Security detection method and system based on artificial intelligence

Abstract

The invention relates to the technical field of data analysis, in particular to a security detection method and system based on artificial intelligence, and the method comprises the steps: S1, constructing a security check model, and training the constructed security check model; s2, a security check picture is obtained, and a security check model is adopted to carry out hidden danger auxiliary decision making; s3, performing real-time security check picture recognition by adopting the security check model; wherein S2 comprises the steps of obtaining a security check picture, performing hidden danger prediction through a security check model, and generating a prediction result; wherein the prediction result comprises a security inspection object and a security inspection hidden danger type; the prediction result is verified, and if the prediction result is matched with the hidden danger identified manually, the prediction result is transferred into a marketing system; and if the prediction result is not matched with the artificially identified hidden danger, feeding back that the prediction result has a problem, and optimizing the security check model. According to the scheme, safety inspection can be assisted, and the quality and efficiency of safety inspection are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to a security detection method and system based on artificial intelligence. Background Art

[0002] Gas is a general term for gaseous fuels, which can burn and release heat for use by residents and industrial enterprises. There are many types of gas, mainly natural gas, artificial gas, liquefied petroleum gas, biogas, and coal-to-gas. In the process of using gas, in order to ensure the safety of gas users, according to relevant regulations, gas companies will regularly carry out safety inspections of gas pipelines, metering instruments, gas combustion appliances, etc. and inform of hidden dangers, that is, gas safety testing.

[0003] Existing gas safety inspections are mainly carried out by personnel going door-to-door to inspect objects that are prone to malfunction, as well as to detect whether there are any safety hazards, and to take photos of areas that need to be inspected. The entire inspection process mainly relies on the experience of personnel, which is prone to missed inspections, and the inspection quality varies. The administrators who manage the inspection results are under great pressure to inspect the safety inspections one by one, and need to check whether there are any unqualified problems, resulting in low efficiency in safety inspections.

[0004] Therefore, there is an urgent need for an artificial intelligence-based security detection method and system that can assist in security inspections and improve the quality and efficiency of security detection. Summary of the invention

[0005] One of the purposes of the present invention is to provide a security detection method based on artificial intelligence, which can assist in security inspections and improve the quality and efficiency of security detection.

[0006] The present invention provides a basic solution 1: a security detection method based on artificial intelligence, including the following contents:

[0007] S1. Build a security inspection model and train the built security inspection model;

[0008] S2. Obtain security inspection pictures and use security inspection models to assist in decision-making on hidden dangers;

[0009] S3, using the security inspection model to perform real-time security inspection image recognition;

[0010] Among them, S2 includes:

[0011] Obtain security inspection pictures, use security inspection models to predict hidden dangers, and generate prediction results; the prediction results include: security inspection objects and security inspection hidden danger types;

[0012] The prediction results are verified. If the prediction results match the hidden dangers identified manually, the prediction results are transferred to the marketing system. If the prediction results do not match the hidden dangers identified manually, feedback is given that there is a problem with the prediction results and the security inspection model is optimized.

[0013] Further, the S1 includes:

[0014] S101, historical security inspection data capture, where the security inspection data is security inspection pictures;

[0015] S102, marking the security inspection data, and uploading the marked security inspection pictures to the big data platform;

[0016] S103, AI platform reads security inspection pictures from the big data platform for training security inspection models;

[0017] S104: The training results are output to the indicator library to provide data mining services.

[0018] Furthermore, the annotation data is divided into: object annotation and hidden danger type standard.

[0019] Furthermore, the security inspection model adopts a Resnet-18-based transfer learning model, and the specific training method is as follows:

[0020] Prepare the environment: Install the library, provide ResNet-18 pre-trained models and image processing tools;

[0021] Load pre-trained model: Load the pre-trained ResNet-18 model and select the model pre-trained on the ImageNet dataset;

[0022] Adjust the model: adjust the model according to the classification task;

[0023] Data preparation: Prepare a data set of security inspection images, including training sets and test sets;

[0024] Data loader: Create a data loader for batch loading data during training and testing;

[0025] Define loss function and optimizer: Set a loss function and optimizer;

[0026] Training model: Use the training set to iteratively train the model and update the model weights;

[0027] Evaluate the model: Use the test set to test the trained model and evaluate the performance of the model. Usually, accuracy is used as the evaluation indicator.

[0028] Reasoning: Classify new images using a model whose performance meets the preset performance requirements;

[0029] Save and load models: Save models whose performance meets the preset performance requirements and load the models when called.

[0030] Furthermore, the prediction results are pushed via SMS or generated reports.

[0031] Furthermore, S3 includes: uploading security inspection pictures stored on the terminal by employees after security inspection through the terminal and taking and uploading real-time security inspection pictures on site through the terminal, using the security inspection model to perform recognition and prediction on the real-time security inspection pictures, and identifying the security inspection objects and the types of security inspection hazards.

[0032] A second objective of the present invention is to provide a safety detection system based on artificial intelligence that can assist in safety inspections and improve the quality and efficiency of safety detection.

[0033] The present invention provides a second basic solution: a security detection system based on artificial intelligence, comprising: a server and a terminal;

[0034] The server is used to build a security inspection model and train the built security inspection model;

[0035] Terminal, used to obtain security inspection pictures and upload them to the server;

[0036] The server is also used to adopt the security inspection model to assist in decision-making on hidden dangers, identify and predict real-time security inspection images, and identify security inspection objects and types of security inspection hidden dangers;

[0037] The security inspection model is used to assist in hidden danger decision-making, including: using the security inspection model to predict hidden dangers for the security inspection pictures uploaded by the terminal and generate prediction results; the prediction results include: security inspection objects and security inspection hidden danger types;

[0038] The prediction results are verified. If the prediction results match the hidden dangers identified manually, the prediction results are transferred to the marketing system. If the prediction results do not match the hidden dangers identified manually, feedback is given that there is a problem with the prediction results and the security inspection model is optimized.

[0039] Furthermore, the server includes: a data acquisition module, a data processing module, a big data platform, an AI platform, and an indicator library;

[0040] The data acquisition module is used to capture historical security inspection data, where the security inspection data is security inspection pictures;

[0041] The data processing module is used to annotate security inspection data and upload the annotated security inspection images to the big data platform;

[0042] The AI ​​platform reads security inspection images from the big data platform to train the security inspection model;

[0043] The AI ​​platform is also used to output training results to the indicator library to provide data mining services.

[0044] Furthermore, the security inspection model adopts a Resnet-18-based transfer learning model, and the specific training method is as follows:

[0045] Prepare the environment: Install the library, provide ResNet-18 pre-trained models and image processing tools;

[0046] Load pre-trained model: Load the pre-trained ResNet-18 model and select the model pre-trained on the ImageNet dataset;

[0047] Adjust the model: adjust the model according to the classification task;

[0048] Data preparation: Prepare a data set of security inspection images, including training sets and test sets;

[0049] Data loader: Create a data loader for batch loading data during training and testing;

[0050] Define loss function and optimizer: Set a loss function and optimizer;

[0051] Training model: Use the training set to iteratively train the model and update the model weights;

[0052] Evaluate the model: Use the test set to test the trained model and evaluate the performance of the model. Usually, accuracy is used as the evaluation indicator.

[0053] Reasoning: Classify new images using a model whose performance meets the preset performance requirements;

[0054] Save and load models: Save models whose performance meets the preset performance requirements and load the models when called.

[0055] Furthermore, the real-time security inspection pictures are identified and predicted to identify the security inspection objects and the types of security inspection hidden dangers, including:

[0056] Employees upload security inspection pictures stored on the terminal after security inspection, and take pictures on site through the terminal and upload them to the AI ​​platform for real-time security inspection. The security inspection model is used to perform real-time security inspection picture recognition and prediction, and identify security inspection objects and security inspection hazard types.

[0057] Beneficial effects: This solution builds a security inspection model and uses the trained security inspection model to assist in hidden danger decision-making. For the collected security inspection pictures, the security inspection model can be used to predict hidden dangers, generate prediction results, and push them; the prediction results include: security inspection objects and security inspection hidden danger types; this solution verifies the prediction results. If the prediction results match the hidden dangers identified manually, the prediction results are transferred to the marketing system; if the prediction results do not match the hidden dangers identified manually, feedback is given that there are problems with the prediction results, the security inspection model is optimized, and the accuracy of recognition is improved, thereby improving the quality of security detection;

[0058] During the security inspection process, the security inspection model is used to perform real-time security inspection picture recognition. First, through the collected security inspection pictures, it can be verified whether the employees take photos and upload them in accordance with the regulations to ensure the basic collection needs; secondly, the administrator can judge whether there are any missed inspections of employees based on the prediction results; finally, the security inspection model is a recognition model that is trained with a large amount of data. It can accurately identify security inspection objects and security inspection hazard types, improve the accuracy of security inspections, and ensure the quality of security inspections. With the assistance of the security inspection system, managers can also improve efficiency and better complete security inspections. This solution opens up the full-link security monitoring automatic identification from the business end to the detection end. Security inspection pictures are obtained through the daily scheduled tasks of the big data platform and then output to the AI ​​platform. The corresponding results are identified and fed back through the security inspection model.

[0059] In summary, this solution can assist in safety inspections and, as an effective means of supervision, improve the quality and efficiency of safety inspections. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a flowchart of an embodiment of a security detection method based on artificial intelligence of the present invention;

[0061] Figure 2 This is a schematic diagram of a prediction result page in an embodiment of a security detection method based on artificial intelligence of the present invention;

[0062] Figure 3 This is a schematic diagram of a security inspection picture upload page in an embodiment of the artificial intelligence-based security detection method of the present invention. DETAILED DESCRIPTION

[0063] The following is further described in detail through specific implementation methods:

[0064] Embodiment 1

[0065] The embodiment is basically as shown in the attached Figure 1 Shown: Security detection methods based on artificial intelligence, including:

[0066] S1. Construct a security inspection model and train the constructed security inspection model; specifically, it includes:

[0067] S101. Capture historical security inspection data, where the security inspection data is security inspection pictures, specifically including:

[0068] Capture the data of security inspection pictures in the marketing system to the designated picture server. Security inspection pictures are generally stored in the marketing system. The number of security inspection pictures required for security inspection training is very large. Directly using the security inspection pictures in the marketing system will affect the business. Therefore, the pictures are downloaded to the picture server through offline download. The reason why they are not uploaded directly is that the pictures stored in the marketing system exist in the form of links and cannot be used directly. Therefore, they need to be downloaded into binary streams first.

[0069] S102. Mark security inspection data, including:

[0070] The security inspection images in the image server are labeled. The labeled data is divided into two categories: object labeling and hidden danger type standards. Object labeling is further divided into seven object subcategories, including: water heaters, gas meters, stoves, hoses under stoves, house numbers, security inspection forms, and kitchen environments. Hidden danger type labeling is divided into multiple types. Business personnel read the images through page operations and select the corresponding objects or hidden danger type options for corresponding markings, and then the marked security inspection images are stored in the marked classification folders on the backend server.

[0071] Upload the annotated security inspection pictures to the big data platform;

[0072] S103, AI platform reads security inspection pictures from the big data platform for training security inspection models;

[0073] Specifically, in this embodiment, the security inspection model uses a Resnet-18-based transfer learning model to realize image recognition. The specific training method is as follows:

[0074] Prepare the environment: Install libraries such as PyTorch and torchvision, which provide pre-trained models of ResNet-18 and common image processing tools;

[0075] Load pre-trained model: Load the pre-trained ResNet-18 model. You can choose a model pre-trained on the ImageNet dataset.

[0076] Adjust the model: Adjust the model to the classification task; typically, replace the fully connected layers with the number of output categories required by the task;

[0077] Data preparation: Prepare a data set of security inspection images, including training sets and test sets; use torchvision's transforms to define image preprocessing steps;

[0078] Data loader: Create a data loader for batch loading data during training and testing;

[0079] Define loss function and optimizer: Choose a loss function and optimizer; for classification tasks, cross entropy loss is usually used;

[0080] Training model: Use the training set to iteratively train the model and update the model weights;

[0081] Evaluate the model: Use the test set to test the trained model and evaluate the performance of the model. Usually, accuracy is used as the evaluation indicator.

[0082] Reasoning: Classify new images using a model whose performance meets the preset performance requirements;

[0083] Save and load models: Save models whose performance meets the preset performance requirements and load the models when called.

[0084] S104, the training results are output to the indicator library to provide data mining services;

[0085] Specifically, the training results are output to a table in the database, and page services are provided through interface calls; for example, employees use the page to determine whether the recognized image is correct.

[0086] S2. Obtain security inspection pictures and use security inspection models to assist in decision-making on hidden dangers, including:

[0087] Specifically, it includes: obtaining security inspection pictures, predicting hidden dangers through security inspection models (AI hidden danger identification auxiliary decision-making functions provided by the AI ​​platform), and generating prediction results; the prediction results include: security inspection objects and security inspection hidden danger types; the prediction results of security inspection object identification include seven categories: water heaters, gas meters, stoves, hoses under stoves, house numbers, security inspection forms, and kitchen environments; the prediction results of security inspection hidden danger type identification include multiple categories, such as corrosion of the front tube of the meter, corrosion of the meter, etc.

[0088] Verify the prediction results. If the prediction results match the hidden dangers identified manually, the prediction results will be transferred to the marketing system. If the prediction results do not match the hidden dangers identified manually, feedback will be given that there are problems with the prediction results, and the security inspection model will be optimized. Specifically, if the prediction results match the hidden dangers identified manually, the prediction will be marked as successful and the prediction results will be transferred to the marketing system. If the prediction results do not match the hidden dangers identified manually, the prediction will be marked as unsuccessful and the prediction results will be fed back to the AI ​​platform to adjust and optimize the security inspection model.

[0089] like Figure 2As shown, the decision-making is assisted by human-machine collaboration through the page, combined with the judgment of experienced employees, and the analysis results of AI (security inspection model) are combined with human intuition, experience and judgment; it helps administrators to conduct security inspections, and the prediction results can be sent through text messages or generated reports to assist administrators in judging whether employees have failed security inspections; and this solution also conducts iterative feedback, experienced employees provide feedback, and the AI ​​platform adjusts and optimizes according to the feedback; the prediction results are verified on-site through security inspections, and if true, they are marked as successful; if not, they are re-labeled for subsequent training.

[0090] S3, using the security inspection model to perform real-time security inspection image recognition;

[0091] Specifically, security inspection pictures stored on the terminal after security inspection are uploaded by employees through the terminal (web page and AP), and real-time security inspection pictures are taken and uploaded on-site through the terminal (AP). The security inspection model is used to perform real-time security inspection picture recognition and prediction, and identify security inspection objects and security inspection hazard types.

[0092] This solution builds a security inspection model and uses the trained security inspection model to assist in hidden danger decision-making. For the collected security inspection pictures, the security inspection model can be used to predict hidden dangers, generate prediction results, and push them; the prediction results include: security inspection objects and security inspection hidden danger types; this solution verifies the prediction results. If the prediction results match the hidden dangers identified manually, the prediction results are transferred to the marketing system; if the prediction results do not match the hidden dangers identified manually, feedback is given that there are problems with the prediction results, the security inspection model is optimized, and the recognition accuracy is improved, thereby improving the quality of security inspection;

[0093] During the safety test, Figure 3 As shown in the figure, the security inspection model is used to perform real-time security inspection picture recognition. First, through the collected security inspection pictures, it can be verified whether the employees take photos and upload them according to the regulations to ensure the basic collection needs; secondly, the administrator can judge whether there are any missed inspections of employees based on the prediction results; finally, the security inspection model is a recognition model trained with a large amount of data, which can accurately identify the security inspection objects and the types of security inspection hazards, improve the accuracy of security inspections, and ensure the quality of security inspections. With the assistance of the security inspection system, managers can also improve efficiency and better complete security inspections.

[0094] In summary, this solution can assist in safety inspections and, as an effective means of supervision, improve the quality and efficiency of safety inspections.

[0095] Embodiment 2

[0096] This embodiment provides a security detection system based on artificial intelligence, including: a server and a terminal;

[0097] The server is used to build a security inspection model and train the built security inspection model;

[0098] Terminal, used to obtain security inspection pictures and upload them to the server;

[0099] The server is also used to adopt security inspection models to assist in decision-making on hidden dangers, identify and predict security inspection images, and identify security inspection objects and types of security inspection hidden dangers.

[0100] Specifically, the server includes: data acquisition module, image server, data processing module, big data platform, AI platform, and indicator library;

[0101] The data acquisition module is used to capture historical security inspection data, where the security inspection data is security inspection pictures, including:

[0102] The data acquisition module grabs the data of security inspection pictures in the marketing system and puts them on the designated picture server. Security inspection pictures are generally stored in the marketing system. The number of security inspection pictures required for security inspection training is very large. Directly using the security inspection pictures in the marketing system will affect the business. Therefore, pictures are downloaded to the picture server through offline download. The reason why pictures are not uploaded directly is that the pictures stored in the marketing system exist in the form of links and cannot be used directly. Therefore, they need to be downloaded into binary streams first.

[0103] The data processing module is used to mark security inspection data, including:

[0104] The security inspection images in the image server are labeled. The labeled data is divided into two categories: object labeling and hidden danger type standards. Object labeling is further divided into seven object subcategories, including: water heaters, gas meters, stoves, hoses under stoves, house numbers, security inspection forms, and kitchen environments. Hidden danger type labeling is divided into multiple types. Business personnel read the images through page operations and select the corresponding objects or hidden danger type options for corresponding markings, and then the marked security inspection images are stored in the marked classification folders on the backend server.

[0105] Upload the annotated security inspection pictures to the big data platform;

[0106] The AI ​​platform reads security inspection images from the big data platform to train the security inspection model;

[0107] Specifically, in this embodiment, the security inspection model uses a Resnet-18-based transfer learning model to realize image recognition. The specific training method is as follows:

[0108] Prepare the environment: Install libraries such as PyTorch and torchvision, which provide pre-trained models of ResNet-18 and common image processing tools;

[0109] Load pre-trained model: Load the pre-trained ResNet-18 model. You can choose a model pre-trained on the ImageNet dataset.

[0110] Adjust the model: Adjust the model to the classification task; typically, replace the fully connected layers with the number of output categories required by the task;

[0111] Data preparation: Prepare a data set of security inspection images, including training sets and test sets; use torchvision's transforms to define image preprocessing steps;

[0112] Data loader: Create a data loader for batch loading data during training and testing;

[0113] Define loss function and optimizer: Choose a loss function and optimizer; for classification tasks, cross entropy loss is usually used;

[0114] Training model: Use the training set to iteratively train the model and update the model weights;

[0115] Evaluate the model: Use the test set to test the trained model and evaluate the performance of the model. Usually, accuracy is used as the evaluation indicator.

[0116] Reasoning: Classify new images using a model whose performance meets the preset performance requirements;

[0117] Save and load models: Save models whose performance meets the preset performance requirements and load the models when called.

[0118] The AI ​​platform is also used to output training results to the indicator library to provide data mining services;

[0119] Specifically, the training results are output to a table in the database, and page services are provided through interface calls; for example, employees use the page to determine whether the recognized image is correct.

[0120] The AI ​​platform provides AI hidden danger identification and decision-making support functions, using security inspection models to assist in hidden danger decision-making, including:

[0121] Specifically, it includes: the terminal collects security inspection pictures and uploads them to the AI ​​platform, and uses the security inspection model (the auxiliary decision-making function of AI hidden danger identification provided by the AI ​​platform) to predict hidden dangers and generate prediction results; the prediction results include: security inspection objects and security inspection hidden danger types; the prediction results of security inspection object identification include seven categories: water heaters, gas meters, stoves, hoses under stoves, house numbers, security inspection forms, and kitchen environments; the prediction results of security inspection hidden danger type identification include multiple categories, such as rust on the front tube of the meter, rust on the meter, etc.

[0122] Verify the prediction results. If the prediction results match the hidden dangers identified manually, the prediction results will be transferred to the marketing system. If the prediction results do not match the hidden dangers identified manually, feedback will be given that there are problems with the prediction results, and the security inspection model will be optimized. Specifically, if the prediction results match the hidden dangers identified manually, the prediction will be marked as successful and the prediction results will be transferred to the marketing system. If the prediction results do not match the hidden dangers identified manually, the prediction will be marked as unsuccessful and the prediction results will be fed back to the AI ​​platform to adjust and optimize the security inspection model.

[0123] Through the page, decision-making is assisted in a human-machine collaborative way, combined with the judgment of experienced employees, and the analysis results of AI (security inspection model) are combined with human intuition, experience and judgment; to help administrators conduct security inspections, the prediction results can be sent through text messages or generated reports to assist administrators in judging whether employees have failed security inspections; and this solution also conducts iterative feedback, experienced employees provide feedback, and the AI ​​platform adjusts and optimizes according to the feedback; through security inspections, the prediction results are verified on-site, and if true, they are marked as successful; if not, they are re-labeled for subsequent training.

[0124] Employees upload security inspection pictures stored on the terminal after security inspection through the terminal (web page and app), and take pictures on site and upload them to the AI ​​platform for real-time security inspection. The security inspection model is used to perform real-time security inspection picture recognition and prediction, and identify security inspection objects and security inspection hazard types.

[0125] The above is only an embodiment of the present invention. The common sense such as the known specific structure and characteristics in the scheme is not described in detail here. The ordinary technicians in the relevant field know all the common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all the existing technologies in the field, and have the ability to apply the conventional experimental means before that date. The ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the protection scope of the present invention, which will not affect the effect of the implementation of the present invention and the practicality of the patent. The protection scope required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A security detection method based on artificial intelligence, characterized in that: It includes the following: S1. Build a security inspection model and train the built security inspection model; S2. Obtain security inspection pictures and use security inspection models to assist in decision-making on hidden dangers; S3, using the security inspection model to perform real-time security inspection image recognition; Among them, S2 includes: Obtain security inspection pictures, use security inspection models to predict hidden dangers, and generate prediction results; the prediction results include: security inspection objects and security inspection hidden danger types; The prediction results are verified. If the prediction results match the hidden dangers identified manually, the prediction results are transferred to the marketing system. If the prediction results do not match the hidden dangers identified manually, feedback is given that there is a problem with the prediction results and the security inspection model is optimized.

2. The artificial intelligence-based security detection method according to claim 1, characterized in that: Said S1 comprises: S101, historical security inspection data capture, where the security inspection data is security inspection pictures; S102, marking the security inspection data, and uploading the marked security inspection pictures to the big data platform; S103, AI platform reads security inspection pictures from the big data platform for training security inspection models; S104: The training results are output to the indicator library to provide data mining services.

3. The artificial intelligence-based security detection method according to claim 1, characterized in that: The labeling data is divided into: object labeling and hidden danger type standards.

4. The artificial intelligence-based security detection method according to claim 2, characterized in that: The security inspection model adopts the Resnet-18 transfer learning model. The specific training method is as follows: Prepare the environment: Install the library, provide ResNet-18 pre-trained models and image processing tools; Load pre-trained model: Load the pre-trained ResNet-18 model and select the model pre-trained on the ImageNet dataset; Adjust the model: adjust the model according to the classification task; Data preparation: Prepare a data set of security inspection images, including training sets and test sets; Data loader: Create a data loader for batch loading data during training and testing; Define loss function and optimizer: Set a loss function and optimizer; Training model: Use the training set to iteratively train the model and update the model weights; Evaluate the model: Use the test set to test the trained model and evaluate the performance of the model. Usually, accuracy is used as the evaluation indicator. Reasoning: Classify new images using a model whose performance meets the preset performance requirements; Save and load models: Save models whose performance meets the preset performance requirements and load the models when called.

5. The artificial intelligence-based security detection method according to claim 1, characterized in that: The prediction results are pushed via SMS or generated reports.

6. The artificial intelligence-based security detection method according to claim 3, characterized in that: The S3 includes: uploading security inspection pictures stored on the terminal by employees after security inspection through the terminal and taking and uploading real-time security inspection pictures on site through the terminal, using the security inspection model to perform recognition and prediction on the real-time security inspection pictures, and identifying the security inspection objects and the types of security inspection hidden dangers.

7. A safety detection system based on artificial intelligence, characterized in that: include: Servers and terminals; The server is used to build a security inspection model and train the built security inspection model; Terminal, used to obtain security inspection pictures and upload them to the server; The server is also used to adopt the security inspection model to assist in decision-making on hidden dangers, identify and predict real-time security inspection images, and identify security inspection objects and types of security inspection hidden dangers; The security inspection model is used to assist in hidden danger decision-making, including: using the security inspection model to predict hidden dangers for the security inspection pictures uploaded by the terminal and generate prediction results; the prediction results include: security inspection objects and security inspection hidden danger types; The prediction results are verified. If the prediction results match the hidden dangers identified manually, the prediction results are transferred to the marketing system. If the prediction results do not match the hidden dangers identified manually, feedback is given that there is a problem with the prediction results and the security inspection model is optimized.

8. The artificial intelligence-based safety detection system according to claim 7, characterized in that: The server includes: a data acquisition module, a data processing module, a big data platform, an AI platform, and an indicator library; The data acquisition module is used to capture historical security inspection data, where the security inspection data is security inspection pictures; The data processing module is used to annotate security inspection data and upload the annotated security inspection images to the big data platform; The AI ​​platform reads security inspection images from the big data platform to train the security inspection model; The AI ​​platform is also used to output training results to the indicator library to provide data mining services.

9. The artificial intelligence-based safety detection system according to claim 8, characterized in that: The security inspection model adopts the Resnet-18 transfer learning model. The specific training method is as follows: Prepare the environment: Install the library, provide ResNet-18 pre-trained models and image processing tools; Load pre-trained model: Load the pre-trained ResNet-18 model and select the model pre-trained on the ImageNet dataset; Adjust the model: adjust the model according to the classification task; Data preparation: Prepare a data set of security inspection images, including training sets and test sets; Data loader: Create a data loader for batch loading data during training and testing; Define loss function and optimizer: Set a loss function and optimizer; Training model: Use the training set to iteratively train the model and update the model weights; Evaluate the model: Use the test set to test the trained model and evaluate the performance of the model. Usually, accuracy is used as the evaluation indicator. Reasoning: Classify new images using a model whose performance meets the preset performance requirements; Save and load models: Save models whose performance meets the preset performance requirements and load the models when called.

10. The artificial intelligence-based safety detection system according to claim 7, characterized in that: The identification and prediction of the real-time security inspection pictures, identification of security inspection objects and types of security inspection risks, include: Employees upload security inspection pictures stored on the terminal after security inspection, and take pictures on site through the terminal and upload them to the AI ​​platform for real-time security inspection. The security inspection model is used to perform real-time security inspection picture recognition and prediction, and identify security inspection objects and security inspection hazard types.

Images