An automatic labeling system and method for the extent of contaminated soil

Abstract

This invention provides an automatic labeling system and method for the extent of contaminated soil, relating to the field of soil pollution investigation technology. It includes: an information acquisition module, used to acquire video of a target site containing contaminated soil; select any frame from the target site video containing the complete boundary of the contaminated soil in the target site video; and label the contaminated soil boundary of the target site image as a standard image; a boundary labeling module, connected to the information acquisition module, used to extract features from the images of the target site to be labeled (excluding the standard image) in the target site video based on the standard image to obtain corresponding soil feature points; use the standard image as the starting tracking frame to track the soil feature points frame-by-frame in each image of the target site to be labeled; and, based on the frame-by-frame tracking results, label the contaminated soil boundary of each image of the target site to be labeled to generate and output a labeled video containing contaminated soil boundary markers. The beneficial effect is the automatic tracking and labeling of the boundary of contaminated soil.

Description

Technical Field

[0001] This invention relates to the field of soil pollution investigation technology, and in particular to an automatic labeling system and method for the extent of polluted soil. Background Technology

[0002] Soil is the material foundation for sustainable economic and social development and is closely related to public health. Conducting soil pollution surveys allows us to understand the quality of soil resources, laying a solid foundation for effective soil pollution prevention and control. Accurate site selection is crucial for accurately identifying the types, levels, and spatial distribution of pollutants in the surveyed soil. Improper site selection can lead to missed pollution and misjudgments. Site selection involves data collection, on-site reconnaissance, and interviews, followed by comprehensive analysis and judgment of the information. Information is typically recorded using paper records and photographs to document the current state of the site. In recent years, the rapid development of aerial photography technology has made the use of drones equipped with navigation devices to capture high-altitude land images, and the application of image processing technology to quickly and accurately obtain soil information a de facto standard.

[0003] However, most soil pollution surveys often suffer from delays in data recording, or the photos taken are not representative and fail to provide a clear and comprehensive view of the site's current condition and boundaries. This leads to the omission of some suspected contaminated areas, and the sampling points not covering all suspected contaminated areas. Alternatively, unclear boundary determination may result in the inclusion of areas adjacent to the site boundary within the scope of suspected contaminated area identification. Currently, there is no system capable of automatically marking the boundaries of contaminated soil contained in video footage of a target site. Summary of the Invention

[0004] To address the problems existing in the prior art, the present invention provides an automatic labeling system for the extent of contaminated soil, comprising:

[0005] The information acquisition module is used to acquire a video of a target plot containing contaminated soil, select any frame of the target plot image containing the complete boundary of the contaminated soil from the video of the target plot, and mark the boundary of the contaminated soil on the target plot image as a standard image.

[0006] The boundary labeling module, connected to the information acquisition module, is used to extract features from the target plot images (excluding the standard image) in the target plot video based on the standard image to obtain corresponding soil feature points, use the standard image as the starting tracking frame to track the soil feature points frame by frame in each target plot image, and perform boundary labeling of the contaminated soil in each target plot image based on the frame-by-frame tracking results to generate and output a labeled video containing contaminated soil boundary markers.

[0007] Preferably, the boundary labeling module includes:

[0008] The sequence segmentation unit is used to take the standard image as the starting tracking frame, and to form a reverse sequence of each of the images to be labeled before the standard image in the target plot video and the standard image in chronological order, and to form a forward sequence of each of the images to be labeled after the standard image in the target plot video and the standard image in chronological order.

[0009] The first tracking and labeling unit, connected to the sequence segmentation unit, is used to extract features from the next frame of the image to be labeled in the positive sequence, starting with the standard image as the first tracking frame, to obtain the corresponding soil feature points and perform frame-by-frame tracking, and to perform boundary labeling on each image to be labeled according to the frame-by-frame tracking results to obtain the corresponding labeled image.

[0010] The second tracking and labeling unit, connected to the sequence segmentation unit, is used to extract features from the previous frame of the image to be labeled in the reverse sequence, starting with the standard image as the tracking frame, to obtain the corresponding soil feature points and perform frame-by-frame tracking, and to perform boundary labeling on each image to be labeled according to the frame-by-frame tracking results to obtain the corresponding labeled image.

[0011] The video synthesis unit is connected to the first tracking and annotation unit and the second tracking and annotation unit respectively. It is used to generate and output an annotated video containing contaminated soil boundary markers based on each of the annotated images after all the images to be annotated have been annotated.

[0012] Preferably, the boundary labeling module further includes a tracking calculation unit, which is connected to the first tracking labeling unit and the second tracking labeling unit respectively, for calculating the similarity between the soil feature points of the starting tracking frame and the adjacent image to be labeled as the frame-by-frame tracking result of the adjacent image to be labeled, and using the image to be labeled as the starting tracking frame.

[0013] Preferably, the video synthesis unit includes:

[0014] The first synthesis unit is used to synthesize the labeled images corresponding to each image to be labeled in the forward sequence into a first video according to the time order of the forward sequence.

[0015] The second synthesis unit is used to synthesize the labeled images corresponding to each image to be labeled in the reverse sequence into a second video according to the time order of the reverse sequence.

[0016] The third synthesis unit, connecting the first synthesis unit and the second synthesis unit, is used to synthesize the first video and the second video using the standard image as video connection points to create the labeled video and output it.

[0017] Preferably, the tracking calculation unit is further configured to generate a tracking success signal when the similarity is greater than a similarity threshold, and to generate a tracking loss signal when the similarity is not greater than the similarity threshold. The first tracking annotation unit and the second tracking unit perform boundary annotation on each of the images to be annotated according to the tracking success signal to obtain the corresponding annotated image. Therefore, the boundary annotation module further includes:

[0018] The adjustment unit is connected to the tracking calculation unit, the first tracking annotation unit, and the second tracking unit, respectively. It is used to provide the operator with a corresponding prompt when the tracking loss signal is received, and then to perform boundary annotation on the image to be annotated adjacent to the starting tracking frame according to the feedback external instructions, as the corresponding annotated image.

[0019] This invention also provides an automatic labeling method for the extent of contaminated soil, using the automatic labeling system described above, comprising:

[0020] Step S1: The automatic labeling system acquires a video of a target plot containing contaminated soil, selects any frame of the target plot image containing the complete boundary of the contaminated soil from the video, and labels the boundary of the contaminated soil on the target plot image as a standard image.

[0021] In step S2, the automatic labeling system extracts features from the target plot images (excluding the standard image) in the video of the target plot based on the standard image to obtain corresponding soil feature points. Using the standard image as the starting tracking frame, the system tracks the soil feature points frame by frame in each of the target plot images. Based on the frame-by-frame tracking results, the system performs boundary labeling of the contaminated soil in each of the target plot images to generate a labeled video containing contaminated soil boundary markers and outputs it.

[0022] Preferably, step S2 includes:

[0023] Step S21: The automatic annotation system uses the standard image as the starting tracking frame, and forms a reverse sequence of the images to be annotated before the standard image in the target plot video and the standard image in chronological order, and forms a forward sequence of the images to be annotated after the standard image in the target plot video and the standard image in chronological order.

[0024] Step S22: For each of the images to be labeled in the ascending sequence, the automatic labeling system extracts features from the next frame of the image to be labeled, starting with the standard image, to obtain the corresponding soil feature points and performs frame-by-frame tracking. Based on the frame-by-frame tracking results, the system performs boundary labeling on each of the images to be labeled to obtain the labeled image.

[0025] Step S23: For each of the images to be labeled in the reverse sequence, the automatic labeling system extracts features from the previous frame of the image to be labeled, starting with the standard image, to obtain the corresponding soil feature points and performs frame-by-frame tracking. Based on the frame-by-frame tracking results, the system performs boundary labeling on each of the images to be labeled to obtain the labeled image.

[0026] Step S24: After all the images to be labeled are labeled, the automatic labeling system generates and outputs a labeled video containing boundary markers of contaminated soil based on each labeled image.

[0027] Preferably, in step S2, when performing feature extraction and frame-by-frame tracking on each of the images to be labeled based on the starting tracking frame, the similarity between the soil feature points of the starting tracking frame and the adjacent images to be labeled is taken as the frame-by-frame tracking result of the adjacent images to be labeled, and the image to be labeled is taken as the starting tracking frame.

[0028] Preferably, step S24 includes:

[0029] Step S241: For the ascending sequence, the automatic annotation system performs video synthesis on the labeled images corresponding to each image to be annotated in the ascending sequence according to the time order of the ascending sequence to obtain the first video.

[0030] Step S242, for the reverse sequence, the automatic annotation system performs video synthesis of the labeled images corresponding to each image to be annotated in the reverse sequence according to the time order of the reverse sequence to obtain a second video;

[0031] In step S243, the automatic annotation system combines the first video and the second video using the standard image as the video connection point to create the annotated video and outputs it.

[0032] Preferably, steps S22 and S23 further include determining whether the similarity is greater than the similarity threshold:

[0033] If so, then boundary annotation is performed on each of the images to be annotated to obtain the corresponding annotated image;

[0034] If not, the operator is given a corresponding prompt, and then the boundary of the image to be labeled adjacent to the starting tracking frame is marked as the corresponding labeled image according to the feedback external instructions.

[0035] The above technical solution has the following advantages or beneficial effects: The automatic annotation system provided by this invention can automatically track and annotate the boundaries of contaminated soil in the target plot video based on soil feature points, avoiding the problem of missing some suspected contaminated areas due to untimely information recording or non-typical photos, and the inability to intuitively and comprehensively display the current status of the plot and its boundary range in the traditional method. Attached Figure Description

[0036] Figure 1 A schematic diagram of the structure of an automatic labeling system for the extent of contaminated soil, as a preferred embodiment of the present invention;

[0037] Figure 2 A flowchart illustrating an automatic labeling method for the extent of contaminated soil, as a preferred embodiment of the present invention.

[0038] Figure 3 A schematic diagram of step S2 of an automatic labeling method for the extent of contaminated soil, as described in a preferred embodiment of the present invention.

[0039] Figure 4 In a preferred embodiment of the present invention, a schematic diagram of step S24 of an automatic labeling method for the extent of contaminated soil is provided. Detailed Implementation

[0040] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The present invention is not limited to this embodiment; other embodiments that conform to the spirit of the present invention may also fall within the scope of the present invention.

[0041] In a preferred embodiment of the present invention, based on the above-mentioned problems existing in the prior art, an automatic labeling system for the extent of contaminated soil is provided, such as... Figure 1 As shown, it includes:

[0042] Information acquisition module 1 is used to acquire video of a target plot containing contaminated soil, select any frame of the target plot image containing the complete boundary of contaminated soil from the target plot video, and mark the boundary of contaminated soil on the target plot image as a standard image.

[0043] The boundary labeling module 2, connected to the information acquisition module 1, is used to extract the corresponding soil feature points from the target plot images other than the standard image in the video of the target plot based on the standard image. The standard image is used as the starting tracking frame to track the soil feature points of each plot image frame by frame. Based on the frame-by-frame tracking results, the boundary labeling of the contaminated soil is performed on each plot image to generate a labeled video containing the boundary markers of the contaminated soil and output it.

[0044] Specifically, in this embodiment, a device with camera function (such as a drone or mobile phone) is used to take pictures of the target plot and its surroundings. Currently, drones are often used for aerial photography to obtain more comprehensive videos of the target plot for the investigation of contaminated soil. During aerial photography, images of the entire area of ​​the target plot are first taken (that is, in addition to capturing the entire boundary of the contaminated soil in the target plot, the surrounding area of ​​the contaminated soil, such as rivers, fields and forests, are also captured in the picture). Then, the target plot is photographed from different spatial directions to avoid the contaminated soil to be investigated being obscured by a single angle, resulting in incomplete information. In this way, multiple videos of the target plot containing contaminated soil are obtained.

[0045] This system first captures video of the target site from a mobile storage device (direct connection to a drone, USB flash drive, etc.). Optionally, the video is decomposed into multiple images to be labeled, forming a sequence of images (or the video can be used directly without decomposition). The decomposition method used here is provided by the OpenCV library (OpenCV is an open-source computer vision library that can be used to process multimedia content such as images and videos).

[0046] Then, PyQt is used to provide a visual interface to the user to display the image sequence (or target plot video) to be labeled. Then, according to the user's selection, a frame image is selected from the image sequence (or target plot video) to be labeled (this frame image is any frame image that contains all the boundaries of the contaminated soil) and the contaminated soil in the image is manually labeled as a standard image.

[0047] Subsequently, based on the standard image, features were extracted sequentially from each image to be labeled to obtain soil feature points, and the soil feature points were tracked frame by frame. Based on the tracking results, the images to be labeled were labeled.

[0048] It can automatically and quickly track and mark the boundaries of contaminated soil in the video of the target site, avoiding the problems of missing some suspected contaminated areas due to untimely information recording or non-typical photos, and the inability to intuitively and comprehensively display the current status and boundary range of the site in traditional methods.

[0049] In a preferred embodiment of the present invention, the image sequence to be labeled includes a forward sequence and a reverse sequence, then as follows: Figure 1 As shown, boundary annotation module 2 includes:

[0050] The sequence segmentation unit 21 is used to take the standard image as the starting tracking frame, form a reverse sequence of each image to be labeled before the standard image in the target plot video according to the time order and the standard image, and form a forward sequence of each image to be labeled after the standard image in the target plot video according to the time order and the standard image.

[0051] The first tracking and labeling unit 22 is connected to the sequence segmentation unit 21. It is used to extract the corresponding soil feature points from the next frame of the image to be labeled in the forward sequence, starting with the standard image as the tracking frame, and to perform frame-by-frame tracking. Based on the frame-by-frame tracking results, it performs boundary labeling on each image to be labeled to obtain the corresponding labeled image.

[0052] The second tracking and labeling unit 23 is connected to the sequence segmentation unit 21. It is used to extract the corresponding soil feature points from the previous frame of the image to be labeled in the reverse sequence, starting with the standard image as the tracking frame, and then perform frame-by-frame tracking. Based on the frame-by-frame tracking results, it performs boundary labeling on each image to be labeled to obtain the corresponding labeled image.

[0053] The video synthesis unit 24 is connected to the first tracking and annotation unit 22 and the second tracking and annotation unit 23 respectively. It is used to generate and output an annotated video containing the boundary marker of polluted soil based on each annotated image after all the images to be annotated have been annotated.

[0054] Specifically, in this embodiment, the selected standard image is generally located in the middle part of the target plot video. In order to save the time of automatic annotation, a bidirectional frame-by-frame tracking annotation method is preferred.

[0055] First, using the standard image as the dividing point, the videos to be labeled in the target plot video are divided into a forward sequence and a reverse sequence (that is, assuming the target plot has 10 frames and the selected standard image is in the 5th frame, then 1-5 are the reverse sequence and 5-10 are the forward sequence. The number of frames here is only for illustrative purposes, and the actual number of frames in the target plot video is much greater than this number).

[0056] Subsequently, feature extraction was performed on the images to be labeled in both the forward and reverse sequences using standard images, yielding the corresponding soil feature points (in the forward sequence, feature extraction was performed on the image to be labeled in the 6th frame using the standard image of the 5th frame; in the reverse sequence, feature extraction was performed on the image to be labeled in the 4th frame using the standard image of the 5th frame); the feature point extraction algorithm used here is a feature point extraction algorithm from the OpenCV library (such as SIFT, SURF, etc.);

[0057] Subsequently, the adjacent images to be labeled will be tracked frame by frame based on the starting tracking point (i.e., in the forward sequence, the soil feature points of the image to be labeled in the 6th frame will be tracked using the standard image of the 5th frame, and in the reverse sequence, the soil feature points of the standard image of the 5th frame will be tracked using the soil feature points of the image to be labeled in the 4th frame). Feature point matching and tracking algorithms (such as optical flow, KLT tracker, etc.) will be used here.

[0058] In summary, in the forward sequence, firstly, based on the standard image of frame 5 (the initial tracking frame), feature extraction is performed on the image to be labeled in frame 6 to obtain soil feature points for frame 6. Then, based on the soil feature points of the standard image of frame 5, the soil feature points of frame 6 are tracked. When the tracking result is reached, the boundary of the image to be labeled in frame 6 is marked based on the soil feature points of frame 6, thus completing the boundary marking of frame 6, and frame 6 is used as the initial tracking frame. Next, based on the standard image of frame 5, feature extraction is performed on the image to be labeled in frame 7 to obtain soil feature points for frame 7. Then, based on the soil feature points of the standard image of frame 6, the soil feature points of frame 7 are tracked. Soil feature points are tracked. When the standard tracking result is reached, the boundary of the image to be labeled in the 7th frame is marked based on the soil feature points of the 6th frame. This completes the boundary labeling of the 7th frame, and the 7th frame image is used as the starting tracking frame. Subsequently, feature extraction and frame-by-frame tracking are performed on the images to be labeled in the 8th, 9th, and 10th frames in the same way, and the boundary of the images to be labeled in the forward sequence is marked based on the tracking results. The same method is used in the reverse sequence, except that feature extraction and frame-by-frame tracking are performed in the forward sequence of 5, 6, 7, 8, 9, 10, and in the reverse sequence of 5, 4, 3, 2, 1.

[0059] In a preferred embodiment of the present invention, such as Figure 1 As shown, the boundary labeling module 2 also includes a tracking calculation unit 25, which is connected to the first tracking labeling unit 22 and the second tracking labeling unit 23 respectively. It is used to calculate the similarity between the soil feature points of the starting tracking frame and the adjacent image to be labeled as the frame-by-frame tracking result of the adjacent image to be labeled, and to use the image to be labeled as the starting tracking frame.

[0060] In a preferred embodiment of the present invention, such as Figure 1 As shown, the tracking calculation unit 25 is also used to generate a tracking success signal when the similarity is greater than the similarity threshold, and to generate a tracking loss signal when the similarity is not greater than the similarity threshold. The first tracking annotation unit 22 and the second tracking annotation unit 23 perform boundary annotation on each image to be annotated according to the tracking success signal to obtain the corresponding annotated image. Therefore, the boundary annotation module 2 further includes:

[0061] The adjustment unit 26 is connected to the tracking calculation unit 25, the first tracking annotation unit 22, and the second tracking unit 23 respectively. It is used to provide the operator with corresponding prompts when a tracking loss signal is received. Then, according to the feedback external instructions, it performs boundary annotation on the images to be annotated adjacent to the starting tracking frame as the corresponding annotated images.

[0062] Specifically, in this embodiment, because the object to be tracked moves gradually throughout the video, its position in adjacent frames will also shift. Therefore, the tracking algorithm mainly uses the similarity of feature points between adjacent images to determine whether the tracking is successful. When the similarity is greater than the similarity threshold, the tracking is successful. At this time, the boundary of the contaminated soil is marked based on the soil feature points extracted from the image to be labeled. When the similarity is not greater than the similarity threshold, the tracking fails (usually because the object to be tracked is occluded, resulting in the absence of the object to be tracked in the next frame). Automatic labeling will stop, and the operator needs to be notified accordingly. Then, based on the feedback of external instructions, the boundary of the image to be labeled adjacent to the starting tracking frame is marked as the corresponding labeled image, so that the automatic labeling of the target plot video can continue.

[0063] In a preferred embodiment of the present invention, such as Figure 1 As shown, the video synthesis unit 24 includes:

[0064] The first synthesis unit 241 is used to synthesize the labeled images corresponding to each image to be labeled in the forward sequence into a first video according to the time order of the forward sequence.

[0065] The second synthesis unit 242 is used to synthesize the labeled images corresponding to each image to be labeled in the reverse sequence into a second video according to the time order of the reverse sequence for the reverse sequence.

[0066] The third synthesis unit 243, connected to the first synthesis unit 241 and the second synthesis unit 242, is used to synthesize the first video and the second video using standard images as video connection points to produce an annotated video and output it.

[0067] This invention also provides an automatic labeling method for the extent of contaminated soil, using the automatic labeling system described above, such as... Figure 2 As shown, it includes:

[0068] Step S1: The automatic annotation system acquires a video of the target plot containing contaminated soil, selects any frame of the target plot image containing the complete boundary of the contaminated soil from the target plot video, and annotates the boundary of the contaminated soil in the target plot image as a standard image.

[0069] In step S2, the automatic annotation system extracts features from the target plot images (excluding the standard image) in the video of the target plot to obtain the corresponding soil feature points. Using the standard image as the starting tracking frame, the system tracks the soil feature points of each plot image frame by frame. Based on the frame-by-frame tracking results, the system marks the boundaries of contaminated soil in each plot image to generate an annotated video containing contaminated soil boundary markers and outputs it.

[0070] In a preferred embodiment of the present invention, such as Figure 3 As shown, step S2 includes:

[0071] Step S21: The automatic annotation system takes the standard image as the starting tracking frame, and forms a reverse sequence of each image to be annotated before the standard image in the target plot video according to the time order and the standard image, and forms a forward sequence of each image to be annotated after the standard image in the target plot video according to the time order and the standard image.

[0072] Step S22: For each image to be labeled in the ascending sequence, the automatic labeling system extracts the corresponding soil feature points from the next frame of the image to be labeled, starting with the standard image as the tracking frame, and performs frame-by-frame tracking. Based on the frame-by-frame tracking results, the system performs boundary labeling on each image to be labeled to obtain the labeled image.

[0073] Step S23: For each image to be labeled in the reverse sequence, the automatic labeling system extracts the corresponding soil feature points from the previous frame of the image to be labeled, starting with the standard image as the tracking frame, and performs frame-by-frame tracking. Based on the frame-by-frame tracking results, the system performs boundary labeling on each image to be labeled to obtain the labeled image.

[0074] Step S24: After all the images to be labeled are labeled, the automatic labeling system generates and outputs a labeled video containing boundary markers of contaminated soil based on each labeled image.

[0075] In a preferred embodiment of the present invention, when performing feature extraction and frame-by-frame tracking on each image to be labeled according to the starting tracking frame in step S2, the similarity between the soil feature points of the starting tracking frame and the adjacent images to be labeled is used as the frame-by-frame tracking result of the adjacent images to be labeled, and the images to be labeled are used as the starting tracking frame.

[0076] In a preferred embodiment of the present invention, such as Figure 4 As shown, step S24 includes:

[0077] Step S241: For the ascending sequence, the automatic annotation system synthesizes the labeled images corresponding to each image to be annotated in the ascending sequence into a video based on the time order of the ascending sequence.

[0078] Step S242: For the reverse sequence, the automatic annotation system synthesizes the labeled images corresponding to each image to be annotated in the reverse sequence into a second video according to the time order of the reverse sequence.

[0079] In step S243, the automatic annotation system combines the first video and the second video using standard images as video connection points to create an annotated video and outputs it.

[0080] In a preferred embodiment of the present invention, steps S22 and S23 further include determining whether the similarity is greater than a similarity threshold:

[0081] If so, then boundary annotations are performed on each image to be annotated to obtain the corresponding annotated image;

[0082] If not, the operator is given the corresponding prompt, and then the boundary annotation of the image to be annotated adjacent to the starting tracking frame is performed according to the feedback external instructions as the corresponding annotated image.

[0083] The above are merely preferred embodiments of the present invention and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should recognize that any equivalent substitutions and obvious changes made using the content of this specification and illustrations should be included within the protection scope of the present invention.

Claims

1. An automatic labeling system for the extent of contaminated soil, characterized in that, include: The information acquisition module is used to acquire videos of target sites containing contaminated soil. The videos of target sites include videos of the contaminated soil and the surrounding area of ​​the target site, as well as videos of the target site taken from different spatial locations. The information acquisition module is also used to decompose the target plot frame by frame to form an image sequence to be labeled, select any frame of the image sequence to be labeled containing the complete boundary of the contaminated soil according to the user's selection, and manually label the target plot image as a standard image. The boundary labeling module, connected to the information acquisition module, is used to extract features from the target plot images (excluding the standard image) in the video of the target plot based on the standard image to obtain corresponding soil feature points. Using the standard image as the starting tracking frame, the module performs frame-by-frame tracking of the soil feature points in each of the target plot images. Upon completion of each tracking, the tracked target plot image is used as the starting tracking frame for the next tracking. Based on the frame-by-frame tracking results, the module performs boundary labeling of the contaminated soil in each of the target plot images to generate a labeled video containing contaminated soil boundary markers and outputs it.

2. The automatic labeling system according to claim 1, characterized in that, The boundary labeling module includes: The sequence segmentation unit is used to take the standard image as the starting tracking frame, and to form a reverse sequence of each of the images to be labeled before the standard image in the target plot video and the standard image in chronological order, and to form a forward sequence of each of the images to be labeled after the standard image in the target plot video and the standard image in chronological order. The first tracking and labeling unit, connected to the sequence segmentation unit, is used to extract features from the next frame of the image to be labeled in the positive sequence, starting with the standard image as the first tracking frame, to obtain the corresponding soil feature points and perform frame-by-frame tracking, and to perform boundary labeling on each image to be labeled according to the frame-by-frame tracking results to obtain the corresponding labeled image. The second tracking and labeling unit, connected to the sequence segmentation unit, is used to extract features from the previous frame of the image to be labeled in the reverse sequence, starting with the standard image as the tracking frame, to obtain the corresponding soil feature points and perform frame-by-frame tracking, and to perform boundary labeling on each image to be labeled according to the frame-by-frame tracking results to obtain the corresponding labeled image. The video synthesis unit is connected to the first tracking and annotation unit and the second tracking and annotation unit respectively. It is used to generate and output an annotated video containing contaminated soil boundary markers based on each of the annotated images after all the images to be annotated have been annotated.

3. The automatic labeling system according to claim 2, characterized in that, The boundary labeling module further includes a tracking calculation unit, which is connected to the first tracking labeling unit and the second tracking labeling unit respectively. It is used to calculate the similarity between the soil feature points of the starting tracking frame and the adjacent image to be labeled as the frame-by-frame tracking result of the adjacent image to be labeled, and to use the image to be labeled as the starting tracking frame.

4. The automatic labeling system according to claim 2, characterized in that, The video synthesis unit includes: The first synthesis unit is used to synthesize the labeled images corresponding to each image to be labeled in the forward sequence into a first video according to the time order of the forward sequence. The second synthesis unit is used to synthesize the labeled images corresponding to each image to be labeled in the reverse sequence into a second video according to the time order of the reverse sequence. The third synthesis unit, connecting the first synthesis unit and the second synthesis unit, is used to synthesize the first video and the second video using the standard image as video connection points to create the labeled video and output it.

5. The automatic annotation system according to claim 3, characterized in that, The tracking calculation unit is further configured to generate a tracking success signal when the similarity is greater than a similarity threshold, and to generate a tracking loss signal when the similarity is not greater than the similarity threshold. The first tracking annotation unit and the second tracking unit perform boundary annotation on each of the images to be annotated based on the tracking success signal to obtain the corresponding annotated image. Therefore, the boundary annotation module further includes: The adjustment unit is connected to the tracking calculation unit, the first tracking annotation unit, and the second tracking unit, respectively. It is used to provide the operator with a corresponding prompt when the tracking loss signal is received, and then to perform boundary annotation on the image to be annotated adjacent to the starting tracking frame according to the feedback external instructions, as the corresponding annotated image.

6. An automatic labeling method for the extent of contaminated soil, characterized in that, The automatic annotation system described in any one of claims 1-5 includes: Step S1: The automatic labeling system acquires a video of a target plot containing contaminated soil, selects any frame of the target plot image containing the complete boundary of the contaminated soil from the video, and labels the boundary of the contaminated soil on the target plot image as a standard image. In step S2, the automatic labeling system extracts features from the target plot images (excluding the standard image) in the video of the target plot based on the standard image to obtain corresponding soil feature points. Using the standard image as the starting tracking frame, the system tracks the soil feature points frame by frame in each of the target plot images. Based on the frame-by-frame tracking results, the system performs boundary labeling of the contaminated soil in each of the target plot images to generate a labeled video containing contaminated soil boundary markers and outputs it.

7. The automatic annotation method according to claim 6, characterized in that, Step S2 includes: Step S21: The automatic annotation system uses the standard image as the starting tracking frame, and forms a reverse sequence of the images to be annotated before the standard image in the target plot video and the standard image in chronological order, and forms a forward sequence of the images to be annotated after the standard image in the target plot video and the standard image in chronological order. Step S22: For each of the images to be labeled in the ascending sequence, the automatic labeling system extracts features from the next frame of the image to be labeled, starting with the standard image, to obtain the corresponding soil feature points and performs frame-by-frame tracking. Based on the frame-by-frame tracking results, the system performs boundary labeling on each of the images to be labeled to obtain the labeled image. Step S23: For each of the images to be labeled in the reverse sequence, the automatic labeling system extracts features from the previous frame of the image to be labeled, starting with the standard image, to obtain the corresponding soil feature points and performs frame-by-frame tracking. Based on the frame-by-frame tracking results, the system performs boundary labeling on each of the images to be labeled to obtain the labeled image. Step S24: After all the images to be labeled are labeled, the automatic labeling system generates and outputs a labeled video containing boundary markers of contaminated soil based on each labeled image.

8. The automatic annotation method according to claim 7, characterized in that, In step S2, when performing feature extraction and frame-by-frame tracking on each of the images to be labeled based on the starting tracking frame, the similarity between the soil feature points of the starting tracking frame and the adjacent images to be labeled is taken as the frame-by-frame tracking result of the adjacent images to be labeled, and the image to be labeled is taken as the starting tracking frame.

9. The automatic annotation method according to claim 7, characterized in that, Step S24 includes: Step S241: For the ascending sequence, the automatic annotation system performs video synthesis on the labeled images corresponding to each image to be annotated in the ascending sequence according to the time order of the ascending sequence to obtain the first video. Step S242, for the reverse sequence, the automatic annotation system performs video synthesis of the labeled images corresponding to each image to be annotated in the reverse sequence according to the time order of the reverse sequence to obtain a second video; In step S243, the automatic annotation system combines the first video and the second video using the standard image as the video connection point to create the annotated video and outputs it.

10. The automatic annotation method according to claim 8, characterized in that, Steps S22 and S23 further include determining whether the similarity is greater than the similarity threshold: If so, then boundary annotation is performed on each of the images to be annotated to obtain the corresponding annotated image; If not, the operator is given a corresponding prompt, and then the boundary of the image to be labeled adjacent to the starting tracking frame is marked as the corresponding labeled image according to the feedback external instructions.

Images