Belt tearing detection method and system based on semantic segmentation

Abstract

The invention discloses a semantic segmentation-based belt tearing detection method and system. The method comprises the following steps of training a UNet segmentation network model by using a model training data set; obtaining a to-be-processed laser ray picture; carrying out slicing processing on the to-be-processed laser ray picture; inputting the plurality of laser line sub-pictures into the trained UNet segmentation network model to carry out parallel laser line contour segmentation processing; performing contour extraction on the plurality of binarized laser line contour segmentation maps to obtain corresponding laser line contours, and constructing convex hulls on the plurality of laser line contours; and selecting a convex hull point with the minimum vertical coordinate value on each laser line contour convex hull. According to the method, the problems that in an existing method for carrying out semantic segmentation on the collected laser line picture through a coding and decoding neural network, when the resolution ratio of the collected laser line picture is large, consumed time for segmentation processing of the coding and decoding neural network is too long, and belt tearing cannot be detected in real time are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of belt tear detection, and in particular to a belt tear detection method and system based on semantic segmentation. Background Art

[0002] Nowadays, the application of segmentation technology is becoming more and more extensive. For the desired object contour, the specific contour can be accurately segmented through segmentation technology. In the project of detecting whether the belt is torn, the laser line on the belt is first segmented by segmentation technology, and then it is judged whether there is a bulge caused by the belt tearing on the segmented laser line, so as to achieve the purpose of detecting whether the belt is torn. There is a prior art that discloses semantic segmentation of the collected laser line image by using a codec neural network to achieve the segmentation of the laser line on the belt. However, in this method, when the resolution of the collected laser line image is relatively large, the time required for the codec neural network segmentation processing will be too long, and it will be impossible to detect the belt tear in real time. In this way, the belt tear point has not been detected yet, and the belt tear point on it has flowed away during the rotation of the belt, which will result in a very large missed detection phenomenon. Summary of the invention

[0003] In view of the above-mentioned defects, the present invention proposes a belt tear detection method and system based on semantic segmentation, aiming to solve the problem that in the existing method of semantic segmentation of collected laser line images by using a codec neural network, when the resolution of the collected laser line images is relatively large, the time required for the codec neural network segmentation processing will be too long, and the belt tear cannot be detected in real time, thus resulting in missed detection.

[0004] To achieve this object, the present invention adopts the following technical solutions:

[0005] A belt tear detection method based on semantic segmentation includes the following steps:

[0006] Step S1: Establish model training data set and U_Net segmentation network model;

[0007] Step S2: Use the model training data set to train the U_Net segmentation network model to obtain a trained U_Net segmentation network;

[0008] Step S3: Obtaining the laser line image to be processed;

[0009] Step S4: Slice the laser line image to be processed to obtain a number of laser line sub-images;

[0010] Step S5: inputting a plurality of laser line sub-images into the trained U_Net segmentation network model to perform parallel laser line contour segmentation processing, and outputting the corresponding binary laser line contour segmentation map;

[0011] Step S6: performing contour extraction on a plurality of binary laser line contour segmentation images respectively to obtain corresponding laser line contours, and constructing convex hulls on the plurality of laser line contours respectively to obtain corresponding laser line contour convex hulls;

[0012] Step S7: Select the convex point with the smallest ordinate value on the convex hull of each laser line profile to obtain the belt tearing point.

[0013] Preferably, in step S1, a model training data set is established, which specifically includes the following sub-steps:

[0014] Step S11: Collect 400 to 600 laser line images and create a blank file;

[0015] Step S12: using a labeling tool to label the position and category name of the laser line in each laser line image;

[0016] Step S13: Create a json file for each laser line image;

[0017] Step S14: storing the position and category name of the laser line in each laser line image in a json file corresponding to the laser line image;

[0018] Step S15: Store 400 to 600 laser line images and the json file of each laser line image in a blank file to complete the establishment of the model training data set.

[0019] Preferably, in step S6, contour extraction is performed on the binary laser line contour segmentation image, which specifically includes the following sub-steps:

[0020] Step S61: Use the contour extraction function fingContours() of OpenCV to extract the laser line contour set conts={C1, C2, C i , …, C n};

[0021] Among them, C i represents the i-th laser line profile, C i ={topx i ,topy i , bottomx i , bottomy i , S i}, topx i Indicates the horizontal coordinate of the upper left corner of the i-th position rectangle, topyi Indicates the upper left corner ordinate of the i-th position rectangle, bottomx i Represents the horizontal coordinate of the lower right corner of the i-th position rectangle, bottomy i Indicates the ordinate of the lower right corner of the i-th position rectangle, S i represents the area of ​​the rectangular box at the ith position; n represents the number of all laser line profiles in the binary laser line profile segmentation image;

[0022] Step S62: selecting the laser line profile with the largest area in the laser line profile set conts in the binary laser line profile segmentation map as the laser line profile corresponding to the binary laser line profile segmentation map.

[0023] Preferably, in step S6, a convex hull is constructed on the laser line profile to obtain the convex hull of the laser line profile, which specifically includes the following sub-steps: a convex hull coordinate point set of the laser line profile is calculated on the laser line profile by using the convex hull function convexHull() of OpenCV.

[0024] Another aspect of the present application provides a belt tear detection system based on semantic segmentation, the system comprising:

[0025] The first building module is used to build a model training data set;

[0026] The second building module is used to build a U Net segmentation network model;

[0027] A model training module is used to train the U Net segmentation network model using a model training data set to obtain a trained U Net segmentation network;

[0028] An acquisition module, used to acquire the laser line image to be processed;

[0029] An image slicing module is used to slice the laser line image to be processed to obtain a number of laser line sub-images;

[0030] An input module is used to input a number of laser line sub-images into the trained U_Net segmentation network model for parallel laser line contour segmentation processing;

[0031] An output module, used for outputting a corresponding binary laser line contour segmentation map;

[0032] A contour extraction module is used to extract contours of several binary laser line contour segmentation images respectively to obtain corresponding laser line contours;

[0033] A construction module, used for constructing convex hulls on several laser line profiles respectively to obtain corresponding convex hulls of the laser line profiles;

[0034] The selection module is used to select the convex point with the smallest ordinate value on the convex hull of each laser line profile to obtain the belt tearing point.

[0035] Preferably, the first establishing module comprises:

[0036] The acquisition submodule is used to collect 400 to 600 laser line images;

[0037] Create a submodule to create a blank file;

[0038] The annotation submodule is used to annotate the position and category name of the laser line in each laser line image using an annotation tool;

[0039] Create a submodule to create a json file for each laser line image;

[0040] A first storage submodule is used to store the position and category name of the laser line in each laser line image in a json file corresponding to the laser line image;

[0041] The second storage submodule is used to store 400 to 600 laser line images and the json file of each laser line image in a blank file to complete the establishment of the model training data set.

[0042] Preferably, the contour extraction module comprises:

[0043] The extraction submodule is used to extract the laser line contour set conts = {C1, C2, C i ,…,C n};

[0044] Among them, C i represents the i-th laser line profile, C i ={topx i ,topy i , bottomx i , bottomy i , S i}, topx i Indicates the horizontal coordinate of the upper left corner of the i-th position rectangle, topy i Indicates the upper left corner ordinate of the i-th position rectangle, bottomx i Represents the horizontal coordinate of the lower right corner of the i-th position rectangle, bottomy i Indicates the ordinate of the lower right corner of the i-th position rectangle, S i represents the area of ​​the rectangular box at the ith position; n represents the number of all laser line profiles in the binary laser line profile segmentation image;

[0045] The selection submodule is used to select the laser line profile with the largest area in the laser line profile set conts in the binary laser line profile segmentation map as the laser line profile corresponding to the binary laser line profile segmentation map.

[0046] Preferably, the building blocks include:

[0047] The calculation submodule is used to calculate the convex hull coordinate point set of the laser line profile through the convex hull function convexHull() of OpenCV on the laser line profile.

[0048] The technical solution provided by the embodiments of the present application may have the following beneficial effects:

[0049] This solution slices the collected laser line image into several laser line sub-images, and uses the trained UNet segmentation network model to perform parallel laser line contour segmentation on all laser line sub-images to obtain the laser line contour segmentation map corresponding to each laser line sub-image. Compared with the traditional method of using an encoding and decoding neural network to perform laser line segmentation on an entire laser line image, this solution can improve the timeliness of the laser line contour in the segmented image, thereby realizing real-time detection of belt tearing and avoiding missed detection of belt tearing points. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flowchart of the steps of a belt tear detection method based on semantic segmentation. DETAILED DESCRIPTION

[0051] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions from beginning to end. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0052] A belt tear detection method based on semantic segmentation includes the following steps:

[0053] Step S1: Establish model training data set and U_Net segmentation network model;

[0054] Step S2: Use the model training data set to train the U_Net segmentation network model to obtain a trained U_Net segmentation network;

[0055] Step S3: Obtaining the laser line image to be processed;

[0056] Step S4: Slice the laser line image to be processed to obtain a number of laser line sub-images;

[0057] Step S5: inputting a plurality of laser line sub-images into the trained U_Net segmentation network model to perform parallel laser line contour segmentation processing, and outputting the corresponding binary laser line contour segmentation map;

[0058] Step S6: performing contour extraction on a plurality of binary laser line contour segmentation images respectively to obtain corresponding laser line contours, and constructing convex hulls on the plurality of laser line contours respectively to obtain corresponding laser line contour convex hulls;

[0059] Step S7: Select the convex point with the smallest ordinate value on the convex hull of each laser line profile to obtain the belt tearing point.

[0060] This scheme is a belt tear detection method based on semantic segmentation, such as Figure 1As shown, the first step is to establish a model training data set and a U_Net segmentation network model. In this embodiment, the establishment of the model training data set is conducive to the subsequent training of the U_Net segmentation network model. The establishment of the U_Net segmentation network model is conducive to the subsequent segmentation of the laser line on the belt. It is further explained that the U_Net segmentation network model is a deep learning network model commonly used for image segmentation tasks. The second step is to train the U_Net segmentation network model using the model training data set to obtain the trained U_Net segmentation network. In this embodiment, the U_Net segmentation network model is trained using the model training data set, which is conducive to improving the accuracy of the U_Net segmentation network model in laser line segmentation. The third step is to obtain the laser line image to be processed. In this embodiment, at the belt tear detection site, a camera is set to collect and obtain the laser line image. The fourth step is to slice the laser line image to be processed to obtain several laser line sub-images. In this embodiment, by slicing the laser line image into four laser line sub-images, it is convenient for the U_Net segmentation network model after subsequent training to perform parallel laser line contour segmentation processing on the four laser line sub-images, thereby improving the timeliness of the trained U_Net segmentation network model in segmenting the laser line. The fifth step is to input several laser line sub-images into the trained U_Net segmentation network model for parallel laser line contour segmentation processing, and output the corresponding binary laser line contour segmentation map. In this embodiment, the trained U_Net segmentation network model can accurately segment the laser line contour segmentation map of each laser line sub-image. In addition, the binary laser line contour segmentation map not only reduces the amount of calculation for subsequent image processing, but also highlights the key features in the image, which is conducive to the subsequent extraction of key features. The sixth step is to extract the contours of several binary laser line contour segmentation images respectively to obtain the corresponding laser line contours, and construct convex hulls on several laser line contours respectively to obtain the corresponding laser line contour convex hulls. In this embodiment, by extracting the laser line contour features in the binary laser line contour segmentation image, it is beneficial to the subsequent analysis and processing of the laser line contour. Constructing a convex hull on the laser line contour can further reduce the complexity and redundant information of the data. The seventh step is to select the convex hull point with the smallest ordinate value on each laser line contour convex hull to obtain the belt tearing point. In this embodiment, the laser line contour convex hull is composed of several convex hull points. By finding the convex hull point with the smallest ordinate value in the laser line contour convex hull, the position of the belt tearing point can be determined, thereby realizing the detection of belt tearing.

[0061] This solution slices the collected laser line image into several laser line sub-images, and uses the trained U_Net segmentation network model to perform parallel laser line contour segmentation on all laser line sub-images to obtain the laser line contour segmentation map corresponding to each laser line sub-image. Compared with the traditional laser line segmentation process of a whole laser line image using a codec neural network, this solution can improve the timeliness of the laser line contour in the segmented image, thereby realizing real-time detection of belt tearing and avoiding missed detection of belt tearing points.

[0062] Preferably, in step S1, a model training data set is established, which specifically includes the following sub-steps:

[0063] Step S11: Collect 400 to 600 laser line images and create a blank file;

[0064] Step S12: using a labeling tool to label the position and category name of the laser line in each laser line image;

[0065] Step S13: Create a json file for each laser line image;

[0066] Step S14: storing the position and category name of the laser line in each laser line image in a json file corresponding to the laser line image;

[0067] Step S15: Store 400 to 600 laser line images and the json file of each laser line image in a blank file to complete the establishment of the model training data set.

[0068] In this embodiment, the annotation tool is a SAM automatic segmentation annotation tool. By using the SAM automatic segmentation annotation tool to automatically annotate each laser line image, the annotation efficiency can be greatly improved and the time and cost of manual annotation can be reduced. The model training data set consisting of 400 to 600 laser line images and the json file of each laser line image is a high-quality data set that can ensure the accuracy of the model output results.

[0069] Preferably, in step S6, contour extraction is performed on the binary laser line contour segmentation image, which specifically includes the following sub-steps:

[0070] Step S61: Use the contour extraction function fingContours() of OpenCV to extract the laser line contour set conts={C1, C2, C i , …, C n};

[0071] Among them, C i represents the i-th laser line profile, C i ={topxi ,topy i , bottomx i , bottomy i , S i}, topx i Indicates the horizontal coordinate of the upper left corner of the i-th position rectangle, topy i Indicates the upper left corner ordinate of the i-th position rectangle, bottomx i Represents the horizontal coordinate of the lower right corner of the i-th position rectangle, bottomy i Indicates the ordinate of the lower right corner of the i-th position rectangle, S i represents the area of ​​the rectangular box at the ith position; n represents the number of all laser line profiles in the binary laser line profile segmentation image;

[0072] Step S62: selecting the laser line profile with the largest area in the laser line profile set conts in the binary laser line profile segmentation map as the laser line profile corresponding to the binary laser line profile segmentation map.

[0073] In this embodiment, in the process of extracting the laser line contour in the binary laser line contour segmentation map using the contour extraction function fingContours() provided by OpenCV, the so-called laser line contour is the white connected area in the binary laser line contour segmentation map, and each laser line contour can be represented by the four coordinates and area of ​​the position rectangular box. In the binary laser line contour segmentation map, the laser line contour with the largest area represents the most important information or target object in the image. Therefore, the laser line contour with the largest area in conts is selected as the laser line contour corresponding to the binary laser line contour segmentation map.

[0074] Preferably, in step S6, a convex hull is constructed on the laser line profile to obtain the convex hull of the laser line profile, specifically comprising the following sub-steps: the convex hull coordinate point set of the laser line profile is calculated by the convex hull function convexHull() of OpenCV on the laser line profile. In this embodiment, the convex hull function convexHull() provided by OpenCV can calculate the convex hull of the laser line profile composed of a number of coordinate points, and then the laser line profile is further analyzed and processed.

[0075] Another aspect of the present application provides a belt tear detection system based on semantic segmentation, the system comprising:

[0076] The first building module is used to build a model training data set;

[0077] The second building module is used to build the U_Net segmentation network model;

[0078] A model training module is used to train the U_Net segmentation network model using a model training data set to obtain a trained U_Net segmentation network;

[0079] An acquisition module, used to acquire the laser line image to be processed;

[0080] An image slicing module is used to slice the laser line image to be processed to obtain a number of laser line sub-images;

[0081] An input module is used to input a number of laser line sub-images into the trained U_Net segmentation network model for parallel laser line contour segmentation processing;

[0082] An output module, used for outputting a corresponding binary laser line contour segmentation map;

[0083] A contour extraction module is used to extract contours of several binary laser line contour segmentation images respectively to obtain corresponding laser line contours;

[0084] A construction module, used for constructing convex hulls on several laser line profiles respectively to obtain corresponding convex hulls of the laser line profiles;

[0085] The selection module is used to select the convex point with the smallest ordinate value on the convex hull of each laser line profile to obtain the belt tearing point.

[0086] The present invention is a belt tear detection system based on semantic segmentation, which realizes real-time detection of belt tearing through the cooperation of the first establishment module, the second establishment module, the model training module, the acquisition module, the image slicing module, the input module, the output module, the contour extraction module, the construction module and the selection module. The present invention slices the collected laser line image into several laser line sub-images, and uses the trained U_Net segmentation network model to perform parallel laser line contour segmentation processing on all laser line sub-images to obtain a laser line contour segmentation map corresponding to each laser line sub-image. Compared with the traditional method of using an encoding and decoding neural network to perform laser line segmentation processing on an entire laser line image, the present invention can improve the timeliness of the laser line contour in the segmented image, thereby realizing real-time detection of belt tearing and avoiding missed detection of belt tearing points.

[0087] Preferably, the first establishing module includes:

[0088] The acquisition submodule is used to collect 400 to 600 laser line images;

[0089] Create a submodule to create a blank file;

[0090] The annotation submodule is used to annotate the position and category name of the laser line in each laser line image using an annotation tool;

[0091] Create a submodule to create a json file for each laser line image;

[0092] A first storage submodule is used to store the position and category name of the laser line in each laser line image in a json file corresponding to the laser line image;

[0093] The second storage submodule is used to store 400 to 600 laser line images and the json file of each laser line image in a blank file to complete the establishment of the model training data set.

[0094] In this embodiment, in the annotation submodule, the annotation tool is the SAM automatic segmentation annotation tool. By using the SAM automatic segmentation annotation tool to automatically annotate each laser line image, the annotation efficiency can be greatly improved and the time and cost of manual annotation can be reduced. The production of the model training data set is achieved through the mutual cooperation of the acquisition submodule, the establishment submodule, the annotation submodule, the creation submodule, the first storage submodule and the second storage submodule. The model training data set consisting of 400 to 600 laser line images and the json file of each laser line image is a high-quality data set that can ensure the accuracy of the model output results.

[0095] Preferably, the contour extraction module comprises:

[0096] The extraction submodule is used to extract the laser line contour set conts = {C1, C2, C i , …, C n};

[0097] Among them, C i represents the i-th laser line profile, C i ={topx i ,topy i , bottomx i , bottomy i , S i}, topx i Indicates the horizontal coordinate of the upper left corner of the i-th position rectangle, topy i Indicates the upper left corner ordinate of the i-th position rectangle, bottomx i Represents the horizontal coordinate of the lower right corner of the i-th position rectangle, bottomy i Indicates the ordinate of the lower right corner of the i-th position rectangle, S i represents the area of ​​the rectangular box at the ith position; n represents the number of all laser line profiles in the binary laser line profile segmentation image;

[0098] The selection submodule is used to select the laser line profile with the largest area in the laser line profile set conts in the binary laser line profile segmentation map as the laser line profile corresponding to the binary laser line profile segmentation map.

[0099] In this embodiment, in the extraction submodule, the laser line contour in the binary laser line contour segmentation map is extracted using the contour extraction function fingContours() provided by OpenCV. The so-called laser line contour is the white connected area in the binary laser line contour segmentation map. Each laser line contour can be represented by the four coordinates and area of ​​the position rectangular box. In the binary laser line contour segmentation map, the laser line contour with the largest area represents the most important information or target object in the image, so the selection submodule is executed.

[0100] Preferably, the construction module includes: a calculation submodule, which is used to calculate the convex hull coordinate point set of the laser line profile through the convex hull function convexHull() of OpenCV on the laser line profile. In this embodiment, the convex hull function convexHull() provided by OpenCV can calculate the convex hull of the laser line profile composed of a number of coordinate points, and then further analyze and process the laser line profile.

[0101] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0102] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A belt tear detection method based on semantic segmentation, characterized in that: The following steps are involved: Step S1: Establish model training data set and U_Net segmentation network model; Step S2: Use the model training data set to train the U_Net segmentation network model to obtain a trained U_Net segmentation network; Step S3: Obtaining the laser line image to be processed; Step S4: Slice the laser line image to be processed to obtain a number of laser line sub-images; Step S5: inputting a plurality of laser line sub-images into the trained U_Net segmentation network model to perform parallel laser line contour segmentation processing, and outputting the corresponding binary laser line contour segmentation map; Step S6: performing contour extraction on a plurality of binary laser line contour segmentation images respectively to obtain corresponding laser line contours, and constructing convex hulls on the plurality of laser line contours respectively to obtain corresponding laser line contour convex hulls; Step S7: Select the convex point with the smallest ordinate value on the convex hull of each laser line profile to obtain the belt tearing point.

2. The belt tear detection method based on semantic segmentation according to claim 1, characterized in that: In step S1, a model training data set is established, which specifically includes the following sub-steps: Step S11: Collect 400 to 600 laser line images and create a blank file; Step S12: using a labeling tool to label the position and category name of the laser line in each laser line image; Step S13: Create a json file for each laser line image; Step S14: storing the position and category name of the laser line in each laser line image in a json file corresponding to the laser line image; Step S15: Store 400 to 600 laser line images and the json file of each laser line image in a blank file to complete the establishment of the model training data set.

3. The belt tear detection method based on semantic segmentation according to claim 1, characterized in that: In step S6, contour extraction is performed on the binary laser line contour segmentation image, which specifically includes the following sub-steps: Step S61: Use the contour extraction function fingContours() of OpenCV to extract the laser line contour set conts={C1, C2, C i ,…,C n }; Among them, C i represents the i-th laser line profile, C i ={topx i ,topy i ,bottomx i ,bottomy i ,S i }, topx i Indicates the horizontal coordinate of the upper left corner of the i-th position rectangle, topy i Indicates the upper left corner ordinate of the i-th position rectangle, bottomx i Represents the horizontal coordinate of the lower right corner of the i-th position rectangle, bottomy i Indicates the ordinate of the lower right corner of the i-th position rectangle, S i represents the area of ​​the rectangular box at the ith position; n represents the number of all laser line profiles in the binary laser line profile segmentation image; Step S62: selecting the laser line profile with the largest area in the laser line profile set conts in the binary laser line profile segmentation map as the laser line profile corresponding to the binary laser line profile segmentation map.

4. The belt tear detection method based on semantic segmentation according to claim 1, characterized in that: In step S6, a convex hull is constructed on the laser line profile to obtain the convex hull of the laser line profile, which specifically includes the following sub-steps: on the laser line profile, a convex hull function convexHull() of OpenCV is used to calculate a set of convex hull coordinate points of the laser line profile.

5. A belt tear detection system based on semantic segmentation, characterized in that: Using the belt tear detection method based on semantic segmentation as described in any one of claims 1 to 4, the system includes: The first building module is used to build a model training data set; The second building module is used to build the U_Net segmentation network model; A model training module is used to train the U_Net segmentation network model using a model training data set to obtain a trained U_Net segmentation network; An acquisition module, used to acquire the laser line image to be processed; An image slicing module is used to slice the laser line image to be processed to obtain a number of laser line sub-images; An input module is used to input a number of laser line sub-images into the trained U_Net segmentation network model for parallel laser line contour segmentation processing; An output module, used for outputting a corresponding binary laser line contour segmentation map; A contour extraction module is used to extract contours of several binary laser line contour segmentation images respectively to obtain corresponding laser line contours; A construction module, used to construct convex hulls on several laser line profiles respectively to obtain corresponding convex hulls of the laser line profiles; The selection module is used to select the convex point with the smallest ordinate value on the convex hull of each laser line profile to obtain the belt tearing point.

6. The belt tear detection system based on semantic segmentation according to claim 5, characterized in that: The first establishment module comprises: The acquisition submodule is used to collect 400 to 600 laser line images; Create a submodule to create a blank file; The annotation submodule is used to annotate the position and category name of the laser line in each laser line image using an annotation tool; Create a submodule to create a json file for each laser line image; A first storage submodule is used to store the position and category name of the laser line in each laser line image in a json file corresponding to the laser line image; The second storage submodule is used to store 400 to 600 laser line images and the json file of each laser line image in a blank file to complete the establishment of the model training data set.

7. The belt tear detection system based on semantic segmentation according to claim 5, characterized in that: The contour extraction module comprises: The extraction submodule is used to extract the laser line contour set conts = {C1, C2, C i ,…,C n }; Among them, C i represents the i-th laser line profile, C i ={topx i ,topy i ,bottomx i ,bottomy i ,S i }, topx i Indicates the horizontal coordinate of the upper left corner of the i-th position rectangle, topy i Indicates the upper left corner ordinate of the i-th position rectangle, bottomx i Represents the horizontal coordinate of the lower right corner of the i-th position rectangle, bottomy i Indicates the ordinate of the lower right corner of the i-th position rectangle, S i represents the area of ​​the rectangular box at the ith position; n represents the number of all laser line profiles in the binary laser line profile segmentation image; The selection submodule is used to select the laser line profile with the largest area in the laser line profile set conts in the binary laser line profile segmentation map as the laser line profile corresponding to the binary laser line profile segmentation map.

8. The belt tear detection system based on semantic segmentation according to claim 5, characterized in that: The building blocks include: The calculation submodule is used to calculate the convex hull coordinate point set of the laser line profile through the convex hull function convexHull() of OpenCV on the laser line profile.

Images