Automatic labeling method and device for data expansion and computer readable storage medium
By iteratively updating weight parameters and data annotations in the yolov10 model, the problem of sample imbalance and automatic annotation accuracy is solved, and efficient and accurate data annotation and model training is achieved, which is suitable for large-scale data processing.
Patent Information
- Application Number
- CN202411872454.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-18
- Publication Date
- 2025-03-14
AI Technical Summary
There are problems in the prior art that sample imbalances, automatic labeling techniques rely on predefined algorithms and rules, and limited accuracy and stability of detection and identification when dealing with small-sized target objects or complex scenarios.
By obtaining the original data and inputting it into the yolov10 model for training, the first weight parameter is obtained; the first expanded data is obtained for object detection, the first annotation is obtained, and then the optimization process is performed and input it into the yolov10 model for training, and the second weight parameter is obtained; the second weight parameter is continuously iteratively updated to generate the target yolov10 model.
It realizes processing large-scale data sets and generates more and more accurate labeled data, improving model training effect and labeling accuracy, and adapting to the needs of modern big data processing.
Smart Images
Figure CN119625431A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present invention relate to the field of image learning technology, and specifically to an automatic annotation method, device and computer-readable storage medium for data expansion. Background Art
[0002] In the era of information explosion, label classification has become an important means of managing and retrieving information. However, for ordinary users, label classification is often a tedious, complex and error-prone task. The emergence of automatic labeling technology will greatly simplify the difficulty and complexity of label classification and improve user experience and efficiency.
[0003] In order to solve the problem that manual labeling of data sets is very tedious in the YOLO series of algorithms, the existing mainstream methods are to achieve automatic labeling through the following methods: (1) data enhancement; (2) Poisson reconstruction densification; (3) semantic labeling using NN algorithm. The above solutions have the following technical problems: The problem of sample imbalance. That is, the number of data samples in some categories far exceeds that of other categories, which may cause the training model to favor categories with higher frequencies and ignore important categories with lower frequencies. This will reduce the accuracy of the model on minority categories and affect the overall detection effect.
[0004] Automatic labeling technology relies on pre-defined algorithms and rules to generate labels. These algorithms may not be able to capture the complex information and context in the image, resulting in errors or inconsistencies in the generated labels. Especially when dealing with unstructured data or complex backgrounds, the accuracy and consistency of automatic labeling often cannot meet actual needs, requiring manual intervention and adjustment.
[0005] When dealing with small-sized objects or complex scenes, the accuracy and stability of detection and recognition are limited. For example, for multiple overlapping objects in a dense scene, automatic labeling technology may not be able to accurately segment and identify each object, thus affecting the final labeling quality and algorithm performance. Summary of the invention
[0006] In view of the above problems, the examples of the present invention provide an automatic labeling method, device and computer-readable storage medium for data expansion, which are used to solve the problems existing in the prior art, such as sample imbalance, automatic labeling technology relying on predefined algorithms and rules to generate labels, and limited accuracy and stability of detection and recognition when processing small-sized target objects or complex scenes.
[0007] According to one aspect of an example of the present invention, there is provided an automatic annotation method for data expansion, the method comprising: Obtaining raw data, and inputting the raw data into a yolov10 model for training to obtain a first weight parameter; Acquire first extended data and perform target detection on the first extended data to obtain a first annotation, perform a first optimization process on the first extended data and the first annotation, and input the first annotation into a yolov10 model for training to obtain a second weight parameter; The second weight parameter is continuously iterated and updated to generate a target yolov10 model.
[0008] In some optional embodiments, the original data includes a plurality of original pictures, and the original pictures have original annotations.
[0009] In some optional ways, the inputting the raw data into the yolov10 model for training to obtain the first weight parameter specifically includes: According to the original image and the original annotation, a first detection threshold is set; By optimizing the yolov10 model and loss function, the position, size and shape of the target detection box are adjusted so that the target detection box corresponds to the original annotation; When the original image is input into the yolov10 model to obtain the original annotation, the weight parameter of the yolov10 model is the first weight parameter.
[0010] In some optional embodiments, the optimizing yolov10 model and loss function specifically includes: Add attention mechanism to yolov10 model; and / or, Perform lightweight convolution on the yolov10 model.
[0011] In some optional embodiments, the first extended data is obtained by processing original data through the parser.add_argument function.
[0012] In some optional methods, obtaining the first extended data and performing target detection on the first extended data to obtain the first annotation, performing the first optimization processing on the first extended data and the first annotation and inputting them into the yolov10 model for training to obtain the second weight parameter, specifically includes: The first expanded data yolov10 model is used to perform target detection according to a first weight parameter to obtain a first annotation, wherein the first annotation includes an annotation position, an annotation category, and an annotation confidence; Performing correction processing on the first annotation to detect and correct the erroneous first annotation, wherein the correction processing includes false detection processing, wrong detection processing and missed detection processing; The first expanded data and the first annotation are input into the yolov10 model for training and prediction to generate the second weight parameter.
[0013] In some optional manners, after obtaining the second weight parameter, the method further includes: The second extended data is obtained and target detection is performed on the first extended data to obtain a second annotation. The second extended data and the second annotation are subjected to a first optimization process and then input into the yolov10 model for training to obtain a third weight parameter.
[0014] In some optional embodiments, the continuously iteratively updating the second weight parameter to generate a target yolov10 model specifically includes: Each time the target data to be detected is detected through the yolov10 model, the second weight parameter is adjusted and iterated in real time in combination with the correction process, and the target yolov10 model is updated.
[0015] According to another aspect of an embodiment of the present invention, there is provided an automatic annotation device for data expansion, the device comprising: A model training module is used to obtain raw data and input the raw data into the yolov10 model for training to obtain a first weight parameter; A parameter updating module, used for acquiring first extended data and performing target detection on the first extended data to obtain a first annotation, performing a first optimization process on the first extended data and the first annotation and inputting the first annotation into a yolov10 model for training to obtain a second weight parameter; And an iteration module, used for continuously iteratively updating the second weight parameter to generate a target yolov10 model.
[0016] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided, wherein at least one executable instruction is stored in the storage medium, and when the executable instruction is executed on the automatic labeling device for data expansion as described above, the automatic labeling device for data expansion performs the operations of the automatic labeling method for data expansion as described above.
[0017] The present invention provides an automatic annotation method, device and computer-readable storage medium for data expansion, and its beneficial effects are: the example of the present invention obtains the original data and inputs the original data into the yolov10 model for training to obtain a first weight parameter; obtains the first expanded data and performs target detection on the first expanded data to obtain a first annotation, and inputs the first expanded data and the first annotation into the yolov10 model for training after the first optimization process to obtain a second weight parameter; continuously iterates and updates the second weight parameter to generate a target yolov10 model. The present invention can not only process large-scale data sets and generate more and more accurate labeled data through an automated data expansion method. As the amount of data increases, the training effect of the model and the accuracy of the annotation are also improved accordingly. In addition, a multi-round iterative training method is adopted, combined with manual optimization and automatic detection, to continuously optimize the model weights and data labels to ensure the accuracy of the annotation and the quality of the data. At the same time, by utilizing the high speed and accuracy of the YOLOv10 algorithm, a large amount of data can be processed in a relatively short time, which meets the needs of modern big data processing.
[0018] The above description is only an overview of the technical solution of the example of the present invention. In order to more clearly understand the technical means of the embodiment of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the example of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are specifically listed below. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] The accompanying drawings are only used to illustrate the embodiments and are not to be considered as limiting the present invention. In addition, the same reference symbols are used to represent the same components throughout the accompanying drawings. In the accompanying drawings: Figure 1 A schematic flow chart of a first embodiment of an automatic annotation method for data expansion provided by the present invention is shown; Figure 2 A schematic diagram of a process of inputting the raw data into a yolov10 model for training to obtain a first weight parameter is shown; Figure 3 A schematic diagram of a process for obtaining a second weight parameter provided by the present invention is shown; Figure 4 A structural diagram of a first embodiment of an automatic annotation device for data expansion provided by the present invention is shown; Figure 5 A flowchart showing a specific example of the automatic annotation method for data expansion provided by the present invention; Figure 6 A data and annotation schematic diagram of a specific case of the automatic annotation method for data expansion provided by the present invention is shown. DETAILED DESCRIPTION
[0020] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein.
[0021] Figure 1 An embodiment of an automatic annotation method for data expansion of the present invention is shown. The present invention is applicable to the field of smart agriculture, and can also be widely used in fields that require large-scale data processing and annotation, such as medical image analysis and industrial detection. In the field of big data processing, the present invention can support real-time monitoring and decision-making during data expansion, and optimize resource utilization and management. In the field of image analysis and industrial detection, it can help quickly identify and analyze key information in images, and support key tasks such as big data diagnosis and maintenance. This method includes: 110, obtaining original data, and inputting the original data into the yolov10 model for training to obtain a first weight parameter; in step 110, the original data includes multiple original pictures, and the original pictures have original annotations. Inputting the original data into the yolov10 model for training to obtain a first weight parameter specifically includes: setting a first detection threshold according to the original picture and the original annotation; adjusting the position, size and shape of the target detection box by optimizing the yolov10 model and the loss function so that the target detection box corresponds to the original annotation; when the original picture is input into the yolov10 model to obtain the original annotation, the weight parameter of the yolov10 model is the first weight parameter.
[0022] 120, obtain the first extended data and perform target detection on the first extended data to obtain the first annotation, perform the first optimization processing on the first extended data and the first annotation and input them into the yolov10 model for training to obtain the second weight parameter; in step 120, the first extended data is obtained by processing the original data through the parser.add_argument function. Obtain the first extended data and perform target detection on the first extended data to obtain the first annotation, perform the first optimization processing on the first extended data and the first annotation and input them into the yolov10 model for training to obtain the second weight parameter, specifically including: perform target detection on the first extended data yolov10 model according to the first weight parameter to obtain the first annotation, wherein the first annotation includes the annotation position, the annotation category and the annotation confidence; perform correction processing on the first annotation to detect and correct the erroneous first annotation, wherein the correction processing includes false detection processing, wrong detection processing and missed detection processing; input the first extended data and the first annotation into the yolov10 model for training and prediction to generate the second weight parameter.
[0023] 130, continuously iteratively update the second weight parameter to generate a target yolov10 model. In step 130, continuously iteratively update the second weight parameter to generate a target yolov10 model, specifically including: each time the target data to be detected is detected by the yolov10 model, combined with the correction processing, the second weight parameter is adjusted and iterated in real time, and the target yolov10 model is updated.
[0024] The present invention can not only process large-scale data sets and generate more and more accurate labeled data through an automated data expansion method. As the amount of data increases, the training effect of the model and the accuracy of the annotations are also improved accordingly. In addition, a multi-round iterative training method is adopted, combined with manual optimization and automatic detection, to continuously optimize the model weights and data labels to ensure the accuracy of the annotations and the quality of the data. At the same time, the high speed and accuracy of the YOLOv10 algorithm can be used to process a large amount of data in a relatively short time, which meets the needs of modern big data processing.
[0025] In some optional ways, the original data includes multiple original pictures, and the original pictures have original annotations. In this embodiment, the original annotations of the original pictures can be obtained by manual fine annotation or machine fine standard, and each original picture can include original annotations. These annotations include accurately locating the target and assigning the correct category label.
[0026] For some alternatives, see Figure 2 , input the original data into the yolov10 model for training, and obtain the first weight parameter, which includes: 210, according to the original image and the original annotation, set the first detection threshold; in step 210, after obtaining the original image and the original annotation, set the first detection threshold according to the location, category and other information of the original annotation or the characteristics of the original annotation itself, so that the original annotation can be identified on the original image through the first detection threshold. According to the specific requirements of the target detection task and the characteristics of the data, adjust the threshold to ensure that the algorithm can accurately identify the target while minimizing the false detection rate and the missed detection rate.
[0027] 220, by optimizing the yolov10 model and the loss function, the position, size and shape of the target detection frame are adjusted so that the target detection frame corresponds to the original annotation; in step 220, after the original annotation is identified, the original annotation frame can be selected within the target detection frame by setting the position, size and shape of the target detection frame, so that the target detection frame corresponds to the position, size and shape of the original annotation. The position, size and shape of the target detection frame are adjusted to match the actual position of the target object as much as possible, and the network structure and loss function are optimized, such as adding an attention mechanism, introducing lightweight convolution, etc. The parameters of each detection frame are dynamically adjusted during the training process so that it can adapt to targets of different sizes and postures, thereby improving the generalization ability and adaptability of the algorithm.
[0028] 230. When the original image is input into the yolov10 model to obtain the original annotation, the weight parameter of the yolov10 model is the first weight parameter. In step 230, after setting the detection threshold and adjusting the parameters of the target detection box, the yolov10 model can normally recognize the original annotation, and the relevant parameters of the yolov10 model can be defined as the original parameters. By obtaining a preliminary optimal model weight W1 in the initial stage, the foundation for our subsequent steps is laid. These weights are obtained after training and adjustment on a small number of manually annotated data sets, and have certain generalization ability and prediction accuracy. Subsequent steps will be based on these preliminary optimized weights, and the performance of the model will be further improved through iterative training weights and data enhancement.
[0029] In some optional ways, the yolov10 model and loss function are optimized, specifically including: adding an attention mechanism to the yolov10 model; and / or, performing lightweight convolution on the yolov10 model. In this embodiment, adding an attention mechanism to the YOLOv10 model can be achieved by a variety of methods, including adding Triplet Attention, Cascaded GroupAttention (CGA), iRMB (Inverse Residual Moving Module), and Spatial Group-wise Enhance (SGE) mechanisms. For lightweight convolution on the yolov10 model, the model can be quantized using the QAT (Quantization Aware Training) method. The following is a simplified example showing how to lightweight the convolution layer of a model in PyTorch: First, a quantization module quantize is created, then all layers of the YOLOv10 model are traversed, the convolution layers are found, and the quantization configurations are set for these convolution layers. Then we prepare and convert these convolution layers to lightweight convolution layers. By adding attention mechanisms and introducing lightweight convolutions, the parameters of each detection box are dynamically adjusted during the training process so that it can adapt to targets of different sizes and postures, thereby improving the generalization and adaptability of the algorithm.
[0030] In some optional ways, the first extended data is obtained by processing the original data through the parser.add_argument function. In this embodiment, the following code is added to the parameter setting: parser.add_argument('--save-txt', action='store_true', default='True' help='save results to *.txt') so that the algorithm saves the label of the detected target while detecting the image, and then hides the detected image target frame and uses it as the existing data to expand the image. Specifically, if save_img or save_crop or view_img:# Add bbox to image c = int(cls) # integer class label = None if hide_labels else (names[c] if hide_conf else f'{names[c]} {conf:.2f}') annotator.box_label(xyxy, label, color=colors(c, True)) if save_crop: save_one_box(xyxy, imc, file=save_dir / 'crops' / names[c] / f'{p.stem}.jpg', BGR=True) Through the above steps, a new round of iterative data sets and labels can be obtained.
[0031] In some optional methods, the first extended data is obtained and target detection is performed on the first extended data to obtain the first annotation. The first extended data and the first annotation are optimized for the first time and then input into the yolov10 model for training to obtain the second weight parameter. Using the first weight parameter obtained in step 110, we perform target detection on the image of the extended data set. The detected data image is then labeled and manually optimized or machine optimized to obtain the optimized first extended data and the first annotation. These optimized data are again sent to the YOLOv10 network model for prediction. This process is iteratively optimized to obtain a more accurate second weight parameter. See Figure 3 , the specific process is as follows: 310, the first extended data YOLOv10 model is used to perform target detection according to the first weight parameter to obtain a first annotation, wherein the first annotation includes the annotation position, the annotation category and the annotation confidence; in step 310, the first extended data is sent to the optimized YOLOv10 algorithm model for target detection. The first extended data can be obtained by converting the original data or newly acquired so that the first extended data includes a wider and more diverse range of scenes and targets. The result of the target detection will include information such as the detected target position, category and confidence of the first extended data.
[0032] 320, perform correction processing on the first annotation, detect and correct the erroneous first annotation, wherein the correction processing includes false detection processing, wrong detection processing and missed detection processing; in step 320, the correction processing can be performed by manual screening or machine screening through the first annotation. The machine screening method can provide screening conditions for the first annotation by setting a threshold and a target detection frame so that the first annotation is detected. After the first annotation is detected, the first annotation can be modified manually so that the first annotation corresponds to the image target.
[0033] 330, input the first extended data and the first label into the YOLOv10 model for training and prediction, and generate a second weight parameter. In step 330, after manual or machine optimization, the optimized first extended data and the first label are obtained. These data correspond to more accurate target detection results after adjustment and correction. Next, we will send the optimized first extended data and the first label into the YOLOv10 network model again for training and prediction. Through the iterative training process, the second weight parameter will be further optimized and adjusted to adapt to the new data set and higher annotation accuracy requirements.
[0034] In some optional methods, after obtaining the second weight parameter, it also includes: obtaining the second expanded data and performing target detection on the first expanded data, obtaining the second annotation, performing the first optimization processing on the second expanded data and the second annotation, and inputting them into the yolov10 model for training to obtain the third weight parameter. In this embodiment, the second weight parameter obtained in step 120 is used to perform data expansion on the first expanded data by performing image horizontal / vertical flipping, random extraction, scale transformation and rotation. Specific expansion methods may include: expansion and scaling cv2.resize(). Translation cv2.warpAffine(). Rotation first uses the cv2.getRotationMatrix2D function to build a rotation matrix, and then uses the cv2.warpAffine() function to move the object. Radial transformation: Get three points from the input image and their corresponding positions in the output image, use cv2. getAffineTransform to first build a 2x3 transformation matrix, and finally the matrix will be passed to the cv2.warpAffine function, etc. By expanding the first expanded data to generate the second expanded data, not only the number of training samples can be expanded, but also the diversity of training samples can be increased. On the one hand, overfitting can be avoided, and on the other hand, model performance will be improved.
[0035] Detect objects on the second augmented data. This step aims to further improve the accuracy and adaptability of the model on new datasets. The detected objects will save the label information of their detection boxes again, including location, category and confidence. Correct the second annotation again. This includes correcting possible false detections or missed detections, and adjusting the labels to ensure that they accurately reflect the real situation in the picture. The manual optimization process is a key step to ensure the quality of data annotation, especially when facing complex scenes or small objects. After further manual optimization, we get the optimized second augmented data and second annotation. These data should be more precise and accurate than the first augmented data and first annotation obtained in the previous step. Finally, the optimized second augmented data and second annotation are sent to the YOLOv10 network model again for training and prediction. This iterative optimization process will further adjust the generated third weight parameters so that it can better adapt to the characteristics and annotation requirements of the augmented dataset.
[0036] In some optional ways, the second weight parameter is continuously iterated and updated to generate a target yolov10 model, specifically including: each time the target data target to be detected is detected through the yolov10 model, combined with the correction process, the second weight parameter is adjusted and iterated in real time, and the target yolov10 model is updated. In this embodiment, as time goes by and the application needs increase, we will continue to expand the data set. The new data will contain more scenes, richer target types and more diverse environmental conditions to ensure the robustness and versatility of the model in various real-life scenarios. The expansion of the data set is the key to keeping the model training and optimization going, providing the system with a more comprehensive data perspective, thereby improving the coverage and accuracy of automatic annotation.
[0037] Example 2 Figure 2 A first embodiment of an automatic annotation device for data expansion is shown. The device is based on embodiment 1 and specifically includes: The model training module 410 is used to execute step 110 of embodiment 1. The model training module 410 specifically includes: the original data includes multiple original pictures, and the original pictures have original annotations. The original data is input into the yolov10 model for training to obtain the first weight parameter, which specifically includes: setting the first detection threshold according to the original picture and the original annotation; adjusting the position, size and shape of the target detection frame by optimizing the yolov10 model and the loss function so that the target detection frame corresponds to the original annotation; when the original picture is input into the yolov10 model to obtain the original annotation, the weight parameter of the yolov10 model is the first weight parameter. ; The parameter updating module 420 is used to execute step 120 of Example 1. The parameter updating module 420 specifically includes the first extended data obtained by processing the original data through the parser.add_argument function. The first extended data is obtained and the target detection is performed on the first extended data to obtain the first annotation. The first extended data and the first annotation are input into the yolov10 model for training after the first optimization process to obtain the second weight parameter, which specifically includes: the first extended data yolov10 model is used to perform target detection according to the first weight parameter to obtain the first annotation, wherein the first annotation includes the annotation position, the annotation category and the annotation confidence; the first annotation is corrected to detect and correct the wrong first annotation, wherein the correction process includes false detection process, wrong detection process and missed detection process; the first extended data and the first annotation are input into the yolov10 model for training and prediction to generate the second weight parameter.
[0038] And an iteration module 430 is used to execute step 130 of Example 1, continuously iteratively update the second weight parameter, and generate a target yolov10 model, specifically including: each time the target data to be detected is detected through the yolov10 model, combined with the correction processing, the second weight parameter is adjusted and iterated in real time, and the target yolov10 model is updated.
[0039] The present invention provides an automatic annotation device for data expansion. The present invention example obtains original data and inputs the original data into a YOLOv10 model for training to obtain a first weight parameter; obtains first expanded data and performs target detection on the first expanded data to obtain a first annotation, and inputs the first expanded data and the first annotation into a YOLOv10 model for training after first optimization processing to obtain a second weight parameter; and continuously iterates and updates the second weight parameter to generate a target YOLOv10 model. The present invention can not only process large-scale data sets and generate more and more accurate labeled data through an automated data expansion method. As the amount of data increases, the training effect of the model and the accuracy of the annotation are also improved accordingly. In addition, a multi-round iterative training method is adopted, combined with manual optimization and automatic detection, to continuously optimize the model weights and data labels to ensure the accuracy of the annotation and the quality of the data. At the same time, by utilizing the high speed and accuracy of the YOLOv10 algorithm, a large amount of data can be processed in a relatively short time, which meets the needs of modern big data processing.
[0040] Embodiment 3: According to another aspect of the embodiments of the present invention, a computer-readable storage medium is provided. Based on Embodiment 1 and Embodiment 2, the storage medium stores at least one executable instruction. When the executable instruction is executed on the automatic annotation device for data expansion as described above, the automatic annotation device for data expansion performs the operations of the automatic annotation method for data expansion as described above. In Embodiment 3 of the present application, the following steps are performed: 110, obtaining original data, and inputting the original data into the yolov10 model for training to obtain a first weight parameter; 120, obtaining first extended data and performing target detection on the first extended data to obtain a first annotation, performing a first optimization process on the first extended data and the first annotation and inputting the first annotation into a yolov10 model for training to obtain a second weight parameter; 140, continuously iteratively update the second weight parameter to generate a target yolov10 model.
[0041] In a specific example, the target yolov10 model can be found in Figure 5 The process shown in the figure is used for generation and optimization. When generating the target yolov10 model, input Figure 6 After the picture to be detected is shown, you can Figure 6 Detect the target annotation box in Figure 6 The black frame part.
[0042] Compared with the prior art, cost-effectiveness: The present invention significantly reduces the cost of manual annotation. Traditional data annotation methods require a large amount of human resources for manual annotation, which is costly and time-consuming. By automating the annotation process, we save a lot of labor costs and can process large-scale data sets in a shorter time. Efficiency: Our method improves the speed of annotation and the efficiency of data processing. Combined with the high speed and accuracy of the YOLOv10 algorithm, we can quickly generate a large amount of labeled data. This is particularly important for application scenarios that require processing large-scale data and rapid response, such as real-time data analysis and decision support in smart agriculture. Annotation accuracy: The automatic annotation method based on big data learning can significantly improve the accuracy of annotation as the amount of data increases. In the process of continuous iterative training and data set expansion, the model learns more data features and annotation patterns, thereby improving the accuracy of target detection. This is particularly important for fields such as autonomous driving and intelligent monitoring that require extremely high data accuracy. Application prospects: The present invention is not only applicable to the field of smart agriculture, but can also be widely used in fields such as medical image analysis and industrial detection that require large-scale data processing and annotation. Its simple and efficient operation process gives it broad market application prospects in both industry and academia.
[0043] In the field of big data processing, the present invention can support real-time monitoring and decision-making during data expansion, optimize resource utilization and management. In the field of image analysis and industrial inspection, it can help quickly identify and analyze key information in images, and support key tasks such as big data diagnosis and maintenance.
[0044] The algorithms or displays provided herein are not inherently related to any particular computer, virtual system or other device. In addition, the embodiments of the present invention are not directed to any particular programming language.
[0045] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. Similarly, in order to simplify the present invention and help understand one or more of the various inventive aspects, in the above description of exemplary embodiments of the present invention, the various features of the embodiments of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. Wherein, the claims that follow the specific embodiment are hereby expressly incorporated into the specific embodiment, wherein each claim itself is a separate embodiment of the present invention.
[0046] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and further may be divided into a plurality of submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive.
[0047] It should be noted that the above embodiments illustrate the present invention rather than limit it, and that those skilled in the art may devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets shall not be construed as a limitation on the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention may be implemented by means of hardware comprising a number of different elements and by means of a suitably programmed computer. In a unit claim enumerating a number of devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order. These words may be interpreted as names. The steps in the above embodiments, unless otherwise specified, should not be understood as limitations on the order of execution.
Claims
1. An automatic annotation method for data expansion, characterized in that: The method comprises: Obtaining raw data, and inputting the raw data into a yolov10 model for training to obtain a first weight parameter; Acquire first extended data and perform target detection on the first extended data to obtain a first annotation, perform a first optimization process on the first extended data and the first annotation, and then input the first annotation into a yolov10 model for training to obtain a second weight parameter; The second weight parameter is continuously iterated and updated to generate a target yolov10 model.
2. The automatic annotation method for data expansion according to claim 1, characterized in that: The original data includes a plurality of original pictures, and the original pictures have original annotations.
3. The automatic annotation method for data expansion according to claim 2, characterized in that: The raw data is input into the yolov10 model for training to obtain the first weight parameter, specifically including: According to the original image and the original annotation, a first detection threshold is set; By optimizing the yolov10 model and loss function, the position, size and shape of the target detection box are adjusted so that the target detection box corresponds to the original annotation; When the original image is input into the yolov10 model to obtain the original annotation, the weight parameter of the yolov10 model is the first weight parameter.
4. The automatic annotation method for data expansion according to claim 3, characterized in that: The optimization of the yolov10 model and loss function specifically includes: Add attention mechanism to yolov10 model; and / or, Perform lightweight convolution on the yolov10 model.
5. The automatic annotation method for data expansion according to claim 1, characterized in that: The first extended data is obtained by processing the original data through the parser.add_argument function.
6. The automatic annotation method for data expansion according to claim 5, characterized in that: Obtaining the first extended data and performing target detection on the first extended data to obtain the first annotation, performing the first optimization processing on the first extended data and the first annotation and inputting them into the yolov10 model for training to obtain the second weight parameter, specifically including: The first expanded data yolov10 model is used to perform target detection according to a first weight parameter to obtain a first annotation, wherein the first annotation includes an annotation position, an annotation category, and an annotation confidence; Performing correction processing on the first annotation to detect and correct the erroneous first annotation, wherein the correction processing includes false detection processing, wrong detection processing and missed detection processing; The first expanded data and the first annotation are input into the yolov10 model for training and prediction to generate the second weight parameter.
7. The automatic annotation method for data expansion according to claim 6, characterized in that: After acquiring the second weight parameter, the method further includes: The second extended data is obtained and target detection is performed on the first extended data to obtain a second annotation. After the second extended data and the second annotation are subjected to a first optimization process, the second annotation and the second annotation are input into the yolov10 model for training to obtain a third weight parameter.
8. The automatic annotation method for data expansion according to claim 6 or 7, characterized in that: The continuously iteratively updating the second weight parameter to generate a target yolov10 model specifically includes: Each time the target data to be detected is detected through the yolov10 model, the second weight parameter is adjusted and iterated in real time in combination with the correction process, and the target yolov10 model is updated.
9. An automatic annotation device for data expansion, characterized in that: The device comprises: A model training module is used to obtain raw data and input the raw data into the yolov10 model for training to obtain a first weight parameter; A parameter updating module, used for acquiring first extended data and performing target detection on the first extended data to obtain a first annotation, performing a first optimization process on the first extended data and the first annotation and inputting the first annotation into a yolov10 model for training to obtain a second weight parameter; And an iteration module, used for continuously iteratively updating the second weight parameter to generate a target yolov10 model.
10. A computer-readable storage medium, characterized in that: The storage medium stores at least one executable instruction. When the executable instruction is executed on the automatic annotation device for data expansion as described in claim 9, the automatic annotation device for data expansion performs the operation of the automatic annotation method for data expansion as described in any one of claims 1-8.