A method and system for analyzing tobacco use based on visual information

Abstract

This invention relates to a method and system for tobacco-related analysis based on visual information. The method involves converting and classifying the collected tobacco-related data, and then detecting and identifying tobacco types in tobacco-related images, extracting tobacco-related license plates from tobacco-related license plate images, and identifying and extracting tobacco-related text from tobacco-related text images. This, combined with personnel information and investigation data, enables rapid tobacco-related analysis of the collected data, helping to quickly identify key personnel.

Description

Technical Field

[0001] This invention relates to the field of smoke monitoring technology, and more specifically, to a method and system for smoke-related analysis based on visual information. Background Technology

[0002] Currently, there is no case analysis and judgment system in China specifically for tobacco-related violations that can quickly and accurately analyze tobacco-related violations from massive amounts of data, helping to quickly identify key personnel and uncover clues. Summary of the Invention

[0003] The problem addressed by this invention is how to quickly and accurately analyze and judge cases involving tobacco-related violations.

[0004] To address the above problems, this invention provides a method for analyzing tobacco use based on visual information, comprising:

[0005] Step 1: Collect tobacco-related data and store it in the database as a raw sample data package. Then, determine the data type of the tobacco-related data. If the tobacco-related data is in a preset format, proceed to Step 2. If the tobacco-related data is in a format other than the preset format, convert the format to the preset format type and then proceed to Step 2.

[0006] Step 2: Classify and store the tobacco-related data that conforms to the preset format type, including personnel information data that stores personnel identity information, multimedia data that stores on-site collected images containing tobacco-related images, license plate images of tobacco-related vehicles, and text images of tobacco-related materials, and investigation data that stores data from operators and banks.

[0007] Step 3: Detect and recognize smoke-related images, smoke-related license plate images, and smoke-related text images in the multimedia data category respectively;

[0008] Step 4: Based on the detection and identification of multimedia data, conduct tobacco-related analysis by combining personnel information data and investigation data, output the analysis results, and visualize the output.

[0009] The beneficial effects of the method of the present invention are as follows: after converting the collected tobacco-related data into a format, it is classified and stored. Then, by detecting and identifying tobacco categories in tobacco-related images, identifying and extracting tobacco-related license plates in tobacco-related license plate images, and identifying and extracting tobacco-related text in tobacco-related text images, combined with personnel information and investigation data, the collected tobacco-related data can be quickly analyzed and judged, which helps to quickly identify key personnel.

[0010] Preferably, step 3, for tobacco-related images in the multimedia data category, specifically employs a YOLOX-CR network constructed by adding a visual attention mechanism to the YOLOX network. This network is used to identify and detect tobacco products in the tobacco-related images, including:

[0011] Step 301A: Scale the smoke-related images;

[0012] Step 302A: Perform X-axis attention calculation and Y-axis attention calculation on the scaled smoke image to obtain the coordinate attention of the smoke image;

[0013] Step 303A: The smoke-affected image with coordinate attention is subjected to the first DA feature extraction to obtain the first layer feature map, the second DA feature extraction to obtain the second layer feature map, and the third DA feature extraction to obtain the third layer feature map.

[0014] Step 304A: Based on PANet, feature maps of the first layer, second layer, and third layer are fused to transmit semantic information at different levels, resulting in small-sized feature maps, medium-sized feature maps, and large-sized feature maps.

[0015] Step 305A: Decode the small-size feature map, medium-size feature map, and large-size feature map respectively to obtain the location coordinates and category confidence of the detected target;

[0016] Step 306A: Based on the location coordinates and category confidence of the detected target, mark the location coordinates and category confidence of the detected target on the original smoke-affected image, output the calibrated smoke-affected image, and store it;

[0017] By incorporating a visual attention mechanism into the YOLOX network, it is possible to extract more discriminative features of targets in smoke-related images.

[0018] Preferably, step 3, which involves recognizing and detecting license plate images of vehicles involved in smoking, specifically includes:

[0019] Step 301B: Input the image of the license plate of the vehicle involved in smoking into a multi-task cascaded convolutional neural network to detect the image of the license plate region involved in smoking;

[0020] Step 302B: Input the image of the license plate area of ​​the vehicle involved in smoking into the STN spatial transformation network, solve the transformation parameters through LocalisationNet, find the position mapping matrix between the output and input features through Gridgenerator based on the transformation parameters and the defined transformation method, and select and output the transformed license plate area image of the vehicle involved in smoking by combining the position mapping matrix and the transformation parameters with bilinear interpolation.

[0021] Step 303B: Use LPRNet to identify and extract license plates involved in smoking from the area map of license plates involved in smoking, extract the license plate data information involved in smoking, and store it;

[0022] A multi-task cascaded convolutional neural network can accurately locate the license plate area of ​​vehicles involved in smoking in images with a lot of data noise. Then, STN spatial transformation network is used to overcome the problems of poor data quality such as image tilt and image distortion. Finally, LPRNet is used to quickly identify and extract the data information of vehicles involved in smoking from the license plate area image.

[0023] Preferably, in step 3, the detection and recognition of tobacco-related text images specifically involves using the DBNet network to identify and extract tobacco-related text information from the tobacco-related text images and storing it.

[0024] Preferably, step 4, based on the detection and identification of multimedia data, combines personnel information data and investigation data data for tobacco-related analysis and judgment, specifically including:

[0025] Step 401) Determine whether there are tobacco-related images in the multimedia data category. If there are, identify and detect the tobacco products in the tobacco-related images according to steps 301A to 305A. Combine the personnel information category and the operator assistance data in the assistance data category to determine whether the personnel have production, sales and operation licenses for the tobacco products in the tobacco-related images, and proceed to step 402. If there are no tobacco products, proceed to step 402.

[0026] Step 402) Determine whether there is a license plate image for tobacco-related vehicles in the multimedia data category. If there is, identify the license plate image for tobacco-related vehicles and extract the license plate data information according to steps 301B to 303B. Combine the operator investigation data in the investigation data category to determine whether the extracted license plate for tobacco-related vehicles has the transportation qualification for the corresponding tobacco category, and proceed to step 403. If there is no such vehicle, proceed to step 403.

[0027] Step 403: Determine whether there are tobacco-related text images in the multimedia data class. If so, use the DBNet network to identify and extract tobacco-related text information from the tobacco-related text images. Combine this with personnel information class and investigation data class to complete the perception, verification and tracking of personnel's economic background, financial transactions and other activities and abnormal economic behaviors, thereby realizing the analysis of tobacco-related cases.

[0028] Step 404: Output the analysis results from steps 401 to 403.

[0029] As a preferred option, it also includes:

[0030] Step 5: Construct a tobacco-related relationship map based on the personnel information and assistance data stored in Step 2 and the detection and identification results of multimedia data in Step 3, to help quickly identify key personnel.

[0031] A visual information-based tobacco use analysis system includes:

[0032] The database module is used to store raw sample data and classify and store tobacco-related data that conforms to preset format types;

[0033] The data conversion module is used to determine the data type of tobacco-related data and transmit tobacco-related data that conforms to the preset format to the database module for classification and storage. It is also used to convert tobacco-related data that does not conform to the preset format into tobacco-related data that conforms to the preset format and transmit it to the database module for classification and storage.

[0034] The tobacco-related case analysis module includes a tobacco-related image analysis unit for detecting and recognizing tobacco-related images in multimedia data and analyzing them in conjunction with personnel information and investigation data; a tobacco-related license plate analysis unit for recognizing and extracting tobacco-related license plate images and analyzing them in conjunction with personnel information and investigation data; and a tobacco-related text analysis unit for detecting and recognizing tobacco-related text images and analyzing them in conjunction with personnel information and investigation data, and performing tobacco-related analysis and judgment based on personnel information and investigation data.

[0035] The tobacco-related knowledge graph module is used to construct a tobacco-related relationship graph based on the detection and identification results of personnel information, investigation data, and multimedia data.

[0036] The analysis results output module is used to output and display the analysis results of tobacco-related cases or to display a tobacco-related relationship graph.

[0037] The beneficial effects of the system of this invention are: it provides an integrated platform for target detection and identification algorithms for tobacco-related data, and by starting with tobacco-related big data, it can quickly identify key personnel, dig out clues, and accurately carry out crackdowns from massive amounts of big data, thus promoting the transformation and upgrading of tobacco anti-counterfeiting work in the era of big data. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating the conversion and categorized storage of tobacco-related data in a specific embodiment 1 of the present invention;

[0039] Figure 2 This is a flowchart of the YOLOX-CR network technology in specific embodiment 1 of the present invention;

[0040] Figure 3 This is a flowchart illustrating the technical process of a multi-task cascaded convolutional neural network, a specific embodiment of the present invention.

[0041] Figure 4 This is a technical flow diagram of the DBNet network in a specific embodiment of the present invention.

[0042] Figure 5 This is a system block diagram of a specific embodiment 2 of the present invention. Detailed Implementation

[0043] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Specific Implementation Example 1

[0045] A method for analyzing tobacco use based on visual information includes:

[0046] Step 1: Collect tobacco-related data. In this embodiment, the tobacco-related data includes on-site collected personnel image data, identity data, mobile phone information, tobacco-related account information, tobacco-related vehicle images and videos, tobacco-related images and videos, postal and logistics information, and tobacco-related information in special characters, etc., which are stored in the database as raw sample data packages; then, as... Figure 1 As shown, the data type of the tobacco-related data is determined. In this embodiment, the data type determination specifically refers to the format of the tobacco-related data. If the tobacco-related data is a preset format type, then proceed to step 2. If the tobacco-related data is in a format other than the preset format, then convert the format to the preset format type and then proceed to step 2. Here, the method for converting tobacco-related data of other format types to the preset format type is existing technology and will not be described in detail here.

[0047] Step 2: Classify and store the tobacco-related data that conforms to the preset format type, including personnel information data that stores personnel identity information, multimedia data that stores on-site collected images containing tobacco-related images, license plate images of tobacco-related vehicles, and text images of tobacco-related materials, and investigation data that stores data from operators and banks.

[0048] Step 3: Detect and recognize smoke-related images, smoke-related license plate images, and smoke-related text images in the multimedia data category; among which,

[0049] For tobacco-related images in multimedia data, a YOLOX-CR network is constructed by adding a visual attention mechanism to the YOLOX network. This network is used to identify and detect tobacco products within the tobacco-related images, including:

[0050] Step 301A: Scale the smoke-related images;

[0051] Step 302A: Perform X-axis attention calculation and Y-axis attention calculation on the scaled smoke image to obtain the coordinate attention of the smoke image;

[0052] Step 303A: The smoke-affected image with coordinate attention is subjected to the first DA feature extraction to obtain a large-size feature map, the second DA feature extraction to obtain a medium-size feature map, and the third DA feature extraction to obtain a small-size feature map.

[0053] Step 304A: Based on PANet, feature fusion is performed on small-sized feature maps, medium-sized feature maps, and large-sized feature maps to transmit semantic information at different levels, resulting in three feature decoders corresponding to the corresponding sizes;

[0054] Step 305A: The three feature decoding inputs corresponding to the respective sizes are used to perform feature decoding on the decoupler network to obtain the location coordinates and category confidence of the detected target;

[0055] Step 306A: Based on the location coordinates and category confidence of the detected target, mark the location coordinates and category confidence of the detected target on the original smoke-affected image, output the calibrated smoke-affected image, and store it;

[0056] By adding a visual attention mechanism to the YOLOX network, it is possible to extract target discriminative features from smoke-related images more effectively.

[0057] Among them, the constructed YOLOX-CR network is as follows Figure 2 As shown, it includes:

[0058] The smoke-related feature extraction unit includes a focusing layer for scaling the smoke-related image, a standard convolutional SiLU activation layer, a DA coordinate attention layer for calculating X-axis and Y-axis attention of the smoke-related image, and an SPP multi-scale pooling layer. After standard focusing of the input smoke-related image, a lossless 12-channel feature map is obtained. This is then subjected to two standard convolutional SiLU activation layer operations, followed by DA coordinate attention calculation to obtain a large-scale feature map. Based on the large-scale feature map, a standard SiLU activation layer and DA coordinate attention layer operation is performed to obtain a medium-scale feature map. Finally, based on the medium-scale feature map, a standard SiLU activation layer, DA coordinate attention layer, and SPP multi-scale pooling operation are performed again to obtain a small-scale feature map. These three feature maps of different sizes store semantic information at different levels of the image.

[0059] The multi-scale fusion unit for tobacco-related features includes concatenation, upsampling, downsampling, a DA coordinate attention layer, and a standard convolutional SiLU layer. PANet upsamples the small-scale feature maps obtained from the feature extraction unit, concatenates them with the medium-sized feature maps, and then recalculates the attention using the DA coordinate attention layer. The recalculated attention-based feature maps are then upsampled again, concatenated with the large-sized feature maps, and recalculated again using the DA coordinate attention layer to obtain feature decoding one. Based on feature decoding one, downsampling is performed, concatenating it with the medium-sized feature maps, and then recalculating the attention using the DA coordinate attention layer to obtain feature decoding two. Based on feature decoding two, downsampling is performed, concatenating it with the small-sized feature maps, and then recalculating the attention using the DA coordinate attention layer to obtain feature decoding three. The feature maps in the feature fusion module undergo bidirectional fusion from bottom to top and top to bottom, resulting in rich semantic information.

[0060] The tobacco-related feature prediction head unit includes a first decoupling layer, a second decoupling layer, and a third decoupling layer that perform feature decoding on small-sized, medium-sized, and large-sized feature maps, respectively, to obtain the bounding box coordinates and category confidence of the predicted target;

[0061] The identification and detection of license plate images of vehicles involved in smoking specifically includes:

[0062] Step 301B: Input the image of the license plate of the vehicle involved in smoking into a multi-task cascaded convolutional neural network to detect the license plate region image of the vehicle involved in smoking. Specifically, this includes: first, scaling the image n-1 times according to a set scaling factor to obtain an image pyramid formed by stacking n license plate images of different sizes; then, inputting the license plate images of different sizes in the image pyramid into a PNet network for filtering to obtain a series of candidate license plate region boxes; then, inputting the license plate image of the vehicle involved in smoking and a series of candidate license plate region boxes into an RNet network for further filtering, and fine-tuning the coordinate positions of the filtered candidate license plate region boxes; finally, inputting all candidate license plate region boxes output by the RNet network into an ONet network to output a license plate region image of the vehicle involved in smoking containing the license plate and license plate key points. Here, the multi-task cascaded convolutional neural network is existing technology and will not be described in detail. The principle diagram is shown below. Figure 3 As shown.

[0063] Step 302B: Input the image of the license plate area of ​​the vehicle involved in smoking into the STN spatial transformation network. Solve the transformation parameters through LocalisationNet. Use Gridgenerator to find the position mapping matrix between the output and input features based on the transformation parameters and the defined transformation method. Sample combines the position mapping matrix and transformation parameters to select and output the transformed license plate area image of the vehicle involved in smoking using bilinear interpolation. Here, the processing flow of the STN spatial transformation network is existing technology and will not be described in detail.

[0064] Step 303B: Use LPRNet to identify and extract license plates involved in smoking from the area map of license plates involved in smoking, extract the license plate data information involved in smoking, and store it;

[0065] A multi-task cascaded convolutional neural network can accurately locate the license plate region of vehicles involved in smoking in images with a lot of data noise. Then, STN spatial transformation network is used to overcome the problems of poor data quality such as image tilt and image distortion. Finally, LPRNet is used to quickly identify and extract data information of vehicles involved in smoking from the license plate region image.

[0066] The detection and recognition of tobacco-related text images specifically employs the DBNet network to identify and extract tobacco-related text information from the images, and stores it, such as... Figure 4 The image of the tobacco-related text shown is of a Golden Leaf cigarette box. The DBNet network can detect and extract the tobacco-related text region. DBNet network is existing technology and will not be described in detail here.

[0067] Step 4: Based on the detection and identification of multimedia data, conduct tobacco-related analysis and judgment by combining personnel information data and assistance data, output the analysis results, and provide a visual output, specifically including:

[0068] Step 401) Determine whether there are tobacco-related images in the multimedia data category. If there are, identify and detect the tobacco products in the tobacco-related images according to steps 301A to 305A. Combine the personnel information category and the operator assistance data in the assistance data category to determine whether the personnel have production, sales and operation licenses for the tobacco products in the tobacco-related images, and proceed to step 402. If there are no tobacco products, proceed to step 402.

[0069] Step 402) Determine whether there is a license plate image for tobacco-related vehicles in the multimedia data category. If there is, identify the license plate image for tobacco-related vehicles and extract the license plate data information according to steps 301B to 303B. Combine the operator investigation data in the investigation data category to determine whether the extracted license plate for tobacco-related vehicles has the transportation qualification for the corresponding tobacco category, and proceed to step 403. If there is no such vehicle, proceed to step 403.

[0070] Step 403: Determine whether there are tobacco-related text images in the multimedia data class. If so, use the DBNet network to identify and extract tobacco-related text information from the tobacco-related text images. Combine this with personnel information class and investigation data class to complete the perception, verification and tracking of personnel's economic background, financial transactions and other activities and abnormal economic behaviors, thereby realizing the analysis of tobacco-related cases.

[0071] Step 404: Output the analysis results from steps 401 to 403;

[0072] Furthermore, based on the personnel information and assistance data stored in step 2, and the detection and identification results of multimedia data in step 3, a tobacco-related relationship map is constructed to help quickly identify key personnel and uncover clues. Specific Implementation Example 2

[0074] A visual information-based tobacco control analysis system, such as Figure 5 As shown, it includes:

[0075] The database module is used to store raw sample data and classify and store tobacco-related data that conforms to preset format types;

[0076] The data conversion module is used to determine the data type of tobacco-related data and transmit tobacco-related data that conforms to the preset format to the database module for classification and storage. It is also used to convert tobacco-related data that does not conform to the preset format into tobacco-related data that conforms to the preset format and transmit it to the database module for classification and storage.

[0077] The tobacco-related case analysis module includes a tobacco-related image analysis unit for detecting and recognizing tobacco-related images in multimedia data and analyzing them in conjunction with personnel information and investigation data; a tobacco-related license plate analysis unit for recognizing and extracting tobacco-related license plate images and analyzing them in conjunction with personnel information and investigation data; and a tobacco-related text analysis unit for detecting and recognizing tobacco-related text images and analyzing them in conjunction with personnel information and investigation data, and performing tobacco-related analysis and judgment based on personnel information and investigation data.

[0078] The tobacco-related knowledge graph module is used to construct a tobacco-related relationship graph based on the detection and identification results of personnel information, investigation data, and multimedia data.

[0079] The analysis results output module is used to output and display the analysis results of tobacco-related cases or to display a tobacco-related relationship graph.

[0080] In this embodiment of the tobacco-related analysis system based on visual information, the tobacco-related knowledge graph module and the analysis result output module are used to output the established relevant knowledge graph. The tobacco-related knowledge graph module includes a knowledge graph model management unit and a knowledge graph data crawling unit. The knowledge graph data crawling unit is used to crawl and generate relevant data information of knowledge graph keywords from the database module, and generate a knowledge graph for display. The knowledge graph management unit is used to manage the generated knowledge graph.

[0081] While the disclosure is as stated above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of this disclosure, and all such changes and modifications will fall within the protection scope of this invention.

Claims

1. A method for analyzing smoke-related incidents based on visual information, characterized in that, include: Step 1: Collect tobacco-related data and store it in the database as a raw sample data package. Then, determine the data type of the tobacco-related data. If the tobacco-related data is in a preset format, proceed to Step 2. If the tobacco-related data is in a format other than the preset format, convert the format to the preset format type and then proceed to Step 2. Step 2: Classify and store the tobacco-related data that conforms to the preset format type, including personnel information data that stores personnel identity information, multimedia data that stores on-site collected images containing tobacco-related images, license plate images of tobacco-related vehicles, and text images of tobacco-related materials, and investigation data that stores data from operators and banks. Step 3: Detect and recognize smoke-related images, smoke-related license plate images, and smoke-related text images in the multimedia data category respectively; For tobacco-related images in multimedia data, a YOLOX-CR network is constructed by adding a visual attention mechanism to the YOLOX network. This network is used to identify and detect tobacco products in the tobacco-related images, including: Step 301A: Scale the smoke-related images; Step 302A: Perform X-axis attention calculation and Y-axis attention calculation on the scaled smoke image to obtain the coordinate attention of the smoke image; Step 303A: The smoke-affected image with coordinate attention is subjected to the first DA feature extraction to obtain the first layer feature map, the second DA feature extraction to obtain the second layer feature map, and the third DA feature extraction to obtain the third layer feature map. Step 304A: Based on PANet, feature maps of the first layer, second layer, and third layer are fused to transmit semantic information at different levels, resulting in small-sized feature maps, medium-sized feature maps, and large-sized feature maps. Step 305A: Decode the small-size feature map, medium-size feature map, and large-size feature map respectively to obtain the location coordinates and category confidence of the detected target; Step 306A: Based on the location coordinates and category confidence of the detected target, mark the location coordinates and category confidence of the detected target on the original smoke image, output the marked smoke image, and store it; The identification and detection of license plate images of vehicles involved in smoking specifically includes: Step 301B: Input the image of the license plate of the vehicle involved in smoking into a multi-task cascaded convolutional neural network to detect the image of the license plate region involved in smoking; Step 302B: Input the image of the license plate area of ​​the vehicle involved in smoking into the STN spatial transformation network, solve the transformation parameters through the Localisation net, find the position mapping matrix between the output and input features through the grid generator in the transformation parameters and the defined transformation method, and select and output the transformed license plate area image of the vehicle involved in smoking by combining the position mapping matrix and the transformation parameters and combining bilinear interpolation. Step 303B: Use LPRNet to identify and extract license plates involved in smoking from the area map of license plates involved in smoking, extract the license plate data information involved in smoking, and store it; The detection and recognition of tobacco-related text images specifically involves using the DBNet network to identify and extract tobacco-related text information from the images and then storing it. Step 4: Based on the detection and identification of multimedia data, conduct tobacco-related analysis in conjunction with personnel information data and investigation data, output the analysis results, and visualize the output.

2. The method for analyzing smoke-related issues based on visual information according to claim 1, characterized in that, Step 4, based on the detection and identification of multimedia data, combines personnel information data and investigation data data to conduct tobacco-related analysis and judgment, specifically including: Step 401: Determine whether there are tobacco-related images in the multimedia data category. If so, identify and detect the tobacco products in the tobacco-related images according to steps 301A to 305A. Combine the personnel information category and the operator assistance data in the assistance data category to determine whether the personnel have production, sales and operation licenses for the tobacco products in the tobacco-related images, and proceed to step 402. If not, proceed to step 402. Step 402: Determine whether there is a license plate image for tobacco-related vehicles in the multimedia data category. If there is, identify the license plate image for tobacco-related vehicles and extract the license plate data information according to steps 301B to 303B. Combine the operator investigation data in the investigation data category to determine whether the extracted license plate for tobacco-related vehicles has the transportation qualification for the corresponding tobacco category, and proceed to step 403. If there is no such vehicle, proceed to step 403. Step 403: Determine whether there are tobacco-related text images in the multimedia data class. If so, use the DBNet network to identify and extract tobacco-related text information from the tobacco-related text images. Combine this with personnel information class and investigation data class to complete the perception, verification and tracking of personnel's economic background, financial transaction activities and abnormal economic behavior, thereby realizing the analysis of tobacco-related cases. Step 404: Output the analysis results from steps 401 to 403.

3. The method for analyzing smoke-related issues based on visual information according to claim 1, characterized in that, Also includes: Step 5: Construct a tobacco-related relationship map based on the personnel information and investigation data stored in Step 2 and the detection and identification results of multimedia data in Step 3.

4. A smoke-related information analysis system based on visual information, implementing the smoke-related information analysis method based on visual information as described in claim 1, characterized in that, include: The database module is used to store raw sample data and classify and store tobacco-related data that conforms to preset format types; The data conversion module is used to determine the data type of tobacco-related data and transmit tobacco-related data that conforms to the preset format to the database module for classification and storage. It is also used to convert tobacco-related data that does not conform to the preset format into tobacco-related data that conforms to the preset format and transmit it to the database module for classification and storage. The tobacco-related case analysis module includes a tobacco-related image analysis unit for detecting and recognizing tobacco-related images in multimedia data and analyzing them in conjunction with personnel information and investigation data; a tobacco-related license plate analysis unit for recognizing and extracting tobacco-related license plate images and analyzing them in conjunction with personnel information and investigation data; and a tobacco-related text analysis unit for detecting and recognizing tobacco-related text images and analyzing them in conjunction with personnel information and investigation data, and performing tobacco-related analysis in conjunction with personnel information and investigation data. The tobacco-related knowledge graph module is used to construct a tobacco-related relationship graph based on the detection and identification results of personnel information, investigation data, and multimedia data. The analysis results output module is used to output and display the analysis results of tobacco-related cases or to display a tobacco-related relationship graph.

Images