios of in-transit security inspection and transmit it back to the server, store it in the form of a file, use the LabelMe annotation tool to box and assign different labels to the packet image data of the relevant scenarios of in-transit security inspection, and extract the features of the packet image data using color histograms.

[0026] Security inspection scenario recognition sub-unit: Use the SVM classification algorithm to separate the packet image data with different labels and use the kernel function to map the features of the packet image data with different labels to a high-dimensional space to construct an optimal hyperplane, so as to better separate the packet image data with different labels in the high-dimensional space, maximize the projection points of the separated packet image data on this hyperplane. The basic formula form of the SVM is:

[0027] f(x) = sign(w * x + b)

[0028] Where f(x) represents the prediction result of the packet image data, w represents the weight vector, which is used to represent the direction of the hyperplane, x represents the feature vector of the packet image data with different labels, b represents the bias term. The label type of the packet image data is judged through the positive and negative class relationships of f(x), and it is fed back to the intelligent security inspection alarm information database through positive and negative class signals, automatically generating intelligent security inspection alarm information, recording the abnormal express items found in the security inspection, reporting them to the postal management department, and reporting to relevant departments according to the situation.

[0029] In a preferred embodiment, the business pressure risk warning module provides the management function of warning rules and supports the management functions of adding, modifying, and deleting multiple types of business rules. After the user with permission completes the rule setting, use historical data to test the rule matching degree and submit the formal rules as the basis for automatic warning according to the rules. Continuously improve the warning management function according to the actual situation. Calculate the data change trajectories of the historical business and the number of couriers at each network point through historical data analysis, and depict the characteristics of the historical business changes and the number of couriers in the daily operation of the network point. Compare with the current situation of the network point in real time, mark the current operation status of each network point, and give timely warnings for the network points that are about to burst, existing, and backlogged. Automatically generate and comprehensively set the classification and grading of burst, existing, and backlogged detections in combination with manual assistance. According to the different types of events, set the event types in advance manually and set the judgment rules.

[0030] In a preferred embodiment, it specifically includes the following steps:

[0031] S101. Collect video data of relevant scenarios in the postal industry and perform regional annotation using the VOTT annotation tool. Construct a target detection network structure based on convolution and pooling and combine it with a residual block, and output the final recognition result of the scenario category;

[0032] S102. Collect picture data of relevant scenarios in the postal industry, use the Label Me annotation tool to perform bounding box selection and specify category labels, select LSTM units to build a recurrent neural network model, output the predicted classification result of the picture data with category labels, and construct an autoencoder model to integrate the predicted classification result of the picture data;

[0033] S103. Collect the bagging image data of relevant scenarios in the over-machine security inspection, extract the features of the bagging image data using the color histogram, use the SVM classification algorithm to separate the bagging image data with different labels, and judge the label type of the bagging image data through the positive and negative class relationship of the prediction result of the bagging image data;

[0034] S104. Provide the management function of warning rules and support the management functions of adding, modifying, and deleting multiple types of business rules. Use historical data to test the rule matching degree, and submit formal rules as the basis for automatic warning according to the rules.

[0035] The beneficial effects of the present invention are as follows: By using the VOTT annotation tool to perform regional annotation on the collected video data of relevant scenarios in the postal industry, it helps to construct a target detection network, ensures that the video data is clear and reduces blur and distortion problems, improves the recognition accuracy. By using multiple convolutional kernels to extract the basic features of the video data, it effectively improves the recognition ability of the scenario category. By using the LabelMe annotation tool to annotate the collected picture data, it helps to construct the labeled data set required for training the model and improves the model accuracy. By selecting LSTM units to build a recurrent neural network model and using an autoencoder model to integrate the predicted classification result of the picture data, it can effectively process the picture data related to time series and extract high-level features, which helps to improve the data retrieval efficiency and management. By extracting the features of the bagging image data using the color histogram, it helps to capture the color distribution information of the image. By using the SVM classification algorithm to classify the bagging image data, it can effectively distinguish the image data with different labels and improve the recognition accuracy of the bagging image data. By judging the label type of the bagging image data through the positive and negative class relationship of the prediction result and combining with the SVM classification algorithm for security inspection anomaly detection, it generates intelligent security inspection alarm information to help security inspection personnel quickly identify abnormal packages. By using historical data to verify the rule matching degree, it can more accurately judge the situations that meet the rule requirements, improve the accuracy and reliability of the warning, and automatically give warnings according to the preset rules to realize the real-time monitoring and warning of the operation status of the network points, help to timely respond to problems such as impending warehouse explosion, existing, and backlog, and improve the operation efficiency and express delivery service quality. Brief Description of the Drawings

[0036] Figure 1 This is the flowchart of the method of the present invention;

[0037] Figure 2 This is the block diagram of the system structure of the present invention. Detailed Embodiments

[0038] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0039] In the description of the present application, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of the described features. In the description of the present application, "a plurality of" means two or more unless otherwise specifically defined.

[0040] In the description of the present application, the term "for example" is used to mean "serving as an example, illustration, or explanation". Any embodiment described as "for example" in the present application is not necessarily construed as being more preferred or having more advantages than other embodiments. In order for any person skilled in the art to implement and use the present invention, the following description is given. In the following description, details are set forth for purposes of explanation. It should be understood that those of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other instances, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope that conforms to the principles and features disclosed in the present application.

[0041] Embodiment 1

[0042] This embodiment provides a comprehensive safety supervision method for the postal industry as shown in Figure 1 and specifically includes the following steps:

[0043] S101. Collect video data of relevant scenarios in the postal industry and perform regional annotation using the VOTT annotation tool, construct a feature extraction network structure based on convolution and pooling and a target detection network combined with residual blocks, and output the final scene category recognition result;

[0044] S102. Collect image data of relevant scenarios in the postal industry, use the Label Me annotation tool to perform bounding box selection and specify category labels, select LSTM units to build a recurrent neural network model, output the predicted classification results of the image data with category labels, and construct an autoencoder model to integrate the predicted classification results of the image data;

[0045] S103. Collect the packet collection image data of relevant scenarios in the checkpoint security inspection, extract the features of the packet collection image data using the color histogram, use the SVM classification algorithm to separate the packet collection image data with different labels, and judge the label type of the packet collection image data through the positive and negative class relationships of the predicted results of the packet collection image data;

[0046] S104. Provide the management function of warning rules and support the management functions of adding, modifying, and deleting multiple types of business rules, use historical data to test the rule matching degree, and submit formal rules as the basis for automatic warning according to the rules.

[0047] Embodiment 2

[0048] This embodiment provides a comprehensive security supervision system for the postal industry as shown in Figure 2 including: a video security inspection and supervision module, a receiving and inspection system supervision module, a checkpoint security inspection system supervision module, and a business pressure risk warning module;

[0049] Video security inspection and supervision module: Use a camera to collect video data of relevant scenarios in the postal industry and use the VOTT annotation tool for area annotation, and construct a target detection network based on the feature extraction network structure of convolution and pooling and combined with residual blocks;

[0050] Receiving and inspection system supervision module: Connect to the image database of relevant scenarios in the postal industry to collect image data of relevant scenarios in the postal industry, use the Label Me annotation tool to perform bounding box selection and specify category labels, select LSTM units to build a recurrent neural network model, and construct an autoencoder model;

[0051] Checkpoint security inspection system supervision module: Use a camera around the security inspection machine equipment to collect the packet collection image data of relevant scenarios in the checkpoint security inspection, use the SVM classification algorithm to separate the packet collection image data with different labels, and automatically generate intelligent security inspection alarm information;

[0052] Business pressure risk warning module: Provide the management function of warning rules and support the management functions of adding, modifying, and deleting multiple types of business rules, use historical data to test the rule matching degree, and submit formal rules as the basis for automatic warning according to the rules;

[0053] S101. Collect video data of relevant scenarios in the postal industry and perform regional annotation using the VOTT annotation tool. Construct a target detection network based on a feature extraction network structure of convolution and pooling and combined with residual blocks, and output the final recognition result of the scenario category.

[0054] Further, use a camera to collect video data of relevant scenarios in the postal industry and transmit it back to the server for storage in the form of a file. Use the VOTT annotation tool to select the rectangular annotation shape to perform regional annotation on the video data, and add corresponding task labels to each annotation area, including target detection, target tracking, and target recognition, so as to quickly and accurately complete the regional annotation task of the video data.

[0055] Further, set the camera resolution ≥ 1080P to ensure that the video data reduces blurring, screen distortion problems. Fix the camera shooting angle so that the scene in the video data is clear and there is no occlusion. In the 1080P video data, ensure that the resolution of the human head of internal enterprise personnel in the picture is ≥ 40×40, and the overlapping and occluded parts of the human head in the picture do not exceed 20%. Ensure that the illumination range of the camera scene is between 70 lx - 600 lx. If the light intensity is insufficient, perform white light filling.

[0056] Further, construct a target detection network based on a feature extraction network structure of convolution and pooling and combined with residual blocks. Input the video data of relevant scenarios in the postal industry, add a convolutional layer to extract the basic features of the video data through multiple convolutional kernels. Combine multiple convolutional layers and skip connections to form residual units, and form residual blocks with the residual units. Use skip connections to effectively solve the gradient vanishing problem in the network structure. After each residual unit, use ReLU as the activation function layer to introduce non-linearity and perform convolution operations on the video data using convolutional kernels to increase the stability and non-linearity of the network structure. The specific formula for the convolution operation is:

[0057]

[0058] Among them, A(i,j) represents an element of the basic features of video data, I(m,n) represents the pixel value of the basic features of video data, K(i - m,j - n) represents the weight value of the basic features of video data, m and n respectively represent the row and column indices inside the convolution kernel, and i and j respectively represent the row and column indices of the basic features of video data. By means of the global average pooling layer of the classification head and performing a downsampling operation on the basic features of the video data of the convolutional layer, it is used to reduce the size of the basic features of the video data and retain important feature information. The first convolution and pooling operations are iteratively repeated to extract the high-level features of the video data. The high-level features of the video data are flattened into a one-dimensional vector through the fully connected layers of the classification head and the regression head and input into the fully connected layer. The fully connected layer is used to map the high-level features of the video data to different scene categories of the video data through a weight matrix, and the final scene category recognition result is output.

[0059] Furthermore, the scene category recognition results are divided into personnel detection, mail detection, operation detection, and behavior detection. Among them, personnel detection includes personnel off-duty detection, face recognition, human identity verification, and hair accessories and clothing detection. Among them, mail detection includes the detection of the stacking state of mails and express parcels, the detection of mails and express parcels landing on the ground, the detection of damaged mails and express parcels, the detection of the opening frequency of postal mailboxes, and the inspection of the contents during opening. Among them, operation detection includes the compliance detection of the operation of the security inspection machine, operation out-of-bounds detection, and the detection of goods falling off the conveyor belt. Among them, behavior detection includes the detection of violent sorting behavior and the detection of illegal behavior in the operation of the conveyor belt.

[0060] S102. Collect the picture data of the relevant scenes in the postal industry, use the Label Me annotation tool to perform bounding box selection and specify category labels, select the LSTM unit to build a recurrent neural network model, output the predicted classification results of the picture data with category labels, and construct an autoencoder model to integrate the predicted classification results of the picture data;

[0061] Furthermore, connect to the picture database of the relevant scenes in the postal industry to collect the picture data of the relevant scenes in the postal industry and send it back to the server, store it in the form of a file, and use the Label Me annotation tool to perform bounding box selection and specify category labels for each object that appears in the picture data of the relevant scenes in the postal industry, including clothes, documents, shoes, food gift boxes, fruits, jewelry, skin care products, tea, data, mobile phones, computers, wines, medicines, meats, and the unopened situation.

[0062] Furthermore, the entire data is divided into multiple data folders according to the category labels of the image data. Each data folder corresponds to a category label. The number of image data in each data folder corresponding to a category label is greater than 50, the total amount of image data is limited within 100,000, the number of targets of the category label in a single image data cannot exceed 1,000, the size of the image data is limited within 14M, the aspect ratio of the image data is within 3:1, and the longest side is less than 4096px and the shortest side is greater than 30px.

[0063] Furthermore, an LSTM unit is selected to build a recurrent neural network model. According to the image data with category labels input at the current timestamp and the hidden state at the previous timestamp, the activation value of the input layer is calculated through the Sigmoid activation function. The specific formula is:

[0064] T l = S(w l *[I t-1 ,x t +b l )

[0065] where T l represents the activation value of the input layer, S() represents the Sigmoid activation function, w l represents the weight matrix, I t-1 represents the hidden state at the previous timestamp, x t represents the image data with category labels input at the current timestamp, and b l represents the bias term. The output layer selects a fully connected layer and calculates the activation value of the output layer through the Sigmoid activation function according to the image data with category labels input at the current timestamp, the hidden state at the previous timestamp, and the cell state of the image data with category labels input at the current timestamp. The specific formula is:

[0066] T o = S(w o *[I t-1 ,x t +b o )

[0067] where T o represents the activation value of the forgetting layer, S() represents the Sigmoid activation function, w o represents the weight matrix, I t-1 represents the hidden state at the previous timestamp, x t represents the image data with category labels input at the current timestamp, and b oIt represents the bias term. The predicted classification result of the picture data with category labels is output through a fully connected layer. An autoencoder model is constructed to input the picture data with category labels, predict the classification result, and map it to the latent space. The autoencoder is supervised and trained using the predicted classification result of the picture data with category labels. For the picture data without category labels, the reconstruction loss is obtained by measuring the difference between the output reconstructed by the decoder and the picture data without category labels. The predicted classification results of the picture data with category labels and the picture data without category labels are integrated to help train the recurrent neural network model and the autoencoder model.

[0068] S103. Collect the packet image data of the relevant scenes of passing-through security inspection, extract the features of the packet image data using the color histogram, separate the packet image data with different labels using the SVM classification algorithm, and determine the label type of the packet image data based on the positive and negative class relationships of the predicted results of the packet image data.

[0069] Furthermore, use a camera around the security inspection machine to collect the packet image data of the relevant scenes of passing-through security inspection and send it back to the server for storage in the form of a file. Use the LabelMe annotation tool to box and assign different labels to the packet image data of the relevant scenes of passing-through security inspection, and extract the features of the packet image data using the color histogram.

[0070] Furthermore, use the SVM classification algorithm to separate the packet image data with different labels and use the kernel function to map the features of the packet image data with different labels to a high-dimensional space to construct an optimal hyperplane, so as to better separate the packet image data with different labels in the high-dimensional space and maximize the projection points of the separated packet image data on this hyperplane. The basic formula form of the SVM is:

[0071] f(x) = sign(w * x + b)

[0072] where f(x) represents the predicted result of the packet image data, w represents the weight vector, which is used to represent the direction of the hyperplane, x represents the feature vector of the packet image data with different labels, b represents the bias term, which is used to adjust the distance between the hyperplane and the origin. The label type of the packet image data is determined by the positive and negative class relationships of f(x). When f(x) < 0, it is predicted as the positive class; when f(x) > 0, it is predicted as the negative class. The positive and negative class signals are fed back to the intelligent security inspection alarm information database to automatically generate intelligent security inspection alarm information. Record the abnormal express packages found during the security inspection, report them to the postal management department, and report to the relevant departments according to the situation. The abnormal security inspection alarm information includes: the time of the problem, the place where it occurred, the type of the problem, the waybill number, the package picture, and the waybill picture.

[0073] S104. Provide a management function for early warning rules and support the management functions of adding, modifying, and deleting multiple types of business rules. Use historical data to verify the rule matching degree and submit formal rules as the basis for automatic early warning according to the rules;

[0074] Further, provide a management function for early warning rules and support the management functions of adding, modifying, and deleting multiple types of business rules. After users with permissions complete the rule settings, use historical data to verify the rule matching degree and submit formal rules as the basis for automatic early warning according to the rules. Continuously improve the early warning management function according to the actual situation. Analyze and calculate the data change trajectories of the historical operations and the number of couriers at each network point through historical data analysis, and depict the characteristics of the historical business changes and the number of couriers in the daily operations of the network points. Compare with the current status of the network points in real time, mark the current operation status of each network point, and give early warnings in a timely manner for the network points that are about to experience warehouse congestion, existing, and backlogged. Automatically generate and comprehensively set the classification and grading of warehouse congestion, existing, and backlogged detections in combination with manual assistance. According to different event types, manually set the event types in advance and set the judgment rules.

[0075] It should be noted that in the above embodiments, the descriptions of each embodiment have their own focuses. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0076] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0077] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram can be realized by computer program instructions, and the combination of the processes and / or blocks in the flowchart and / or block diagram can also be realized by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded computers, or other programmable data processing devices to generate a machine, so that the instructions executed by the processors of the computer or other programmable data processing devices generate means for realizing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0078] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to work in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction device that implements the functions specified in one or more processes and / or blocks Figure 1 in the process Figure 1 or processes and / or blocks

[0079] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, such that a series of operational steps are performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one or more processes and / or blocks Figure 1 in the process Figure 1 or processes and / or blocks

[0080] Although the preferred embodiments of the present invention have been described, additional changes and modifications can be made by those skilled in the art once they learn of the basic inventive concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the present invention

[0081] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations

Claims

1. A comprehensive safety supervision system for the postal industry, characterized in that: Specifically include: It includes video security inspection and supervision module, mail and mail inspection and supervision module, machine security inspection and supervision module, and business pressure risk warning module; Video security inspection and supervision module: Use cameras to collect video data of postal-related scenes and use VOTT annotation tools to annotate regions, build a feature extraction network structure based on convolution and pooling, and combine it with a target detection network with residual blocks; Input video data of postal industry-related scenes, add a convolution layer to extract the basic features of the video data through multiple convolution kernels, use ReLU as the activation function layer to introduce nonlinearity and use the convolution kernel to perform convolution operations on the video data, and downsample the basic features of the video data in the convolution layer, repeat the first convolution and pooling operations, flatten the high-level features of the video data into a one-dimensional vector through the fully connected layer of the classification head and the regression head, and input it into the fully connected layer, use the fully connected layer to map the high-level features of the video data to different scene categories of the video data through the weight matrix, and output the final scene category recognition result; Supervision module of the mail inspection system: connect to the image database of postal industry-related scenes to collect image data of postal industry-related scenes, use the LabelMe annotation tool to select and specify category labels, select LSTM units to build a recurrent neural network model, and build an autoencoder model; The autoencoder model is used to input image data with category labels to predict the classification results and map them to the latent space. The autoencoder is supervised and trained by predicting the classification results with image data with category labels. Machine security inspection system supervision module: Use cameras around the security inspection equipment to collect package image data of machine security inspection related scenes, use the SVM classification algorithm to separate the package image data with different labels, and automatically generate intelligent security inspection alarm information; Business pressure risk warning module: provides warning rule management functions and supports the addition, modification and deletion of multiple types of business rules. It uses historical data to verify the degree of rule matching and submits formal rules as the basis for automatic warning according to the rules.

2. A postal industry comprehensive safety supervision system according to claim 1, characterized in that: The specific formula of the convolution operation is: Among them, A(i,j) represents an element of the basic feature of the video data, I(m,n) represents the basic feature pixel value of the video data, K(im,jn) represents the basic feature weight value of the video data, m and n represent the row and column indexes inside the convolution kernel respectively, and i and j represent the row and column indexes of the basic feature of the video data respectively.

3. A postal industry comprehensive safety supervision system according to claim 1, characterized in that: The collection and inspection system supervision module is connected to the image database of postal industry-related scenes to collect image data of postal industry-related scenes, uses the LabelMe annotation tool to select each object appearing in the image data of postal industry-related scenes and assign category labels, selects the LSTM unit to build a recurrent neural network model, inputs the image data with category labels and the hidden state of the previous timestamp according to the current timestamp, calculates the activation value of the input layer through the Sigmoid activation function, selects the fully connected layer as the output layer, and inputs the image data with category labels according to the current timestamp, the hidden state of the previous timestamp, and the cell state of the image data with category labels at the current timestamp, calculates the activation value of the output layer through the Sigmoid activation function, and outputs the image data with category labels through the fully connected layer to predict the classification result.

4. A postal industry comprehensive safety supervision system according to claim 3, characterized in that: The specific formula for calculating the activation value of the input layer is: T l =S(w l *[I t-1 ,x t ]+b l ) Among them, T l represents the activation value of the input layer, S() represents the Sigmoid activation function, and w l represents the weight matrix, I t-1 represents the hidden state of the previous timestamp, x t Indicates the image data with category label input at the current timestamp, b l Represents the bias term.

5. A postal industry comprehensive safety supervision system according to claim 3, characterized in that: The specific formula for calculating the activation value of the output layer is: T o =S(w o *[I t-1 ,x t ]+b o ) Among them, T o represents the activation value of the forgetting layer, S() represents the Sigmoid activation function, and w o represents the weight matrix, I t-1 represents the hidden state of the previous timestamp, x t Indicates the image data with category label input at the current timestamp, b o Represents the bias term.

6. A postal industry comprehensive safety supervision system according to claim 1, characterized in that: The machine security inspection system supervision module uses cameras around the security inspection machine equipment to collect packaged image data of machine security inspection related scenes, uses the LabelMe annotation tool to select the packaged image data of machine security inspection related scenes and assign different labels, uses the SVM classification algorithm to separate the packaged image data with different labels and uses the kernel function to map the features of the packaged image data with different labels to a high-dimensional space to construct an optimal hyperplane, so as to better separate the packaged image data with different labels in the high-dimensional space and maximize the projection point of the separated packaged image data on the hyperplane.

7. A postal industry comprehensive safety supervision system according to claim 6, characterized in that: The basic formula form of the SVM is: f(x)=sign(w*x+b) Where f(x) represents the prediction result of the packet image data, w represents the weight vector used to represent the direction of the hyperplane, x represents the feature vector of the packet image data with different labels, and b represents the bias term.

8. A postal industry comprehensive safety supervision system according to claim 1, characterized in that: The business pressure risk warning module provides the management function of warning rules and supports the management functions of adding, modifying and deleting multiple types of business rules. It continuously improves the warning management function according to the actual situation, calculates the data change trajectory of the historical business and the number of couriers of each outlet through historical data analysis, and describes the characteristics of the historical business changes and the number of couriers of the daily operation of the outlets, compares them with the current status of the outlets in real time, marks the current operation status of each outlet, automatically generates and combines manual assistance to comprehensively set the classification and grading of warehouse explosion, existing and backlog detection, and manually sets the event type in advance according to the different event types, and sets the judgment rules.

9. A postal industry comprehensive safety supervision method, applied to a postal industry comprehensive safety supervision system as claimed in any one of claims 1 to 8, characterized in that: The specific steps include: S101. Collect video data of postal-related scenes and use VOTT annotation tools to annotate regions, build a feature extraction network structure based on convolution and pooling and combine it with a target detection network of residual blocks, and output the final scene category recognition results; S102. Collect image data of postal industry-related scenes, use LabelMe annotation tool to select and specify category labels, select LSTM unit to build recurrent neural network model, output image data prediction and classification results with category labels, and build autoencoder model to integrate image data prediction and classification results; S103. Collecting package image data of scenes related to machine security inspection, extracting features of package image data using color histogram, separating package image data with different labels using SVM classification algorithm, and determining the label type of package image data by the positive and negative class relationship of the prediction results of package image data; S104. Provides management functions for early warning rules and supports the addition, modification, and deletion of multiple types of business rules. It uses historical data to verify the degree of rule matching and submits formal rules as the basis for automatic early warning according to the rules.

Images