Picture labeling training method and device

Abstract

The present application belongs to the technical field of image recognition, and particularly relates to a picture labeling training method and device, the method comprising: in response to a user's uploading operation, obtaining training pictures. The present application first responds to the user's uploading operation, then obtains the training pictures and classifies them according to preset classification label groups to generate corresponding labeling tasks, obtains the labeling dataset generated by the user's labeling operation on the training pictures, and uses the labeling dataset to train a generated result model. The present application classifies labeling tasks by setting different classification label groups, avoids the situation of reduced efficiency and erroneous classification results caused by manual classification, helps the user quickly and conveniently complete the labeling task, and makes the subsequent model training process more accurate and efficient. The present application not only improves the labeling efficiency, but also helps the user to process a large amount of data more easily and quickly.

Description

Technical Field

[0001] This invention belongs to the field of image recognition technology, and in particular relates to an image annotation training method and apparatus. Background Technology

[0002] Spring Boot is an open-source framework based on the Spring framework for rapidly building web applications and microservices. It simplifies the development of Spring applications by providing default configurations and conventions, enabling developers to build, test, and deploy applications faster. Deep learning, or artificial intelligence, is a machine learning method that uses multi-layered neural networks to perform complex tasks. Deep learning excels in natural language processing, image recognition, and speech recognition. With increasing computing power and data volume, its applications are expanding to include autonomous driving, medical diagnosis, and financial risk management.

[0003] With the increasing widespread application of deep learning, more and more application scenarios require image datasets for model training. However, existing technologies for processing image datasets still have many inconveniences, especially when dealing with large-scale image datasets. Current technologies typically employ manual classification when processing image datasets. When dealing with large datasets, manual initial processing is required, involving filtering and classifying different image datasets one by one, and manually recording classification information. This can easily lead to low classification efficiency and classification errors, affecting the efficiency and convenience of subsequent annotation based on classification information, and ultimately resulting in low model training efficiency. Summary of the Invention

[0004] The purpose of this invention is to provide an image annotation training method and apparatus, which aims to solve the problems of low classification efficiency and accuracy, inconvenient annotation, and low model training efficiency in the existing technology.

[0005] To achieve the above objectives, embodiments of the present invention provide an image annotation training method, comprising the following steps:

[0006] Step S100: In response to the user's upload operation, obtain the training image;

[0007] Step S200: Classify the training images according to the preset classification label group and generate the corresponding annotation task, wherein the classification label group is preset;

[0008] Step S300: In response to the user's annotation operation on the training image, obtain the annotation dataset generated based on the training image and the annotation operation, wherein the annotation operation is performed according to the annotation task;

[0009] Step S400: Train the labeled dataset to generate the resulting model.

[0010] Optionally, step S200: classifying the training images according to a preset classification label group and generating a corresponding annotation task, specifically includes:

[0011] Step S210: Determine whether the training image contains historical annotation information;

[0012] Step S220: If the determination is negative, then classify the training images according to the preset classification label group;

[0013] Step S230: Create a labeling task based on the classification label group.

[0014] Optionally, step S200, which involves classifying the training images according to a preset classification label group and generating a corresponding annotation task, further includes:

[0015] Step S240: If the determination is yes, then perform reverse parsing on the training image to obtain historical annotation information;

[0016] Step S250: Create a re-annotation task based on the historical annotation information.

[0017] Optionally, step S240: if the determination is yes, then reverse parsing the training image to obtain historical annotation information, specifically including:

[0018] Step S241: Perform reverse parsing on the training image based on OpenCV to obtain historical annotation information, wherein the historical annotation information includes classification label information and segmentation coordinate information;

[0019] Step S242: Store the historical annotation information in a MySQL database.

[0020] Optionally, step S300: In response to the user's annotation operation on the training image, obtain the annotation dataset generated based on the training image and the annotation operation, specifically including:

[0021] Step S310: In response to the user's annotation operation on the training image, obtain the annotation information generated according to the annotation operation;

[0022] Step S320: Using the Java-based Polygon tool, fill the labeled image with the labeled information to generate a labeled dataset;

[0023] Optionally, step S400: training the labeled dataset to generate the resulting model specifically includes:

[0024] Step S410: Generate a training task based on the labeled dataset;

[0025] Step S420: Run the Docker training model based on the preset remote training server, and generate the result model according to the training task.

[0026] Optionally, step S420: running a Docker training model based on a preset remote training server, and generating a result model according to the training task, specifically includes:

[0027] Step S421: Upload the training dataset to the training directory using JSch and Docker;

[0028] Step S422: Run Docker to train the model and generate the result model based on the training task;

[0029] Step S423: Send the result model to an external display terminal, wherein the external display terminal is used to display the result model.

[0030] Optionally, the image annotation training method further includes:

[0031] Step S500: Based on Alibaba OSS files, store the training images, the labeled dataset, and the result model in the cloud.

[0032] Optionally, step S100: in response to the user's upload operation, acquiring training images specifically includes:

[0033] Step S110: In response to the user's upload operation, receive the training image set based on the Spring Boot MultipartFile class;

[0034] Step S120: Analyze the training image set to obtain training images.

[0035] The present invention also provides an image annotation training device, the device comprising:

[0036] The training image acquisition module is used to acquire training images in response to user upload operations;

[0037] The annotation task generation module is used to classify the training images according to a preset classification label group and generate corresponding annotation tasks, wherein the classification label group is preset.

[0038] The annotation data generation module is used to respond to the user's annotation operation on the training image and obtain the annotation dataset generated based on the training image and the annotation operation, wherein the annotation operation is performed according to the annotation task;

[0039] The result model generation module is used to train the labeled dataset to generate a result model;

[0040] The image annotation training method and apparatus provided in this embodiment of the invention have at least the following technical effects:

[0041] This invention first responds to the user's upload operation, then acquires training images and classifies them according to preset classification label groups to generate corresponding annotation tasks. After the user performs annotation operations on the training images, the generated annotation dataset is obtained and used to train the resulting model. This invention classifies annotation tasks by setting different classification label groups, avoiding the inefficiency and errors caused by manual classification. It can also help users complete annotation tasks quickly and conveniently, and make the subsequent model training process more accurate and efficient. It not only improves annotation efficiency, but also helps users process large amounts of data more easily and quickly. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is an overall flowchart of the image annotation training method provided in the embodiments of the present invention;

[0044] Figure 2 This is a flowchart of a process for acquiring training images in response to a user's upload operation, provided in an embodiment of the present invention.

[0045] Figure 3 This is a flowchart of a task for classifying training images and generating corresponding annotations based on preset classification label groups, provided in an embodiment of the present invention.

[0046] Figure 4 This is a flowchart illustrating the process of obtaining historical annotation information from reverse-analyzed training images, as provided in an embodiment of the present invention.

[0047] Figure 5 A flowchart illustrating the generated labeled dataset provided in this embodiment of the invention;

[0048] Figure 6 This is a flowchart illustrating the model for generating the training labeled dataset provided in an embodiment of the present invention.

[0049] Figure 7This is a flowchart illustrating the generation of a result model based on the training task, provided in an embodiment of the present invention.

[0050] Figure 8 This is a structural block diagram of the image annotation training device provided in an embodiment of the present invention.

[0051] The following are the labeling elements in the figure:

[0052] 1- Training image acquisition module; 2- Labeling task generation module; 3- Labeled data generation module; 4- Result model generation module; 5- Data cloud storage module. Detailed Implementation

[0053] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0054] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0055] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0056] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0057] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0058] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0059] In another embodiment of the invention, such as Figure 1 As shown, an image annotation training method is provided, including the following steps:

[0060] Step S100: In response to the user's upload operation, obtain the training image;

[0061] In this step, the uploaded training images are obtained, which can be a single file or a compressed package.

[0062] Step S200: Classify the training images according to the preset classification label group and generate the corresponding annotation task, wherein the classification label group is preset;

[0063] In this step, the classification label group is used to store the classification categories required by the business. The training images are classified according to their different categories, so that the training images have clear category divisions and corresponding annotation tasks are generated, which facilitates the subsequent annotation process.

[0064] Step S300: In response to the user's annotation operation on the training image, obtain the annotation dataset generated based on the training image and the annotation operation, wherein the annotation operation is performed according to the annotation task;

[0065] In this step, users independently annotate the training images, allowing multiple users to annotate simultaneously and from different locations anytime, anywhere, significantly improving annotation efficiency. User annotation of the training images generates a corresponding labeled dataset, which is then used for subsequent deep learning training.

[0066] Step S400: Train the labeled dataset to generate the resulting model.

[0067] In this step, after the labeled dataset is labeled, it is uploaded to the designated location for training the model, and training parameters are created. Training is then performed on the specified labeled dataset, and the training progress is published to an external display terminal in real time during the training process. Upon completion of training, the training result model is available for download, and preset packaging parameters are provided to assist in packaging the training result model and generating a .db file for deployment on the appropriate machine. This ensures the efficiency of the training process and the real-time openness of training information, enhancing the resilience of the training.

[0068] In this embodiment, the present invention first responds to the user's upload operation, then acquires training images and classifies them according to preset classification label groups to generate corresponding annotation tasks. After the user performs annotation operations on the training images, the generated annotation dataset is acquired and used to train the resulting model. The present invention classifies annotation tasks by setting different classification label groups, avoiding the inefficiency and errors caused by manual classification. At the same time, it can help users complete annotation tasks quickly and conveniently, and make the subsequent model training process more accurate and efficient. It not only improves annotation efficiency, but also helps users process large amounts of data more easily and quickly.

[0069] In another embodiment of the invention, such as Figure 2 As shown, step S100: In response to the user's upload operation, acquiring training images specifically includes:

[0070] Step S110: In response to the user's upload operation, receive the training image set based on the Spring Boot MultipartFile class;

[0071] Step S120: Analyze the training image set to obtain training images.

[0072] In this embodiment, the parsing operation of the training image set is performed by using Spring Boot's MultipartFile class to receive image files and packaged compressed files, and using Java's built-in ZipFile and ZipInputStream to parse the uploaded compressed files to obtain training images, thus achieving convenient and fast acquisition of the training image set, making the entire annotation and training process faster.

[0073] In another embodiment of the invention, such as Figure 3 As shown, step S200: classifying the training images according to a preset classification label group and generating a corresponding annotation task, specifically includes:

[0074] Step S210: Determine whether the training image contains historical annotation information;

[0075] Step S220: If the determination is negative, then classify the training images according to the preset classification label group;

[0076] Step S230: Create a labeling task based on the classification label group.

[0077] In this embodiment, each category label group contains multiple label information items, which are further classifications of the category label group. Creating annotation tasks based on these category label groups makes the classification of annotation tasks clearer and ensures the efficiency of subsequent processes.

[0078] In another embodiment of the invention, such as Figure 3 As shown, step S200, which involves classifying the training images according to a preset classification label group and generating corresponding annotation tasks, further includes:

[0079] Step S240: If the determination is yes, then perform reverse parsing on the training image to obtain historical annotation information;

[0080] Step S250: Create a re-annotation task based on the historical annotation information.

[0081] In this embodiment, it is determined whether the training image has historical annotation information. If the historical annotation information exists, the training image with the historical annotation information is obtained, and the re-annotation task is created according to the historical annotation information to perform iterative annotation on the training image again to ensure the integrity of the information in the training image.

[0082] In another embodiment of the invention, such as Figure 4 As shown, step S240: if the determination is yes, then reverse parsing the training image to obtain historical annotation information, specifically including:

[0083] Step S241: Perform reverse parsing on the training image based on OpenCV to obtain historical annotation information, wherein the historical annotation information includes classification label information and segmentation coordinate information;

[0084] Step S242: Store the historical annotation information in a MySQL database.

[0085] In this embodiment, reverse parsing refers to processing the pre-annotated segmented image to obtain the historical annotation information and extract the segmentation coordinate information, which can then be viewed and edited on a pre-built Vue framework page. Specifically, the reverse parsing uses OpenCV technology, decompiles the Windows opencv_java453.dll file, and generates a libopencv_java453.so file for deployment on a Linux server. The coordinate information is then parsed using the Imgproc.findContours method, which OpenCV depends on, and subjected to interval filtering. The training image containing the historical annotation information is obtained through reverse parsing, and a corresponding re-annotation task is created based on this information. The training image is then iteratively annotated multiple times to further ensure the integrity of the information in the training image.

[0086] In another embodiment of the invention, such as Figure 5 As shown, step S300: In response to the user's annotation operation on the training image, obtain the annotation dataset generated based on the training image and the annotation operation, specifically including:

[0087] Step S310: In response to the user's annotation operation on the training image, obtain the annotation information generated according to the annotation operation;

[0088] Step S320: Using the Java-based Polygon tool, fill the labeled image with the labeled information to generate a labeled dataset;

[0089] In this embodiment, the user performs annotation operations on the training images according to the annotation task. The annotation task is divided into a classification annotation task and a segmentation annotation task. The classification annotation task involves assigning corresponding label information to each training image and generating a classification image. The segmentation task involves using a pre-built Vue front-end framework to perform custom pattern selection segmentation annotation on multiple small objects in each training image, and generating the segmented image after specifying the label information corresponding to the current task. After the user annotates the training images, the annotated segmented images are used to generate a training dataset in a specified format. Specifically, this invention uses the Polygon utility class built into Java to fill the original image (different colors for the same type / same color for the same type) according to the coordinate information stored in the MySQL database, generating SegmentationClass PNG and SegmentationObject PNG, which are the annotated dataset. The above operations fully ensure the accuracy of the training dataset and provide more effective training materials for the subsequent training process.

[0090] In another embodiment of the invention, such as Figure 6 As shown, step S400: training the labeled dataset to generate the result model specifically includes:

[0091] Step S410: Generate a training task based on the labeled dataset;

[0092] Step S420: Run the Docker training model based on the preset remote training server, and generate the result model according to the training task.

[0093] In this embodiment, a training task is generated based on the labeled dataset, and then the Docker training model is run on a remote training server. Finally, the result model is generated based on the training task. Training on a remote training server effectively avoids performance limitations caused by insufficient local server resources and further improves the stability of the training task.

[0094] In another embodiment of the invention, such as Figure 7 As shown, step S420: running a Docker training model based on a preset remote training server, and generating a result model according to the training task, specifically includes:

[0095] Step S421: Upload the training dataset to the training directory using JSch and Docker;

[0096] Step S422: Run Docker to train the model and generate the result model based on the training task;

[0097] Step S423: Send the result model to an external display terminal, wherein the external display terminal is used to display the result model.

[0098] In this embodiment, the training task uses JSch technology to remotely operate a Linux server and run instructions. A pre-deployed training model is launched on the remote training server using Docker, and training is performed on the training dataset in a specified directory. The training progress is sent to an external display terminal in real time. The training result file is obtained through the ChannelSftp class, a dependency of JSch, which is part of the core SFTP class implemented by JSch.

[0099] In another embodiment of the invention, such as Figure 1 As shown, the image annotation training method further includes:

[0100] Step S500: Based on Alibaba OSS files, store the training images, the labeled dataset, and the result model in the cloud.

[0101] In this embodiment, the training images, the labeled dataset, and the resulting model are uploaded and saved to an OSS cloud server for easy file use and transfer. Specifically, the use of an OSS cloud server for storage ensures the security of the file data system and convenient resource scalability, avoiding resource waste.

[0102] In another embodiment of the invention, such as Figure 8 As shown, an image annotation training device includes a training image acquisition module 1, an annotation task generation module 2, an annotation data generation module 3, and a result model generation module 4.

[0103] The training image acquisition module 1 is used to acquire training images in response to the user's upload operation. The annotation task generation module 2 is used to classify the training images according to a preset classification label group and generate corresponding annotation tasks. The annotation data generation module 3 is used to acquire an annotation dataset generated based on the training images and the annotation operation in response to the user's annotation operation on the training images. The result model generation module 4 is used to train the annotation dataset to generate a result model.

[0104] In another embodiment of the invention, such as Figure 8 As shown, the device also includes a data cloud storage module 5.

[0105] The cloud storage module 5 is used to store the training images, the labeled dataset, and the result model in the cloud.

[0106] In another embodiment of the invention, such as Figure 8 As shown, the annotation task generation module 2 is further configured to: determine whether the training image has historical annotation information; if the determination is no, classify the training image according to a preset classification label group; and create an annotation task according to the classification label group.

[0107] In another embodiment of the invention, such as Figure 8 As shown, the annotation task generation module 2 is further configured to: if the determination is yes, perform reverse parsing on the training image to obtain historical annotation information; and create a re-annotation task based on the historical annotation information.

[0108] In another embodiment of the invention, such as Figure 8 As shown, the annotation task generation module 2 is further configured to: perform reverse parsing on the training image based on OpenCV to obtain historical annotation information, wherein the historical annotation information includes classification label information and segmentation coordinate information; and store the historical annotation information in a MySQL database.

[0109] In another embodiment of the invention, such as Figure 8 As shown, the annotation data generation module 3 is further configured to: in response to the user's annotation operation on the training image, obtain annotation information generated according to the annotation operation; use the Java-based Polygon tool to fill the annotation image with the annotation information to generate an annotation dataset.

[0110] In another embodiment of the invention, such as Figure 8 As shown, the result model generation module 4 is further configured to: generate a training task based on the labeled dataset; run a Docker training model based on a preset remote training server; and generate a result model based on the training task.

[0111] In another embodiment of the invention, such as Figure 8 As shown, the result model generation module 4 is further configured to: upload the training dataset to the training directory via JSch and Docker; run the Docker training model and generate a result model based on the training task; and send the result model to an external display terminal, wherein the external display terminal is used to display the result model.

[0112] To achieve the above objectives, embodiments of the present invention also provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the image annotation training method.

[0113] To achieve the above objectives, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the image annotation training method.

[0114] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0115] The modules described as separate components may or may not be physically separate. Similarly, the components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0116] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0117] If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned readable storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0118] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A picture labeling training method, characterized in that, Includes the following steps: Step S100: In response to the user's upload operation, obtain the training image; Step S200: Classify the training images according to the preset classification label group and generate the corresponding annotation task, wherein the classification label group is preset; Step S300: In response to the user's annotation operation on the training image, obtain the annotation dataset generated based on the training image and the annotation operation, wherein the annotation operation is performed according to the annotation task; Step S400: Train the labeled dataset to generate the resulting model; Step S100 specifically includes: Step S110: In response to the user's upload operation, receive the training image set based on the Spring Boot MultipartFile class; Step S120: Parse the training image set to obtain training images; Step S200 specifically includes: Step S210: Determine whether the training image contains historical annotation information; Step S220: If the determination is negative, classify the training images according to the preset classification label group; Step S230: Create a labeling task based on the classification label group; Step S240: If the determination is yes, then perform reverse parsing on the training image to obtain historical annotation information; Step S241: Perform reverse parsing on the training image based on OpenCV to obtain historical annotation information, wherein the historical annotation information includes classification label information and segmentation coordinate information; Step S242: Store the historical annotation information in a MySQL database; Step S250: Create a re-annotation task based on the historical annotation information; Step S300 specifically includes: Step S310: In response to the user's annotation operation on the training images, obtain the annotation information generated according to the annotation operation; the user performs annotation operation on the training images according to the annotation task, the annotation task is divided into classification annotation task and segmentation annotation task, the classification annotation task is to assign corresponding label information to each training image and generate a classification image; the segmentation annotation task is to use a pre-built Vue front-end framework to perform custom pattern selection segmentation annotation on multiple small objects in each training image, and generate a segmentation image after specifying the label information corresponding to the current task; Step S320: Using the Java-based Polygon tool, fill the training image with the annotation information to generate an annotation dataset; fill the original image with the coordinate information stored in the MySQL database to generate SegmentationClassPNG and SegmentationObjectPNG. Step S400 specifically includes: Step S410: Generate a training task based on the labeled dataset; Step S420: Run the Docker training model on a preset remote training server and generate the result model according to the training task; Step S421: Upload the labeled dataset to the training directory using JSch and Docker; Step S422: Run Docker to train the model and generate the result model based on the training task; Step S423: Send the result model to an external display terminal, wherein the external display terminal is used to display the result model.

2. The picture labeling training method of claim 1, wherein, Also includes: Step S500: Based on Alibaba OSS files, store the training images, the labeled dataset, and the result model in the cloud.

3. An image annotation training device, characterized in that, The device includes: The training image acquisition module is used to acquire training images in response to user upload operations, specifically including: In response to the user's upload operation, the MultipartFile class based on Spring Boot receives the training image set; The training image set is parsed to obtain training images; The annotation task generation module is used to classify the training images according to a preset classification label group and generate corresponding annotation tasks. The classification label group is preset and specifically includes: Determine whether the training images contain historical annotation information; If the determination is negative, the training images are classified according to the preset classification label group; Create a labeling task based on the aforementioned category label group; If the determination is yes, then the training images are reverse-analyzed to obtain historical annotation information; The training images are reverse-parsed using OpenCV to obtain historical annotation information, which includes classification label information and segmentation coordinate information. Store the historical annotation information in a MySQL database; Create a re-annotation task based on the historical annotation information; The annotation data generation module is used to respond to the user's annotation operation on the training images, and obtain an annotation dataset generated based on the training images and the annotation operation, wherein the annotation operation is performed according to the annotation task, specifically including: In response to the user's annotation operation on the training images, the annotation information generated according to the annotation operation is obtained; the user performs annotation operation on the training images according to the annotation task, which is divided into classification annotation task and segmentation annotation task. The classification annotation task is to assign corresponding label information to each training image and generate a classification image; the segmentation annotation task is to use a pre-built Vue front-end framework to perform custom pattern selection segmentation annotation on multiple small objects in each training image, and generate a segmentation image after specifying the label information corresponding to the current task. The Java-based Polygon tool is used to fill in the training images according to the annotation information to generate an annotated dataset; the original image is filled according to the coordinate information stored in the MySQL database to generate SegmentationClassPNG and SegmentationObjectPNG. The result model generation module is used to train the labeled dataset to generate a result model, specifically including: Generate training tasks based on the labeled dataset; The Docker training model is run on a preset remote training server, and the resulting model is generated according to the training task. Upload the labeled dataset to the training directory using JSch and Docker; Run the Docker training model and generate the resulting model based on the training task. The resulting model is sent to an external display terminal, wherein the external display terminal is used to display the resulting model.

Images