Efficient data labeling method and system, electronic equipment and storage medium

Abstract

The invention discloses an efficient data annotation method and system, electronic equipment and a storage medium, and relates to the technical field of image data annotation, and the method comprises the following steps: carrying out the frame selection of a sample image through an image annotation platform, and carrying out the manual image annotation, and obtaining an annotation sample; constructing an AI big data annotation model, and training the AI big data annotation model by analyzing the contour features of the annotation samples; after the training is completed, performing efficient and automatic annotation on the image through an AI big data annotation model; the method is used for solving the problems that the intelligence of image data annotation is insufficient and the data basis of AI data annotation is insufficient due to the fact that an existing image data annotation technology depends too much on manual annotation and image features are difficult to unify.

Description

Technical Field

[0001] The present invention relates to the technical field of image data annotation, and in particular to an efficient data annotation method, system, electronic device and storage medium. Background Art

[0002] Image data labeling technology refers to the process of adding labels, annotations or key information to image data by using artificial intelligence and machine learning technologies to identify, classify and locate image content. These labels and annotations can help computer systems understand and interpret image content, thereby realizing functions such as automatic identification, search and classification of image content. Image data labeling technology has a wide range of applications in computer vision, image recognition, natural language processing and other fields, and is one of the important means to achieve intelligent image processing and analysis.

[0003] Existing image data annotation technology usually extracts entities in the image by computer and then annotates the entities in the image in the form of manual annotation. Existing image data annotation technology relies too much on manual annotation and lacks automatic and intelligent image annotation technology. At the same time, since image annotation technology needs to automatically annotate different entities in multiple fields or targets in surveillance videos, and the features between different types and individuals are difficult to unify, the extracted image features are not representative enough and cannot effectively provide a data basis for AI data annotation. For example, in a patent application with publication number CN118690194A, a data annotation method, device, equipment, medium and program product are disclosed. The solution still has a large number of manual annotation steps in the application process and relies too much on human resources, resulting in the data annotation method being not intelligent enough. Existing image data annotation technology also has the problems of relying too much on manual annotation and difficulty in unifying image features, resulting in insufficient intelligence of image data annotation and insufficient data basis for AI data annotation. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems in the prior art to a certain extent. A sample image is framed and manually annotated through an image annotation platform to obtain an annotated sample, and then an AI big data annotation model is constructed to extract the contour of the framed target in the annotated sample to obtain the target contour, and a feature distribution map is constructed based on the target contour. The feature distribution map is then analyzed to extract the contour features of the framed target. After the training is completed, the image is efficiently and automatically annotated through the AI ​​big data annotation model to solve the problems that the existing image data annotation technology still relies too much on manual annotation and the image features are difficult to unify, resulting in insufficient intelligence of the image data annotation and insufficient data basis for AI data annotation.

[0005] To achieve the above objectives, in a first aspect, the present application provides an efficient data labeling method, comprising the following steps: The sample image is framed and manually annotated on the image through the image annotation platform to obtain annotated samples; Build an AI big data annotation model and train it by analyzing the contour features of the annotated samples; After training is completed, the images are efficiently and automatically annotated using the AI ​​big data annotation model.

[0006] Furthermore, the sample image is framed and manually annotated on the image through the image annotation platform to obtain the annotated sample, which includes the following sub-steps: Connect data with the image annotation platform; The sample image is framed by an image annotation platform to obtain a framed target, where the framed target is a rectangular area; Annotate the image of the selected target through internal self-operated annotation, third-party annotation, and crowdsourcing annotation to obtain the annotation content; For any selected target, the annotation content with the highest percentage is selected as the final annotation. If the three annotation contents are different, the internal self-operated annotation, third-party annotation, and crowdsourcing annotation need to negotiate to obtain the final annotation; The selected target and the corresponding final annotation are the labeled samples.

[0007] Furthermore, constructing an AI big data annotation model and training the AI ​​big data annotation model by analyzing the contour features of the annotation samples includes the following sub-steps: Build an AI big data annotation model, extract the contour of the box-selected target in the annotated sample, obtain the target contour, and build a feature distribution map based on the target contour; Analyze the feature distribution map and extract the contour features of the selected target.

[0008] Furthermore, an AI big data annotation model is constructed to extract the contour of the box-selected target in the annotation sample to obtain the target contour. The feature distribution map is constructed based on the target contour, which includes the following sub-steps: Build AI big data annotation model; Get the boxed target in the labeled sample, extract the contour of the boxed target using the OpenCV contour detection algorithm, and obtain the target contour; A plane rectangular coordinate system is established with the lower left vertex of the selected object as the origin, the number of pixels in the horizontal direction as the X-axis, and the number of pixels in the vertical direction as the Y-axis, and is named a feature distribution map. The left side and bottom side of the selected object overlap with the Y-axis and the X-axis respectively. The coordinates of each pixel point of the selected target in the feature distribution map are obtained and marked as pixel coordinates. The pixel coordinates are expressed as (X, Y), where (X, Y) represents that the pixel point is at the Xth pixel point in the horizontal direction and the Yth pixel point in the vertical direction.

[0009] Furthermore, the feature distribution map is analyzed to extract the contour features of the framed target, including the following sub-steps: Find the pixel coordinates belonging to the target contour, mark them as contour coordinates, mark the X of the contour coordinates as the contour horizontal coordinate, and mark the Y of the contour coordinates as the contour vertical coordinate; The contour abscissas are numbered in ascending order, using the symbol T n Indicates that the contour ordinates are numbered in ascending order, using the symbol L n It means, where n∈Z+ and n is the serial number of T and L; With n as the horizontal axis, T n Establish a plane rectangular coordinate system for the vertical axis, named as the horizontal feature map, and convert T n Enter the horizontal feature map according to n; take n as the horizontal axis, L n Establish a plane rectangular coordinate system for the vertical axis, named as the vertical feature map, and set L n Enter into the longitudinal feature map according to n; The horizontal feature map and the vertical feature map are regressed by linear regression in discrete regression analysis to obtain a horizontal feature function and a vertical feature function respectively; the format of the horizontal feature function is YT=at×n+bt, where YT is T n , at is the slope of the horizontal characteristic function, marked as the horizontal slope, bt is the intercept of the horizontal characteristic function, marked as the horizontal intercept; the format of the longitudinal characteristic function is YL=al×n+bl, where YL is L n , al is the slope of the longitudinal characteristic function, marked as the longitudinal slope, bl is the intercept of the longitudinal characteristic function, marked as the longitudinal intercept; Establish a contour feature data set to record the horizontal feature function, vertical feature function of the selected target and the corresponding final annotation in the annotation sample; Merge the box-selected objects with the same final annotations. After the merging is completed, the final annotations are a first-level directory. The horizontal feature function and the vertical feature function constitute a feature data. Each first-level directory includes several feature data. When a user inputs a new frame selection target into the AI ​​big data annotation model, it is marked as a target to be annotated; the horizontal feature function and the vertical feature function of the target to be annotated are analyzed, and they are named as the horizontal function to be analyzed and the vertical function to be analyzed respectively, and a piece of data to be analyzed is composed of the horizontal function to be analyzed and the vertical function to be analyzed; Find out whether there is feature data in the contour feature data set that is exactly the same as the data to be analyzed. If so, the output data has a signal. If not, the output data does not have a signal. If there is a signal in the output data, the final annotation corresponding to the feature data that is exactly the same as the data to be analyzed is obtained and output; if there is no signal in the output data, the data to be analyzed and the feature data are matched and analyzed through a feature matching algorithm.

[0010] Furthermore, the feature matching algorithm includes the following sub-steps: Re-establish the horizontal feature map and the vertical feature map, and name them as the horizontal comparison map and the vertical comparison map respectively; For any characteristic data, the straight line corresponding to the horizontal characteristic function in the characteristic data is named as the horizontal characteristic straight line, and the straight line corresponding to the vertical characteristic function in the characteristic data is named as the vertical characteristic straight line; the straight line corresponding to the horizontal function to be analyzed is named as the horizontal straight line to be analyzed, and the straight line corresponding to the vertical function to be analyzed is named as the vertical straight line to be analyzed; Enter the horizontal feature straight line and the horizontal straight line to be analyzed into the horizontal comparison diagram, and enter the vertical feature straight line and the vertical straight line to be analyzed into the vertical comparison diagram. When entering each straight line, follow the value range of the number of pixels n in the corresponding frame selection target, that is, convert the straight line into a line segment based on the value range of n in the frame selection target; Rename the horizontal characteristic straight line, the horizontal straight line to be analyzed, the vertical characteristic straight line and the vertical straight line to be analyzed to the horizontal characteristic line segment, the horizontal line segment to be analyzed, the vertical characteristic line segment and the vertical line segment to be analyzed; Connect the left end point of the horizontal characteristic line segment and the horizontal line segment to be analyzed by a straight line, and connect the right end point of the horizontal characteristic line segment and the horizontal line segment to be analyzed, and obtain at most two closed areas after the connection is completed, and obtain the area of ​​the closed area and mark it as the horizontal comparison index; Connect the left endpoints of the longitudinal characteristic straight line and the longitudinal straight line to be analyzed by a straight line, and connect the right endpoints of the longitudinal characteristic straight line and the longitudinal straight line to be analyzed by a straight line, and obtain at most two closed areas after the connection is completed, and obtain the area of ​​the closed area and mark it as the longitudinal comparison index; The horizontal comparison index and the vertical comparison index are added together to obtain the annotation deviation index, the data to be analyzed are analyzed with all the characteristic data in turn to obtain several annotation deviation indexes, and the minimum value among the annotation deviation indexes is found and marked as the minimum deviation index; The feature data corresponding to the minimum deviation index is marked as the target data, and the final annotation corresponding to the target data is output as the annotation result.

[0011] Furthermore, after the training is completed, the efficient automatic labeling of images through the AI ​​big data labeling model includes the following sub-steps: The user extracts the input image through the image annotation platform to obtain the framed target in the input image; Input the selected target into the AI ​​big data annotation model, and the AI ​​big data annotation model will output the annotation results for automatic annotation.

[0012] In a second aspect, the present application provides an efficient data annotation system, including a platform extraction module, an annotation training module and an automatic annotation module; the platform extraction module and the automatic annotation module are respectively connected to the annotation training module data; The platform extraction module is used to select the sample image through the image annotation platform and manually annotate the image to obtain annotated samples; The annotation training module is used to build an AI big data annotation model, and train the AI ​​big data annotation model by analyzing the contour features of the annotation samples; The automatic annotation module is used to efficiently and automatically annotate images through an AI big data annotation model after training is completed.

[0013] In a third aspect, the present application provides an electronic device, comprising a processor and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps in the above method are performed.

[0014] In a fourth aspect, the present application provides a storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps in the above method are performed.

[0015] Beneficial effects of the present invention: The present invention selects a sample image through an image annotation platform and manually annotates the image to obtain annotated samples, then constructs an AI big data annotation model, extracts the contour of the selected target in the annotated sample, and obtains the target contour. The advantage is that the manual annotation method is only used to make annotated samples for training the AI ​​big data annotation model. After the training is completed, intelligent annotation can be performed without manual intervention. At the same time, when training the AI ​​big data annotation model, different entities can be effectively distinguished by extracting and analyzing the contour as a feature of the annotation sample, thereby improving the accuracy and reliability of intelligent image data annotation. The present invention constructs a feature distribution map based on the target contour, and then analyzes the feature distribution map to extract the contour features of the selected target. After training, the image is efficiently and automatically annotated through the AI ​​big data annotation model. The advantage is that when the contour is extracted and analyzed as the feature of the annotation sample, it is difficult to distinguish all entities by conventional comparison and analysis of the contour. At the same time, conventional comparison and analysis usually performs similarity analysis on the contour. This method will also greatly increase the complexity of the analysis process, thereby reducing the efficiency of intelligent image annotation. The present invention can effectively distinguish various entities by analyzing the distribution characteristics of the contour. At the same time, the analysis process is more concise, which improves the accuracy and efficiency of intelligent image data annotation. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a functional block diagram of the system of the present invention; Figure 2 is a sample image of the present invention; Figure 3 Frame selection of the object of the present invention; Figure 4 The target outline of the present invention; Figure 5 is a characteristic distribution diagram of the present invention; Figure 6 A schematic diagram of placing a target profile into a feature distribution map according to the present invention; Figure 7 is a schematic diagram of a lateral characteristic diagram of the present invention; Figure 8 It is a horizontal comparison diagram of the present invention; Fig. 9 is a schematic diagram of a closed area of ​​the present invention; Fig.10 The figure is a flow chart of the steps of the method of the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] Example 1, please refer to Figure 1 As shown, the present application provides an efficient data annotation system, including a platform extraction module, an annotation training module and an automatic annotation module; the platform extraction module and the automatic annotation module are respectively connected to the annotation training module data.

[0019] The platform extraction module is used to select the sample image through the image annotation platform and manually annotate the image to obtain annotated samples; The platform extraction module is configured with platform extraction strategies, which include: Connect data with the image annotation platform; See also Figures 2 to 3 As shown, the sample image is framed by the image annotation platform to obtain a framed target, which is a rectangular area; In practical applications, the image annotation platform is a network platform provided by existing online image annotation service providers. In the image data annotation process, it is first necessary to extract the target to be annotated in the image, that is, to select the target. The process of extracting the selected target can use the algorithm provided by the image annotation platform. This embodiment focuses on the feature analysis of the selected target and uses this as a basis for data annotation of the selected target. Therefore, the extraction process of the selected target is not described. Figure 2 is a sample image, Figure 3 The image annotation platform selects a sample image and obtains one of the selected targets. Figure 3 The black border in the image is the border of the selected object. In subsequent analysis, the processing of the selected object starts with the black border, not the Figure 3 The borders of Figure 3 The reserved interval is to show the specific format of the frame selection target. Each frame selection target has its own frame. When analyzing, only the image within the frame needs to be analyzed. For example, the lower left vertex of the frame selection target is taken as the origin. Figure 3 The lower left vertex of the black border is the origin, not Figure 3 The lower left vertex of the image; Annotate the image of the selected target through internal self-operated annotation, third-party annotation, and crowdsourcing annotation to obtain the annotation content; For any selected target, the annotation content with the highest percentage is selected as the final annotation. If the three annotation contents are different, the internal self-operated annotation, third-party annotation, and crowdsourcing annotation need to negotiate to obtain the final annotation; The selected target and the corresponding final annotation are the labeled samples; In practical applications, most of the existing annotation methods use manual annotation, such as internal self-operated annotation, third-party annotation and crowdsourcing annotation. Figure 3The boxed target in the image is annotated, and the annotated contents are all cars, and the proportion of annotated contents as cars is 100%, and the final annotation is car; there is another situation, for example, the annotation contents given by internal self-operated annotation, third-party annotation and crowdsourcing annotation are car, private car and vehicle respectively, then the internal self-operated annotation, third-party annotation and crowdsourcing annotation need to negotiate to give the final annotation.

[0020] The annotation training module is used to build an AI big data annotation model, and train the AI ​​big data annotation model by analyzing the contour features of the annotation samples; the annotation training module includes a feature distribution construction unit, a feature distribution analysis unit, and a feature matching analysis unit; The feature distribution construction unit is used to construct an AI big data annotation model, extract the contour of the framed target in the annotation sample, obtain the target contour, and construct a feature distribution map based on the target contour; The feature distribution construction unit is configured with a feature distribution construction strategy, which includes: Build AI big data annotation model; See also Figure 4 As shown, the framed target in the labeled sample is obtained, and the contour of the framed target is extracted using the OpenCV contour detection algorithm to obtain the target contour; See also Figure 5 As shown in the figure, a plane rectangular coordinate system is established with the lower left vertex of the selected object as the origin, the number of pixels in the horizontal direction as the X-axis, and the number of pixels in the vertical direction as the Y-axis. It is named the feature distribution map, and the left side and bottom side of the selected object overlap with the Y-axis and the X-axis respectively. Get the coordinates of each pixel of the selected target in the feature distribution map, marked as pixel coordinates, and the pixel coordinates are expressed as (X, Y), where (X, Y) means that the pixel is at the Xth pixel in the horizontal direction and the Yth pixel in the vertical direction; In practical applications, the contour detection algorithm is implemented using the existing OpenCV contour detection technology. The target contour obtained by contour extraction is as follows: Figure 4 As shown in the figure, the constructed feature distribution diagram is as follows Figure 5 As shown in the figure, since X and Y represent the number of pixels in the horizontal and vertical directions of the framed target, and the resolution of the framed target is 473×448, that is, there are 473 pixels in the horizontal direction and 448 pixels in the vertical direction, the maximum value of X is 473 and the maximum value of Y is 448. There are 473×448=211904 pixels in the framed target, that is, 211904 pixel coordinates are obtained; The feature distribution analysis unit is used to analyze the feature distribution map and extract the contour features of the frame-selected target; The feature distribution analysis unit is configured with a feature distribution analysis strategy, which includes: See also Figure 6 As shown, find the pixel coordinates belonging to the target contour, mark them as contour coordinates, mark the X of the contour coordinates as the contour horizontal coordinate, and mark the Y of the contour coordinates as the contour vertical coordinate; The contour abscissas are numbered in ascending order, using the symbol T n Indicates that the contour ordinates are numbered in ascending order, using the symbol L n It means, where n∈Z+ and n is the serial number of T and L; In practical applications, when searching for contour coordinates, the target contour is placed in the feature distribution map, and then the coordinates of the pixels with grayscale values ​​other than 255 are extracted, except for the coordinates of the border, to obtain the contour coordinates, such as Figure 6 As shown in the figure, 10595 contour coordinates are extracted and the coordinates of T1 to T are obtained by numbering. 10595 and L1 to L 10595 , 1≤n≤10595; See also Figure 7 As shown, with n as the horizontal axis, T n Establish a plane rectangular coordinate system for the vertical axis, named as the horizontal feature map, and convert T n Enter the horizontal feature map according to n; take n as the horizontal axis, L n Establish a plane rectangular coordinate system for the vertical axis, named as the vertical feature map, and set L n Enter into the longitudinal feature map according to n; Through the linear regression in discrete regression analysis, the horizontal feature map and the vertical feature map are regressed to obtain the horizontal feature function and the vertical feature function respectively; the format of the horizontal feature function is YT=at×n+bt, where YT is T n , at is the slope of the horizontal characteristic function, marked as the horizontal slope, bt is the intercept of the horizontal characteristic function, marked as the horizontal intercept; the format of the vertical characteristic function is YL=al×n+bl, where YL is L n , al is the slope of the longitudinal characteristic function, marked as the longitudinal slope, bl is the intercept of the longitudinal characteristic function, marked as the longitudinal intercept; In practical applications, since the horizontal feature map and the vertical feature map adopt the same principle and the processing process is exactly the same, this embodiment only shows the horizontal feature map. Since there is a lot of data and it is too dense in the image, this embodiment only uses part of the data to show the distribution format of the coordinate points in the horizontal feature map. The horizontal feature map is constructed as shown in FIG. Figure 7 As shown, T nThe value range is 1 to 473. The horizontal feature map shows the distribution of 10595 contour coordinates on the X-axis in the feature distribution map. Similarly, the vertical feature map shows the distribution of 10595 contour coordinates on the Y-axis in the feature distribution map. The horizontal feature function obtained by linear regression is YT=0.0428×n+15.812, where the horizontal slope at=0.0428 and the horizontal intercept bt=15.812. Establish a contour feature data set to record the horizontal feature function, vertical feature function of the selected target and the corresponding final annotation in the annotation sample; Merge the box-selected objects with the same final annotations. After the merging is completed, the final annotations are a first-level directory. The horizontal feature function and the vertical feature function constitute a feature data. Each first-level directory includes several feature data. In practical applications, some data of the contour feature data set are shown in Table 1 below: Table 1 Partial data of contour feature dataset Among them, YT=0.0428×n+15.812 and YL=0.2144×n+43.8625 constitute a feature data. Table 1 is only to show the storage format and corresponding relationship of the data in the contour feature data set. Therefore, the specific data is not shown too much. It is only necessary to know the arrangement of the data. Each final annotation includes several feature data and records the value range of n of these feature data. When a user inputs a new frame selection target into the AI ​​big data annotation model, it is marked as a target to be annotated; the horizontal feature function and the vertical feature function of the target to be annotated are analyzed, and they are named as the horizontal function to be analyzed and the vertical function to be analyzed respectively, and a piece of data to be analyzed is composed of the horizontal function to be analyzed and the vertical function to be analyzed; Find out whether there is feature data in the contour feature data set that is exactly the same as the data to be analyzed. If so, the output data has a signal. If not, the output data does not have a signal. If there is a signal in the output data, the final annotation corresponding to the feature data that is exactly the same as the data to be analyzed is obtained and output; if there is no signal in the output data, the data to be analyzed and the feature data are matched and analyzed by a feature matching algorithm; In practical applications, when a user inputs a new framed target into the AI ​​big data annotation model, it is marked as a target to be annotated, and the horizontal function to be analyzed and the vertical function to be analyzed of the target to be annotated are extracted. It is then searched whether there is feature data in the contour feature data set whose horizontal feature function is the same as the horizontal function to be analyzed and whose vertical feature function is the same as the vertical function to be analyzed. If so, it means that the contour feature of the target to be annotated is exactly the same as a certain annotated sample in the contour feature data set. Therefore, the final annotation of the annotated sample can be directly output as the result. If not, it means that the contour features of the annotated sample and the target to be annotated are different. In this case, further analysis is required through a feature matching algorithm. The feature matching analysis unit is configured with a feature matching analysis strategy, which includes: See also Figure 8 As shown, the horizontal feature map and the vertical feature map are re-established and named as the horizontal comparison map and the vertical comparison map respectively; For any characteristic data, the straight line corresponding to the horizontal characteristic function in the characteristic data is named as the horizontal characteristic straight line, and the straight line corresponding to the vertical characteristic function in the characteristic data is named as the vertical characteristic straight line; the straight line corresponding to the horizontal function to be analyzed is named as the horizontal straight line to be analyzed, and the straight line corresponding to the vertical function to be analyzed is named as the vertical straight line to be analyzed; Enter the horizontal feature straight line and the horizontal straight line to be analyzed into the horizontal comparison diagram, and enter the vertical feature straight line and the vertical straight line to be analyzed into the vertical comparison diagram. When entering each straight line, follow the value range of the number of pixels n in the corresponding frame selection target, that is, convert the straight line into a line segment based on the value range of n in the frame selection target; Rename the horizontal characteristic straight line, the horizontal straight line to be analyzed, the vertical characteristic straight line and the vertical straight line to be analyzed to the horizontal characteristic line segment, the horizontal line segment to be analyzed, the vertical characteristic line segment and the vertical line segment to be analyzed; In practical applications, since the analysis process of the horizontal comparison graph and the vertical comparison graph is exactly the same, this embodiment only takes the horizontal comparison graph as an example to explain the specific implementation process of the feature matching algorithm; Figure 8 The figure shows the horizontal straight line to be analyzed and the horizontal characteristic straight line of a feature data in the contour feature data set, where the solid line is the straight line to be analyzed and the dotted line is the horizontal characteristic straight line. Since the value range of n of the target to be labeled is [1,10582], the horizontal straight line to be analyzed also only takes the part between [1,10582] to obtain the horizontal line segment to be analyzed; the value range of n corresponding to the horizontal characteristic straight line is [1,11496], so the horizontal characteristic straight line also only takes the part between [1,11496] to obtain the horizontal characteristic line segment; See also Fig. 9As shown, the left end point of the horizontal characteristic line segment and the horizontal line segment to be analyzed are connected by a straight line, and the right end point of the horizontal characteristic line segment and the horizontal line segment to be analyzed are connected at the same time. After the connection is completed, at most two closed areas are obtained, and the area of ​​the closed area is obtained and marked as the horizontal comparison index; Connect the left endpoints of the longitudinal characteristic straight line and the longitudinal straight line to be analyzed by a straight line, and connect the right endpoints of the longitudinal characteristic straight line and the longitudinal straight line to be analyzed by a straight line, and obtain at most two closed areas after the connection is completed, and obtain the area of ​​the closed area and mark it as the longitudinal comparison index; The horizontal comparison index and the vertical comparison index are added together to obtain the annotation deviation index, the data to be analyzed are analyzed with all the characteristic data in turn to obtain several annotation deviation indexes, and the minimum value among the annotation deviation indexes is found and marked as the minimum deviation index; The feature data corresponding to the minimum deviation index is marked as the target data, and the final annotation corresponding to the target data is output as the annotation result; In practical applications, Fig. 9 The closed area generated by connecting the endpoints of the horizontal feature line segment and the horizontal line segment to be analyzed is shown. If the two line segments do not intersect, a closed area is obtained. If the two line segments intersect, two closed areas are generated. The area of ​​the closed area is obtained to obtain the horizontal comparison index. The horizontal comparison index represents the degree of deviation between the two line segments. If the horizontal line segment to be analyzed is closer to the horizontal feature line segment, it means that the contour features of the target to be labeled and the labeled sample are closer in the horizontal direction, and the horizontal comparison index is also smaller, and vice versa. The meaning of the longitudinal comparison index is the same as that of the transverse comparison index. The annotation deviation index obtained by adding the meaning of the longitudinal comparison index to the transverse comparison index reveals the overall deviation degree of the contour features between the target to be labeled and the labeled sample. The minimum value can be found to obtain the labeled sample that is closest to the contour features of the target to be labeled, that is, the target data, and the final annotation of the target data is the annotation of the target to be labeled.

[0021] The automatic annotation module is used to efficiently and automatically annotate images through the AI ​​big data annotation model after training is completed; The automatic annotation module is configured with automatic annotation strategies, which include: The user extracts the input image through the image annotation platform to obtain the framed target in the input image; Input the selected target into the AI ​​big data annotation model, and the AI ​​big data annotation model will output the annotation results for automatic annotation; In actual applications, the AI ​​big data annotation model is connected to the image annotation platform for data. After the input image is extracted through the image annotation platform, the selected target in the input image can be intelligently annotated through the AI ​​big data annotation model, which greatly reduces the dependence on manual annotation.

[0022] Example 2, please refer to Fig.10 As shown, the present application provides an efficient data labeling method, comprising the following steps: Step S1, select the sample image through the image annotation platform and manually annotate the image to obtain an annotated sample; Step S1 includes the following sub-steps: Step S101, establishing data connection with an image annotation platform; Step S102, the sample image is framed by an image annotation platform to obtain a framed target, where the framed target is a rectangular area; Step S103, annotating the image of the selected object through internal self-operated annotation, third-party annotation and crowdsourcing annotation to obtain annotation content; Step S104: for any selected target, select the annotation content with the highest proportion as the final annotation. If the three annotation contents are different, the internal self-operated annotation, the third-party annotation and the crowdsourcing annotation need to negotiate to obtain the final annotation; Step S105, the selected target and the corresponding final annotation are the annotation samples; Step S2, constructing an AI big data annotation model, and training the AI ​​big data annotation model by analyzing the contour features of the annotated samples; Step S2 includes the following sub-steps: Step S201, constructing an AI big data annotation model, extracting the contour of the box-selected target in the annotation sample, obtaining the target contour, and constructing a feature distribution map based on the target contour; Step S201 includes the following sub-steps: Step S201.1, building an AI big data annotation model; Step S201.2, obtaining the framed target in the labeled sample, and extracting the contour of the framed target using the OpenCV contour detection algorithm to obtain the target contour; Step S201.3, establish a plane rectangular coordinate system with the lower left vertex of the selected object as the origin, the number of pixels in the horizontal direction as the X-axis, and the number of pixels in the vertical direction as the Y-axis, named as the feature distribution map, and the left side and bottom side of the selected object overlap with the Y-axis and the X-axis respectively; Step S201.4, obtaining the coordinates of each pixel point of the framed target in the feature distribution map, marked as pixel coordinates, and the pixel coordinates are expressed as (X, Y), where (X, Y) represents that the pixel point is at the Xth pixel point in the horizontal direction and the Yth pixel point in the vertical direction; Step S202, analyzing the feature distribution map to extract the contour features of the framed target; Step S202 includes the following sub-steps: Step S202.1, find pixel coordinates belonging to the target contour, mark them as contour coordinates, mark the X of the contour coordinates as the contour abscissa, and mark the Y of the contour coordinates as the contour ordinate; Step S202.2: number the contour horizontal coordinates in ascending order, using the symbol T n Indicates that the contour ordinates are numbered in ascending order, using the symbol L n It means, where n∈Z+ and n is the serial number of T and L; Step S202.3, with n as the horizontal axis, T n Establish a plane rectangular coordinate system for the vertical axis, named as the horizontal feature map, and convert T n Enter the horizontal feature map according to n; take n as the horizontal axis, L n Establish a plane rectangular coordinate system for the vertical axis, named as the vertical feature map, and set L n Enter into the longitudinal feature map according to n; Step S202.4, performing regression analysis on the horizontal feature map and the vertical feature map by linear regression in discrete regression analysis, and obtaining a horizontal feature function and a vertical feature function respectively; the format of the horizontal feature function is YT=at×n+bt, where YT is T n , at is the slope of the horizontal characteristic function, marked as the horizontal slope, bt is the intercept of the horizontal characteristic function, marked as the horizontal intercept; the format of the vertical characteristic function is YL=al×n+bl, where YL is L n , al is the slope of the longitudinal characteristic function, marked as the longitudinal slope, bl is the intercept of the longitudinal characteristic function, marked as the longitudinal intercept; Step S202.5, establishing a contour feature data set, recording the horizontal feature function, the vertical feature function of the framed target, and the corresponding final annotation in the annotation sample; Step S202.6, merging the box-selected objects with the same final labeling. After the merging is completed, the final labeling is a first-level directory. The horizontal feature function and the vertical feature function constitute a feature data. Each first-level directory includes a number of feature data. Step S202.7, when the user inputs a new box selection target into the AI ​​big data annotation model, it is marked as a target to be annotated; the horizontal feature function and the vertical feature function of the target to be annotated are analyzed, and they are named as the horizontal function to be analyzed and the vertical function to be analyzed, respectively, and the horizontal function to be analyzed and the vertical function to be analyzed form a piece of data to be analyzed; Step S202.8, searching whether there is feature data in the contour feature data set that is identical to the data to be analyzed, if so, outputting a data presence signal, if not, outputting a data non-existence signal; Step S202.9, if the output data has a signal, obtain the final annotation corresponding to the feature data that is exactly the same as the data to be analyzed and output it; if the output data does not have a signal, perform matching analysis on the data to be analyzed and the feature data using a feature matching algorithm; Step S202.9 includes the following sub-steps: Step S202.9.1, re-establishing the horizontal feature map and the vertical feature map, and naming them as the horizontal comparison map and the vertical comparison map respectively; Step S202.9.2, for any feature data, name the straight line corresponding to the horizontal feature function in the feature data as the horizontal feature straight line, and name the straight line corresponding to the vertical feature function in the feature data as the vertical feature straight line; name the straight line corresponding to the horizontal function to be analyzed as the horizontal straight line to be analyzed, and name the straight line corresponding to the vertical function to be analyzed as the vertical straight line to be analyzed; Step S202.9.3, enter the horizontal feature straight line and the horizontal straight line to be analyzed into the horizontal comparison diagram, and enter the vertical feature straight line and the vertical straight line to be analyzed into the vertical comparison diagram. When entering each straight line, the value range of the number of pixels n in the corresponding frame selection target is followed, that is, the straight line is converted into a line segment based on the value range of n in the frame selection target; Step S202.9.4, rename the horizontal characteristic straight line, the horizontal straight line to be analyzed, the vertical characteristic straight line and the vertical straight line to be analyzed as the horizontal characteristic line segment, the horizontal line segment to be analyzed, the vertical characteristic line segment and the vertical line segment to be analyzed; Step S202.9.5, connect the left endpoint of the horizontal feature line segment and the horizontal line segment to be analyzed by a straight line, and connect the right endpoint of the horizontal feature line segment and the horizontal line segment to be analyzed, and after the connection is completed, obtain at most two closed areas, obtain the area of ​​the closed area and mark it as the horizontal comparison index; Step S202.9.6, connect the left endpoints of the longitudinal characteristic straight line and the longitudinal straight line to be analyzed by a straight line, and connect the right endpoints of the longitudinal characteristic straight line and the longitudinal straight line to be analyzed by a straight line, and after the connection is completed, at most two closed areas are obtained, and the area of ​​the closed area is obtained and marked as the longitudinal comparison index; Step S202.9.7, add the horizontal comparison index and the vertical comparison index to obtain the annotation deviation index, analyze the data to be analyzed with all the feature data in turn to obtain several annotation deviation indices, find the minimum value among the annotation deviation indices, and mark it as the minimum deviation index; Step S202.9.8, marking the feature data corresponding to the minimum deviation index as target data, and outputting the final annotation corresponding to the target data as the annotation result; Step S3, after the training is completed, the image is efficiently and automatically annotated through the AI ​​big data annotation model; Step S3 includes the following sub-steps: Step S301, the user extracts the input image through the image annotation platform to obtain the framed target in the input image; Step S302: input the selected target into the AI ​​big data annotation model, and the AI ​​big data annotation model outputs the annotation result for automatic annotation.

[0023] Embodiment 3, the present application provides an electronic device, which may include: a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus. The memory stores computer-readable instructions, and the processor can call the instructions in the memory. When the computer-readable instructions are executed by the processor, the steps in an efficient data annotation method are executed to achieve the following functions: the sample image is framed through the image annotation platform and the image is manually annotated to obtain an annotated sample; an AI big data annotation model is constructed, and the AI ​​big data annotation model is trained by analyzing the contour features of the annotated sample; after the training is completed, the image is efficiently and automatically annotated through the AI ​​big data annotation model.

[0024] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0025] Embodiment 4. The present application also provides a computer-readable storage medium. The present application provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in the above efficient data labeling method are executed to achieve the following functions: a sample image is framed and manually labeled by an image labeling platform to obtain a labeled sample; an AI big data labeling model is constructed, and the AI ​​big data labeling model is trained by analyzing the contour features of the labeled sample; after the training is completed, the image is efficiently and automatically labeled by the AI ​​big data labeling model.

[0026] Through the description of the above implementation methods, the embodiments of the present invention can be provided as methods, systems or computer program products. Based on such an understanding, the above technical solutions can be essentially or partly contributed to the prior art in the form of software products, which can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and include several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0027] In the embodiments provided in the present application, it should be understood that the disclosed system or method can be implemented in other ways. The embodiments described above are merely schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules or units can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of systems, modules and units can be electrical, mechanical or other forms.

[0028] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An efficient data labeling method, characterized in that: The steps include: The sample images are framed and manually annotated on the image annotation platform to obtain annotated samples; Build an AI big data annotation model and train it by analyzing the contour features of the annotated samples; After training is completed, the images are efficiently and automatically annotated using the AI ​​big data annotation model.

2. The efficient data labeling method according to claim 1, characterized in that: The sample image is framed and manually annotated on the image annotation platform to obtain the annotated sample, which includes the following sub-steps: Connect data with the image annotation platform; The sample image is framed by an image annotation platform to obtain a framed target, where the framed target is a rectangular area; Annotate the image of the selected target through internal self-operated annotation, third-party annotation, and crowdsourcing annotation to obtain the annotation content; For any selected target, the annotation content with the highest percentage is selected as the final annotation. If the three annotation contents are different, the internal self-operated annotation, third-party annotation, and crowdsourcing annotation need to negotiate to obtain the final annotation; The selected target and the corresponding final annotation are the labeled samples.

3. The efficient data labeling method according to claim 2, characterized in that: Building an AI big data annotation model and training the AI ​​big data annotation model by analyzing the contour features of the annotation samples includes the following sub-steps: Build an AI big data annotation model, extract the contour of the box-selected target in the annotated sample, obtain the target contour, and build a feature distribution map based on the target contour; Analyze the feature distribution map and extract the contour features of the selected target.

4. The efficient data labeling method according to claim 3, characterized in that: Construct an AI big data annotation model, extract the contour of the box-selected target in the annotation sample, obtain the target contour, and construct a feature distribution map based on the target contour, including the following sub-steps: Build AI big data annotation model; Get the boxed target in the labeled sample, extract the contour of the boxed target using the OpenCV contour detection algorithm, and obtain the target contour; A plane rectangular coordinate system is established with the lower left vertex of the selected object as the origin, the number of pixels in the horizontal direction as the X-axis, and the number of pixels in the vertical direction as the Y-axis, and is named a feature distribution map. The left side and bottom side of the selected object overlap with the Y-axis and the X-axis respectively. The coordinates of each pixel point of the selected target in the feature distribution map are obtained and marked as pixel coordinates. The pixel coordinates are expressed as (X, Y), where (X, Y) represents that the pixel point is at the Xth pixel point in the horizontal direction and the Yth pixel point in the vertical direction.

5. The efficient data labeling method according to claim 4, characterized in that: Analyzing the feature distribution map and extracting the contour features of the selected target includes the following sub-steps: Find the pixel coordinates belonging to the target contour, mark them as contour coordinates, mark the X of the contour coordinates as the contour horizontal coordinate, and mark the Y of the contour coordinates as the contour vertical coordinate; The contour abscissas are numbered in ascending order, using the symbol T n Indicates that the contour ordinates are numbered in ascending order, using the symbol L n It means, where n∈Z+ and n is the serial number of T and L; With n as the horizontal axis, T n Establish a plane rectangular coordinate system for the vertical axis, named as the horizontal feature map, and convert T n Enter the horizontal feature map according to n; take n as the horizontal axis, L n Establish a plane rectangular coordinate system for the vertical axis, named as the vertical feature map, and set L n Enter into the longitudinal feature map according to n; The horizontal feature map and the vertical feature map are regressed by linear regression in discrete regression analysis to obtain a horizontal feature function and a vertical feature function respectively; the format of the horizontal feature function is YT=at×n+bt, where YT is T n , at is the slope of the horizontal characteristic function, marked as the horizontal slope, bt is the intercept of the horizontal characteristic function, marked as the horizontal intercept; the format of the longitudinal characteristic function is YL=al×n+bl, where YL is L n , al is the slope of the longitudinal characteristic function, marked as the longitudinal slope, bl is the intercept of the longitudinal characteristic function, marked as the longitudinal intercept; Establish a contour feature data set to record the horizontal feature function, vertical feature function of the selected target and the corresponding final annotation in the annotation sample; Merge the box-selected objects with the same final annotations. After the merging is completed, the final annotations are a first-level directory. The horizontal feature function and the vertical feature function constitute a feature data. Each first-level directory includes several feature data. When a user inputs a new frame selection target into the AI ​​big data annotation model, it is marked as a target to be annotated; the horizontal feature function and the vertical feature function of the target to be annotated are analyzed, and they are named as the horizontal function to be analyzed and the vertical function to be analyzed respectively, and a piece of data to be analyzed is composed of the horizontal function to be analyzed and the vertical function to be analyzed; Find out whether there is feature data in the contour feature data set that is exactly the same as the data to be analyzed. If so, the output data has a signal. If not, the output data does not have a signal. If there is a signal in the output data, the final annotation corresponding to the feature data that is exactly the same as the data to be analyzed is obtained and output; if there is no signal in the output data, the data to be analyzed and the feature data are matched and analyzed through a feature matching algorithm.

6. The efficient data labeling method according to claim 5, characterized in that: The feature matching algorithm includes the following sub-steps: Re-establish the horizontal feature map and the vertical feature map, and name them as the horizontal comparison map and the vertical comparison map respectively; For any characteristic data, the straight line corresponding to the horizontal characteristic function in the characteristic data is named as the horizontal characteristic straight line, and the straight line corresponding to the vertical characteristic function in the characteristic data is named as the vertical characteristic straight line; the straight line corresponding to the horizontal function to be analyzed is named as the horizontal straight line to be analyzed, and the straight line corresponding to the vertical function to be analyzed is named as the vertical straight line to be analyzed; Enter the horizontal feature straight line and the horizontal straight line to be analyzed into the horizontal comparison diagram, and enter the vertical feature straight line and the vertical straight line to be analyzed into the vertical comparison diagram. When entering each straight line, follow the value range of the number of pixels n in the corresponding frame selection target, that is, convert the straight line into a line segment based on the value range of n in the frame selection target; Rename the horizontal characteristic straight line, the horizontal straight line to be analyzed, the vertical characteristic straight line and the vertical straight line to be analyzed to the horizontal characteristic line segment, the horizontal line segment to be analyzed, the vertical characteristic line segment and the vertical line segment to be analyzed; Connect the left end point of the horizontal characteristic line segment and the horizontal line segment to be analyzed by a straight line, and connect the right end point of the horizontal characteristic line segment and the horizontal line segment to be analyzed, and obtain at most two closed areas after the connection is completed, and obtain the area of ​​the closed area and mark it as the horizontal comparison index; Connect the left endpoints of the longitudinal characteristic straight line and the longitudinal straight line to be analyzed by a straight line, and connect the right endpoints of the longitudinal characteristic straight line and the longitudinal straight line to be analyzed by a straight line, and obtain at most two closed areas after the connection is completed, and obtain the area of ​​the closed area and mark it as the longitudinal comparison index; The horizontal comparison index and the vertical comparison index are added together to obtain the annotation deviation index, the data to be analyzed are analyzed with all the characteristic data in turn to obtain several annotation deviation indexes, and the minimum value among the annotation deviation indexes is found and marked as the minimum deviation index; The feature data corresponding to the minimum deviation index is marked as the target data, and the final annotation corresponding to the target data is output as the annotation result.

7. The efficient data labeling method according to claim 6, characterized in that: After training is completed, the AI ​​big data annotation model is used to efficiently and automatically annotate images, including the following sub-steps: The user extracts the input image through the image annotation platform to obtain the framed target in the input image; Input the selected target into the AI ​​big data annotation model, and the AI ​​big data annotation model will output the annotation results for automatic annotation.

8. An efficient data labeling system, used to implement an efficient data labeling method according to any one of claims 1 to 7, characterized in that: It includes a platform extraction module, a labeling training module and an automatic labeling module; the platform extraction module and the automatic labeling module are respectively connected with the labeling training module data; The platform extraction module is used to select the sample image through the image annotation platform and manually annotate the image to obtain annotated samples; The annotation training module is used to build an AI big data annotation model, and train the AI ​​big data annotation model by analyzing the contour features of the annotation samples; The automatic annotation module is used to efficiently and automatically annotate images through an AI big data annotation model after training is completed.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the steps in the method according to any one of claims 1 to 7 are executed.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps in the method according to any one of claims 1 to 7 are executed.

Images