Visual image analysis system and method based on YOLOv7 and variant genetic algorithm
By using the combination of YOLOv7 and variant genetic algorithms in the image analysis model training system, the problem of complex hyperparameter parameter adjustment and high time cost is solved, and efficient and accurate image analysis is achieved.
Patent Information
- Application Number
- CN202411887632.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-20
- Publication Date
- 2025-05-06
AI Technical Summary
In the existing image analysis model training system, the operation of adjusting hyperparameters is complex and time-consuming.
A visual image analysis system based on YOLOv7 and variant genetic algorithm is adopted. Through variant genetic algorithm, only variant operations are performed, no cross-operations are performed, and parameters are updated in combination with the Adam optimizer to optimize hyperparameters.
While ensuring the optimal solution quality, the efficiency of hyperparameter parameter adjustment is significantly improved, the time required for hyperparameter parameter adjustment is significantly reduced, and relatively accurate image analysis is achieved.
Smart Images
Figure CN119942295A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of computer image analysis, and in particular to a visual image analysis system and method based on YOLOv7 and a variant genetic algorithm. Background Art
[0002] Image analysis is an important branch in the field of artificial intelligence. It mainly involves importing annotated images, processing images and converting them into tensor data, selecting appropriate deep learning algorithm models, adjusting parameters and hyperparameters by comparing accuracy to train the algorithm model, evaluating the performance of the trained model, and finally selecting the model with good results for application. It is an effective method for image recognition to analyze and determine whether the image contains annotated content. Hyperparameters mainly affect the model structure and training methods. Parameters usually affect the weights and biases of the neural network. The existing image analysis model training methods mainly adjust hyperparameters manually based on experience and use various optimizers to automatically adjust parameters. Take the training of the YOLOv7 model as an example:
[0003] The first is to adjust the learning rate, starting from 0.01, and continue to try 0.001, 0.0001, 0.00001, 0.000001, and select the learning rate with better effect according to the error rate of the validation set;
[0004] The second method is to adjust the batch size to affect the efficiency of gradient descent during parameter optimization. Try the value of 2 to the power of 2 from small to large, and choose the batch size with better effect according to the error rate of the validation set.
[0005] The third method is to adjust the number of training rounds. Usually, a basic value is set according to the number of samples and the value is continuously increased. When the results of several consecutive rounds of training verification do not improve, the training is terminated.
[0006] Other hyperparameters include momentum, image size, image enhancement rate, etc., which are similar to those mentioned above.
[0007] The existing image analysis model training system imports labeled images into the system, sets the model according to the above-mentioned method of manually adjusting hyperparameters, and then uses the optimizer to train the parameters in the system, compares the analysis accuracy to find the appropriate hyperparameters, and trains the algorithm model parameters. This has the problems of complex operation and high time cost. Summary of the invention
[0008] Based on the above, the purpose of the present invention is to provide a visual image analysis system and method based on YOLOv7 and a variant genetic algorithm. By combining the YoloV7 algorithm model with a variant genetic algorithm that only performs high-probability and small-variance mutation operations, the system can quickly adjust the hyperparameters while performing accurate image analysis, thereby solving the problem of complex and time-consuming hyperparameter adjustment in related technologies.
[0009] To achieve the above object, the present invention adopts the following technical solutions:
[0010] In a first aspect, an embodiment of the present application provides a visual image analysis system based on YOLOv7 and a variant genetic algorithm, the system comprising a front-end page, a back-end service module and a database, the back-end service module comprising:
[0011] A data set management module, used to import and manage a data set, wherein the data set includes a picture and actual annotation information, the actual annotation information is annotated on the picture, and the data set is randomly divided into a training set, a validation set, and a test set according to a proportion;
[0012] An algorithm model management module is used to configure the YOLOv7 algorithm model and perform version management, configure the model name, model profile, model file, creation time and version number, and store them in the database;
[0013] A model training module is used to adjust hyperparameters through a variant genetic algorithm, wherein the variant genetic algorithm only performs mutation operations but does not perform crossover operations, and simultaneously optimizes the YOLOv7 algorithm model by updating parameters through an Adam optimizer;
[0014] The model evaluation module evaluates the training results of the YOLOv7 algorithm model according to the average mean accuracy standard, and saves the model if the evaluation passes, otherwise the training iteration continues;
[0015] The model application module is used to put the trained YOLOv7 algorithm model into practical application, import materials, analyze and determine whether the actual annotation information exists in the materials, and perform annotation.
[0016] In a second aspect, an embodiment of the present application provides a visual image analysis method based on YOLOv7 and a variant genetic algorithm, the method being applied to the system of claim 1, comprising the following steps:
[0017] Creating and processing a data set: calling the data management module, creating the data set and storing it in the database, the data set including the image and the actual annotation information, and randomly dividing the data set into the training set, the validation set and the test set according to a proportion;
[0018] Create an algorithm model: call the algorithm model management module, configure the YOLOv7 algorithm model and perform version management, configure the model name, the model profile, the model file, the creation time and the version number and store them in the database;
[0019] Model training: calling the model training module, adjusting hyperparameters through the variant genetic algorithm, wherein the variant genetic algorithm only performs mutation operations but does not perform crossover operations, and optimizing the YOLOv7 algorithm model by updating parameters through the Adam optimizer;
[0020] Model evaluation: calling the model evaluation module to evaluate the training results of the YOLOv7 algorithm model according to the average mean accuracy standard, and saving the model if the evaluation passes, otherwise continuing the training iteration;
[0021] Model application: calling the model application module, putting the trained YOLOv7 algorithm model into practical application, importing the material and analyzing and determining whether the actual annotation information exists in the material and annotating it.
[0022] As a preferred solution of a visual image analysis method based on YOLOv7 and a variant genetic algorithm, the data management module is called to create the data set and store it in the database, the data set includes the image and the actual annotation information, and the data set is randomly divided into the training set, the verification set and the test set according to a proportion, specifically including:
[0023] Create the data set on the front-end page, configure the name of the data set, the path of the picture and the actual annotation information on the front-end page, execute a first mouse click event to call the data management module to carry out a first back-end service, and store the name of the data set, the path of the picture and the actual annotation information in the database; select the data set on the front-end page, execute a second mouse click event to call the data management module to carry out a second back-end service, and randomly divide the data set into a training set, a validation set and a test set in a ratio of 7:2:1.
[0024] As a preferred solution of a visual image analysis method based on YOLOv7 and variant genetic algorithm, the algorithm model management module is called, the YOLOv7 algorithm model is configured and version management is performed, and the model name, the model profile, the model file, the creation time and the version number are configured and stored in the database, specifically including:
[0025] Create a YOLOv7 algorithm model on the front-end page and configure the model name, the model introduction, the model file, the creation time and the version number, execute a third mouse click event to call the algorithm model management module to carry out a third back-end service, and store the model name, the model introduction, the model file, the creation time and the version number in the database.
[0026] As a preferred solution of a visual image analysis method based on YOLOv7 and a variant genetic algorithm, the model training module is called, hyperparameters are adjusted by the variant genetic algorithm, the variant genetic algorithm only performs mutation operations but does not perform crossover operations, and the parameters are updated by the Adam optimizer to optimize the YOLOv7 algorithm model, specifically including:
[0027] Select the YOLOv7 algorithm model and the data set on the front-end page, execute the fourth mouse click event to call the model training module to carry out the fourth back-end service, obtain the model file and initialize the YOLOv7 algorithm model; obtain the paths of the training set, the validation set and the test set of the data set, and read the picture; adopt the variant genetic algorithm, only perform mutation operations with high probability and small variance, do not perform crossover operations, retain multiple offspring with the best performance for iteration, form a reference after the iteration reaches the average mean accuracy standard, automatically optimize hyperparameters, and use the Adam optimizer to optimize parameters; select the hyperparameters with the best performance and relatively high concentration and save them as a yaml file, and save the matching optimal parameters, model structure and the state of the Adam optimizer together as a pt file, write the yaml file and the pt file into the model file, and then write the new version number and save time of the model file on the front-end page and store them in the database.
[0028] As a preferred solution of a visual image analysis method based on YOLOv7 and a variant genetic algorithm, the model evaluation module is called to evaluate the training results of the YOLOv7 algorithm model according to the average mean accuracy standard. If the evaluation passes, the model is saved; otherwise, the training iteration is continued, which specifically includes:
[0029] The YOLOv7 algorithm model to be evaluated is selected on the front-end page, and a fifth mouse click event is executed to call the model evaluation module to perform a fifth back-end service. The YOLOv7 algorithm model analyzes the image to obtain reference annotation information, and compares and calculates the reference annotation information with the actual annotation information according to the average mean accuracy standard, and stores the calculation results in the database.
[0030] As a preferred solution of a visual image analysis method based on YOLOv7 and a variant genetic algorithm, the model application module is called, the trained YOLOv7 algorithm model is put into practical application, the material is imported, and the material is analyzed and determined to determine whether the actual annotation information exists and is annotated, specifically including:
[0031] On the front-end page, select the YOLOv7 algorithm model to be applied, import the material, execute the sixth mouse click event to call the model application module to carry out the sixth back-end service, and the YOLOv7 algorithm model analyzes the material to see whether there is corresponding information and marks it. The material includes pictures and videos, and the video will be converted into sliced pictures.
[0032] As a preferred solution of a visual image analysis method based on YOLOv7 and variant genetic algorithm, the first back-end service, the second back-end service, the third back-end service, the fourth back-end service, the fifth back-end service and the sixth back-end service will return success information to the front-end page after they are successfully carried out.
[0033] As a preferred solution of a visual image analysis method based on YOLOv7 and a variant genetic algorithm, the variant genetic algorithm performs a high-probability small-variance mutation operation with an 80% mutation probability and a 0.04 variance.
[0034] As a preferred solution of a visual image analysis method based on YOLOv7 and a variant genetic algorithm, the average mean accuracy standard is mAP50, that is, the cross-overlap threshold of the reference rectangular box and the actual rectangular box is not less than 0.5.
[0035] The beneficial effects of the present invention are:
[0036] 1. Compared with manual hyperparameter adjustment and traditional genetic algorithm hyperparameter adjustment, the variant genetic algorithm improves the solution efficiency while ensuring the quality of the optimal solution. It can not only complete relatively accurate image analysis, but also significantly reduce the time required for hyperparameter adjustment;
[0037] 2. Realize the visual integration of the entire process of data set creation, data set processing, model creation, model training, model evaluation, and model application;
[0038] 3. Can clearly display the version number and storage time of the model to be trained, which is convenient for version management;
[0039] 4. It can visually implement algorithm model training by filling in parameters and checking conditions, display the training process and result data, and reduce the difficulty of deep learning training models. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the contents of the embodiments of the present invention and these drawings without paying any creative work.
[0041] Figure 1 It is a structural schematic diagram of a visual image analysis system based on YOLOv7 and a variant genetic algorithm provided by the present invention; DETAILED DESCRIPTION
[0042] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It will be appreciated that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings. In the description of the present invention, unless otherwise specified, the meaning of "multiple" is two or more. In addition, the terms "first" and "second" are only used to distinguish in the description and have no special meaning.
[0043] Some terms used in the embodiments are described below to facilitate understanding by those skilled in the art.
[0044] Vue.js: Vue.js is a popular JavaScript front-end framework for building user interfaces. It uses a component-based development approach that can help developers build interactive single-page applications (SPAs). Vue.js provides a wealth of tools and libraries that enable developers to easily build modern front-end applications.
[0045] Spring Cloud: Spring Cloud is a microservice framework based on Spring Boot, which is a development framework for quickly building Java applications. They provide rich features and components for building and managing distributed systems. By using Spring Boot and Spring Cloud, you can build scalable and reliable RESTful services.
[0046] YOLO: is an open source algorithm for deep learning image detection. The core concept is to predict the target bounding box and category through a single neural network. It has been upgraded many times with the efforts of researchers around the world. The image annotation information storage format required by this algorithm is also called YOLO, which is expressed as a txt text file. The stored information includes the category ID, the coordinate information of the target box normalized relative to the height and width of the image, and the height and width information of the target box normalized relative to the height and width of the image.
[0047] This embodiment provides a visual image analysis system based on YOLOv7 and variant genetic algorithm. Figure 1As shown in the figure, the system is built based on Vue.js and Spring Cloud that integrates Python interface services. The main feature of the system is to quickly adjust hyperparameters through variant genetic algorithms and manual intervention, and automatically optimize parameters through the Adam optimizer to train the YOLOv7 algorithm model. The system includes a front-end page, a back-end service module and a database. The front-end page is used to configure operations, trigger the execution of operations and display execution results. The back-end service module is used to execute operations and return execution results. The database is used to store information.
[0048] Optionally, the database is MySQL, and the information is stored in the database by executing SQL statements.
[0049] Furthermore, the backend service module includes:
[0050] The data set management module is used to import annotated image data and manage basic data set information, including data set name, sample number, actual annotation information, creation time and other data. The actual annotation information includes VOC, COCO, YOLO and other annotation formats. This system is mainly used to train the YOLOV7 algorithm model. Other forms of annotation information will eventually be converted into YOLO format during model training and saved as txt text; data management functions include adding, deleting, modifying and querying data set information;
[0051] The algorithm model management module is used to configure the algorithm model information and version management. The algorithm model information includes model name, algorithm type, model introduction, preset model, creation time and version number. Version management includes saving the version number, model file and creation time after model training. The preset model can select the V5, V6 and V7 versions of the YOLO algorithm model. This system is mainly used to train the YOLOv7 algorithm model. The algorithm model management function includes adding, deleting, modifying and querying the algorithm model information.
[0052] The model training module is used to adjust hyperparameters through variant genetic algorithms and manual intervention, and also to optimize the YOLOv7 algorithm model by updating parameters through the Adam optimizer, and to select the average mean accuracy standard to complete the training iteration. It contains a lot of deep learning data analysis operations. Since the framework content for implementing data analysis in the Java language is not comprehensive, it needs to be implemented in Python language code, that is, in the JAVA code, the exec() method of the Runtime class is used to execute external Python scripts and pass parameters, process.getInputStream() is used to capture the input stream, process.getErrorStream() is used to capture the error stream, process.waitFor() is used to obtain the return value, and sys.argv is used in the Python code to obtain parameters and execute the program;
[0053] The model evaluation module is used to evaluate the quality of the trained model, select the algorithm model, set the evaluation average mean accuracy standard, and calculate the average mean accuracy of the algorithm model to verify the model training results;
[0054] The model application module is used to actually apply the trained YOLOv7 algorithm model, select the model, and import the materials. The materials can be pictures and videos. The videos will be sliced into pictures at certain intervals according to the set parameters. The system will analyze whether there is corresponding information based on the trained YOLOv7 algorithm model and mark it on the picture.
[0055] Optionally, the average mean accuracy standard used during model training iteration and model evaluation comparison is an evaluation index such as mAP50 and mAP75, that is, the cross-overlap threshold between the reference rectangular box and the actual rectangular box is not less than 0.5 and 0.75. In this embodiment, the average mean accuracy mAP50 is used as the standard.
[0056] This embodiment also provides a visual image analysis method based on YOLOv7 and a variant genetic algorithm, which is applied to the aforementioned system and specifically includes the following steps:
[0057] S101 creates a data set: creates a data set on the front-end page, the data set includes a picture and actual annotation information, and the picture and the actual annotation information are stored in a database; configures the name of the data set, the path of the picture and the actual annotation information on the front-end page, executes a first mouse click event, sends an HTTP request through JavaAPI to call a data management module to carry out a first back-end service, stores the name of the data set, the path of the picture and the actual annotation information in the database, and returns a success message;
[0058] S102 Dataset processing: Select a dataset on the front-end page, execute the second mouse click event, send an HTTP request through JavaAPI to call the data management module to carry out the second back-end service, call the Python script through the exec() method of the Runtime class in the JAVA code, obtain the image path and annotation information according to the selected dataset, the image will be reconstructed into a 640*640 square and converted into tensor format, the annotation information will be parsed and uniformly converted into YOLO format, the dataset will be randomly divided into training set, validation set and test set in a ratio of 7:2:1, the validation set is used to debug hyperparameters, regulate the model training process, and avoid problems such as overfitting or underfitting that affect the performance of the YOLOv7 algorithm model, the training set is used to debug parameters, and the test set is used to evaluate and test the trained model, and a success message is returned after the dataset division is completed;
[0059] S103 creates an algorithm model: creates a YOLOv7 algorithm model on the front-end page and configures the model name, model introduction, model file, creation time and version number, and sets the initial value of the hyperparameter. The initial value can be manually set according to the default value or the previously trained project, executes a third mouse click event, sends an HTTP request through the Java API to call the algorithm model management module to carry out a third back-end service, stores the model name, model introduction, model file, creation time and version number in the database, and returns a success message;
[0060] S104 model training: select the YOLOv7 algorithm model and data set on the front-end page, execute the fourth mouse click event, send an HTTP request through the JavaAPI to call the model training module to carry out the fourth back-end service, read the yaml file containing hyperparameters in the model file and the pt file containing parameters, model architecture, Adam optimizer status and other information, initialize the YOLOv7 algorithm model, and initialize it according to the preset model if it cannot be read; obtain the paths of the training set, validation set, and test set of the data set, and read the image; use a variant genetic algorithm, only perform mutation operations, do not perform crossover operations, set 80% mutation probability, 0.04 variance, retain the three offspring with the best performance for iteration, and form a reference after the iteration reaches the standard of average mean accuracy of mAP50, automatically optimize the hyperparameters, and compare the traditional The genetic algorithm is used to directly select the top 3 individual iterations in terms of performance, which can reduce the computational cost and achieve rapid convergence. At the same time, because only mutation operations with high probability and small variance are used, the algorithm will not converge too much and fall into the local optimal solution. At the same time, the Adam optimizer is used to optimize the parameters, and the parameters are adjusted by the moving average of the first-order moment and the second-order moment of the gradient. Compared with the traditional gradient descent method, it has stronger learning efficiency and versatility in processing sparse data, and can accelerate convergence and reduce turbulence. The hyperparameters with the best performance and relatively high concentration are manually selected and saved as yaml files. At the same time, the matching optimal parameters, model structure and optimizer status are saved together as pt files. The yaml file and pt file are written into the model file, and then the new version number and save time of the model file are written into the front-end page and stored in the database, and finally a success message is returned.
[0061] S105 Model evaluation: Select the YOLOv7 algorithm model to be evaluated on the front-end page, fill in the parameters, including whether to use the test set divided during training, the custom test set path, the minimum confidence threshold and the average mean accuracy standard. If not set, the evaluation is performed by default. The fifth mouse click event is executed, and the model evaluation module is called through the Java API to send an HTTP request to carry out the fifth back-end service. In the JAVA code, the Python script is called through the exec() method of the Runtime class. The YOLOv7 algorithm model analyzes the image to obtain reference annotation information, compares and calculates the reference annotation information with the actual annotation information according to the standard, stores the calculation results in the database, and returns a success message.
[0062] S106 Model application: Select the YOLOv7 algorithm model to be applied on the front-end page, import the material, execute the sixth mouse click event, send an HTTP request through the Java API to call the model application module to carry out the sixth back-end service, call the Python script through the exec() method of the Runtime class in the JAVA code, and the YOLOv7 algorithm model analyzes the material to obtain the corresponding annotation information and returns a success message. The imported materials include pictures and videos, and the videos will be converted into sliced pictures, and the sliced pictures will also be converted into tensor format.
[0063] Furthermore, the first backend service, the second backend service, the third backend service, the fourth backend service, the fifth backend service and the sixth backend service are microservices built based on Spring Cloud.
[0064] In a specific embodiment, manual operation, traditional genetic algorithm (crossover probability 80%, mutation probability 2%, higher average mean accuracy of the offspring retention probability) and variant genetic algorithm (no crossover, mutation probability 80%, retaining the three offspring with the highest average accuracy for the next iteration) are used for hyperparameter optimization. When 5,000 sample pictures of people wearing helmets are trained for production operation behavior image analysis on the same machine, the test results of average mean accuracy mAP50 and calculation time are shown in the following table:
[0065] Hyperparameter Tuning Methods Manual parameter adjustment Traditional genetic algorithm Variant Genetic Algorithm mAP50 0.706 0.861 0.856 time Unable to calculate 51 hours, 47 minutes and 33 seconds 38 hours, 16 minutes and 17 seconds
[0066] From the above table, it can be seen that the visual image analysis system and method based on YOLOv7 and variant genetic algorithm provided by the present application can not only complete relatively accurate image analysis, but also significantly reduce the time required for hyperparameter adjustment.
[0067] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the technical field not disclosed in the present disclosure. The specification and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.
[0068] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.
Claims
1. A visual image analysis system based on YOLOv7 and variant genetic algorithm, characterized in that: The system includes a front-end page, a back-end service module and a database, and the back-end service module includes: A data set management module, used to import and manage a data set, wherein the data set includes a picture and actual annotation information, the actual annotation information is annotated on the picture, and the data set is randomly divided into a training set, a validation set, and a test set according to a proportion; An algorithm model management module is used to configure the YOLOv7 algorithm model and perform version management, configure the model name, model profile, model file, creation time and version number, and store them in the database; A model training module is used to adjust hyperparameters through a variant genetic algorithm, wherein the variant genetic algorithm only performs mutation operations but does not perform crossover operations, and simultaneously optimizes the YOLOv7 algorithm model by updating parameters through an Adam optimizer; The model evaluation module evaluates the training results of the YOLOv7 algorithm model according to the average mean accuracy standard, and saves the model if the evaluation passes, otherwise the training iteration continues; The model application module is used to put the trained YOLOv7 algorithm model into practical application, import materials, analyze and determine whether the actual annotation information exists in the materials, and perform annotation.
2. A visual image analysis method based on YOLOv7 and variant genetic algorithm, the method being applied to the system according to claim 1, characterized in that: The following steps are involved: Creating and processing a data set: calling the data management module, creating the data set and storing it in the database, the data set including the image and the actual annotation information, and randomly dividing the data set into the training set, the validation set and the test set according to a proportion; Create an algorithm model: call the algorithm model management module, configure the YOLOv7 algorithm model and perform version management, configure the model name, the model profile, the model file, the creation time and the version number and store them in the database; Model training: calling the model training module, adjusting hyperparameters through the variant genetic algorithm, wherein the variant genetic algorithm only performs mutation operations but does not perform crossover operations, and optimizing the YOLOv7 algorithm model by updating parameters through the Adam optimizer; Model evaluation: calling the model evaluation module to evaluate the training results of the YOLOv7 algorithm model according to the average mean accuracy standard, and saving the model if the evaluation passes, otherwise continuing the training iteration; Model application: calling the model application module, putting the trained YOLOv7 algorithm model into practical application, importing the material and analyzing and determining whether the actual annotation information exists in the material and annotating it.
3. A visual image analysis method based on YOLOv7 and variant genetic algorithm according to claim 2, characterized in that: The calling of the data management module, creating the data set and storing it in the database, wherein the data set includes the image and the actual annotation information, and randomly dividing the data set into the training set, the validation set, and the test set according to a proportion, specifically includes: Create the data set on the front-end page, configure the name of the data set, the path of the picture and the actual annotation information on the front-end page, execute a first mouse click event to call the data management module to carry out a first back-end service, and store the name of the data set, the path of the picture and the actual annotation information in the database; select the data set on the front-end page, execute a second mouse click event to call the data management module to carry out a second back-end service, and randomly divide the data set into a training set, a validation set and a test set in a ratio of 7:2:
1.
4. The visual image analysis method based on YOLOv7 and variant genetic algorithm according to claim 3, characterized in that: The calling of the algorithm model management module, configuring the YOLOv7 algorithm model and performing version management, configuring the model name, the model profile, the model file, the creation time and the version number and storing them in the database specifically includes: Create a YOLOv7 algorithm model on the front-end page and configure the model name, the model introduction, the model file, the creation time and the version number, execute a third mouse click event to call the algorithm model management module to carry out a third back-end service, and store the model name, the model introduction, the model file, the creation time and the version number in the database.
5. A visual image analysis method based on YOLOv7 and variant genetic algorithm according to claim 4, characterized in that: The calling of the model training module, adjusting the hyperparameters by using the variant genetic algorithm, wherein the variant genetic algorithm only performs mutation operation but does not perform crossover operation, and optimizing the YOLOv7 algorithm model by updating the parameters by using the Adam optimizer, specifically includes: Select the YOLOv7 algorithm model and the data set on the front-end page, execute the fourth mouse click event to call the model training module to carry out the fourth back-end service, obtain the model file and initialize the YOLOv7 algorithm model; obtain the paths of the training set, the validation set and the test set of the data set, and read the picture; adopt the variant genetic algorithm, only perform mutation operations with high probability and small variance, do not perform crossover operations, retain multiple offspring with the best performance for iteration, form a reference after the iteration reaches the average mean accuracy standard, automatically optimize hyperparameters, and use the Adam optimizer to optimize parameters; select the hyperparameters with the best performance and relatively high concentration and save them as a yaml file, and save the matching optimal parameters, model structure and the state of the Adam optimizer together as a pt file, write the yaml file and the pt file into the model file, and then write the new version number and save time of the model file on the front-end page and store them in the database.
6. A visual image analysis method based on YOLOv7 and variant genetic algorithm according to claim 5, characterized in that: Call the model evaluation module to evaluate the training results of the YOLOv7 algorithm model according to the average mean accuracy standard. If the evaluation passes, save the model, otherwise continue training iteration, specifically including: The YOLOv7 algorithm model to be evaluated is selected on the front-end page, and a fifth mouse click event is executed to call the model evaluation module to perform a fifth back-end service. The YOLOv7 algorithm model analyzes the image to obtain reference annotation information, and compares and calculates the reference annotation information with the actual annotation information according to the average mean accuracy standard, and stores the calculation results in the database.
7. The visual image analysis method based on YOLOv7 and variant genetic algorithm according to claim 6, characterized in that: Calling the model application module, putting the trained YOLOv7 algorithm model into practical application, importing the material and analyzing and determining whether the material contains the actual annotation information and annotating it, specifically includes: On the front-end page, select the YOLOv7 algorithm model to be applied, import the material, execute the sixth mouse click event to call the model application module to carry out the sixth back-end service, and the YOLOv7 algorithm model analyzes the material to see whether there is corresponding information and marks it. The material includes pictures and videos, and the video will be converted into sliced pictures.
8. The visual image analysis method based on YOLOv7 and variant genetic algorithm according to claim 7, characterized in that: After the first backend service, the second backend service, the third backend service, the fourth backend service, the fifth backend service and the sixth backend service are successfully implemented, they will return success information to the front-end page.
9. The visual image analysis method based on YOLOv7 and variant genetic algorithm according to claim 7, characterized in that: The variant genetic algorithm performs a high-probability, small-variance mutation operation with a mutation probability of 80% and a variance of 0.
04.
10. The visual image analysis method based on YOLOv7 and variant genetic algorithm according to claim 7, characterized in that: The average mean accuracy standard is mAP50, that is, the cross-overlap threshold between the reference rectangular box and the actual rectangular box is not less than 0.5.