Semantic segmentation pseudo label generation method for identifying road surface strip repair area and application thereof
By processing grayscale images of road surfaces to generate pseudo-labels that clearly distinguish striped repair areas from the background, the problem of high-cost labeling is solved, high-quality pseudo-labels are generated, and the recognition accuracy of deep learning models is improved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTHWEST FORESTRY UNIVERSITY
- Filing Date
- 2025-06-09
- Publication Date
- 2026-06-05
AI Technical Summary
In existing technologies, the semantic segmentation labeling for identifying patchy repair areas on asphalt roads is costly, and the quality of pseudo-labels is difficult to match that of real labels. There is a lack of pseudo-label generation schemes for identifying patchy repair areas on asphalt roads.
By performing grayscale mean normalization, median filtering, threshold segmentation, and pixel value inversion on the initial road surface grayscale image, combined with multi-size downsampling, median filtering, and image fusion, a semantic segmentation pseudo-label with the same size as the initial road surface grayscale image is generated. The pixel coordinates are determined by the bounding box and set to zero, generating a pseudo-label that clearly distinguishes the strip repair area from the background.
It reduces the cost of label annotation, generates pseudo-labels of near-real label quality, significantly reduces manual annotation time and cost, and improves the recognition accuracy of deep learning models for strip-shaped patch regions.
Smart Images
Figure CN120543978B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of image processing technology, and in particular to a method for generating semantic segmentation pseudo-labels for recognizing noodle-like repair areas on roads and its application. Background Technology
[0002] Asphalt pavements are prone to developing strip-shaped repairs over prolonged use. These repairs refer to the strip-shaped areas formed after patching existing cracks in the pavement. For pavements, the two main types of defects are strip-shaped repairs and strip cracks. While pavements with strip cracks pose a greater risk, strip-shaped repairs occur more frequently, indicating a higher repair frequency and continued potential road hazards. Because strip-shaped repair areas are similar in color to the intact pavement, they are not easily visible to the naked eye. However, with the development of deep learning technology, some studies have applied it to the identification of strip-shaped repair areas on asphalt pavements, replacing the relatively inefficient method of manual on-site inspection and recording.
[0003] Currently, semantic segmentation based on deep learning is one of the mainstream methods for automatically identifying strip-shaped repair defects. However, the creation of semantic segmentation labels is very labor-intensive and time-consuming. Compared with classification tasks and object detection tasks, the labeling cost of semantic segmentation labels is higher.
[0004] In related technical solutions, existing solutions reduce the workload of labeling by generating pseudo-labels that are similar to real labels. However, the inventors found in the process of conceiving and implementing this application that the quality of pseudo-labels generated by the commonly used pseudo-label generation methods is difficult to match that of real labels, and there is a lack of pseudo-label generation solutions for identification scenarios of patchy repair areas on asphalt roads.
[0005] Therefore, this application aims to propose a pseudo-label generation method for semantic segmentation of strip repair of road surface defects, so as to reduce the labeling cost. Summary of the Invention
[0006] The main purpose of this application is to provide a method for generating semantic segmentation pseudo-labels for identifying noodle-like repair areas on roads, aiming to solve the problem of how to reduce the cost of label annotation.
[0007] To achieve the above objectives, this application provides a method for generating semantic segmentation pseudo-labels for identifying noodle-like repair areas on roads, the method comprising:
[0008] Obtain an initial road surface grayscale image, wherein the initial road surface grayscale image includes at least one strip-shaped repair marked with a bounding box, and the image size is a preset size;
[0009] The initial road surface grayscale image is subjected to grayscale mean normalization and median filtering in sequence to obtain the first image;
[0010] The first image is sequentially subjected to threshold segmentation and pixel value inversion to obtain the second image;
[0011] The second image is subjected to multi-size downsampling, median filtering, and image fusion to obtain a third image with the same size as the initial road surface grayscale image.
[0012] Determine the coordinates of the pixels within and including the bounding box associated with the stripe patch in the third image;
[0013] In the third image, the pixel values corresponding to pixels other than the pixel coordinates are set to zero to obtain semantic segmentation pseudo-labels for identifying the noodle-like repair area on the road.
[0014] Optionally, the step of performing multi-size downsampling, median filtering, and image fusion on the second image to obtain a third image with the same size as the initial road surface grayscale image includes:
[0015] The second image is modified into at least two downsampled images of different sizes, and each downsampled image is processed with a sliding window of at least two different sizes;
[0016] The downsampled images after median filtering are upsampled and restored to the same size as the second image.
[0017] The non-zero pixel values in the downsampled image after upsampling are set as the target pixel values, and then the downsampled images after upsampling are fused into a third image with the same size as the initial road surface grayscale image according to their respective preset weights.
[0018] Optionally, the target pixel value is 128.
[0019] Optionally, the downsampled image includes two downsampled images with sizes of 256*256 and 128*128, wherein the sliding window size corresponding to the 256*256 downsampled image is 7*7, and the sliding window size corresponding to the two 128*128 downsampled images is 3*3.
[0020] Optionally, the first image is obtained by processing a 5*5 sliding window.
[0021] In addition, to achieve the above objectives, this application also provides a semantic segmentation pseudo-label generated using the semantic segmentation pseudo-label generation method for identifying road surface patch areas as described above.
[0022] Furthermore, to achieve the above objectives, this application also provides an application of semantic segmentation pseudo-labels generated using the semantic segmentation pseudo-label generation method for identifying road surface patch regions as described above in deep learning model training.
[0023] In addition, to achieve the above objectives, this application also provides a computer system, the computer system comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, it implements the steps of the semantic segmentation pseudo-label generation method for identifying road surface patch areas as described above.
[0024] In addition, to achieve the above objectives, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the semantic segmentation pseudo-label generation method for identifying road surface patch areas as described in any of the preceding claims.
[0025] This application has at least the following beneficial effects:
[0026] By generating semantic segmentation pseudo-labels that clearly distinguish the strip-shaped repair area from the background road surface area, the labeling cost is reduced, thus facilitating the deep learning model to learn the strip-shaped repair features in the semantic segmentation pseudo-labels. Attached Figure Description
[0027] Figure 1 This is a flowchart illustrating the first embodiment of the semantic segmentation pseudo-label generation method for identifying road surface patch areas according to this application.
[0028] Figure 2 This is a schematic diagram of the image processing flow for identifying semantic segmentation pseudo-labels for road surface patch areas, as described in an embodiment of this application.
[0029] Figure 3 This is a schematic diagram comparing the semantic segmentation pseudo-labels involved in the embodiments of this application with the initial road surface grayscale image;
[0030] Figure 4 This is a schematic diagram comparing the semantic segmentation pseudo-labels involved in the embodiments of this application with the initial road surface grayscale image;
[0031] Figure 5 This is a schematic diagram comparing the semantic segmentation pseudo-labels involved in the embodiments of this application with the initial road surface grayscale image;
[0032] Figure 6 This is a comparison chart of the segmentation accuracy of U-net trained on weakly supervised and fully supervised datasets in the embodiments of this application on the test set;
[0033] Figure 7This is a schematic diagram of the hardware operating environment of the computer system involved in the embodiments of this application.
[0034] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0035] To better understand the above technical solutions, exemplary embodiments of this disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of this disclosure to those skilled in the art. Example
[0036] Reference Figure 1 The flowchart shown is a semantic segmentation pseudo-label generation method for identifying noodle-shaped repair areas according to this application. In this embodiment, the method includes the following steps:
[0037] Step S10: Obtain an initial road surface grayscale image, wherein the initial road surface grayscale image contains at least one strip-shaped repair and the size is a preset size;
[0038] In this step, the initial road surface grayscale image refers to the original road surface image after grayscale processing, which has not undergone the image processing method in this embodiment. It serves as a template for generating pseudo-labels. This image contains at least one strip-shaped repair and its size is a preset fixed value.
[0039] Strip repair refers to the strip-shaped area formed after repairing strip-shaped cracks in the road surface. For road surfaces, the two main types of defects include strip repair and strip cracks. While road surfaces with strip cracks pose a greater risk, strip repair occurs more frequently, indicating a higher repair frequency and the continued existence of potential road hazards. Compared to strip cracks, strip repair areas are more difficult to distinguish from the background road surface when identified by deep learning models because their color is similar to that of the intact road surface.
[0040] It should be noted that the preset size is a size value that is easy to input into the deep learning model for training. The purpose is to ensure that the pseudo-labels generated after processing by the semantic segmentation pseudo-label generation method in this embodiment are the same as the preset size, so as to facilitate input into the deep learning model.
[0041] In some alternative implementations, the preset size is 512*512.
[0042] Step S20: Perform grayscale mean averaging and median filtering on the initial road surface grayscale image sequentially to obtain the first image;
[0043] In this step, the initial road surface grayscale image is first averaged and then subjected to median filtering. The purpose of this step is to distinguish the striped patch in the image from the background.
[0044] In some alternative implementations, the specific steps of grayscale averaging are as follows: calculate the neighborhood average of each pixel and replace the center pixel value.
[0045] In some optional implementations, the specific steps of median filtering are as follows: select a sliding window of a preset size to divide the initial road surface grayscale image, sort all pixel values within the window, and then replace the value of the center pixel of the sliding window with the median value.
[0046] Further, and optionally, the first image is obtained by processing a 5*5 sliding window.
[0047] Step S30: The first image is subjected to threshold segmentation and pixel value inversion in sequence to obtain the second image;
[0048] In this step, the first image is first thresholded to further distinguish the striped patch in the image from the background.
[0049] In some optional implementations, the specific steps of threshold segmentation are as follows: Pixels with grayscale values in the [0:20] range are uniformly set to 0; pixels with grayscale values in the [20:30] range are uniformly set to 42; pixels with grayscale values in the [30:40] range are uniformly set to 85; pixels with grayscale values in the [40:50] range are uniformly set to 170; and the remaining unprocessed pixels are uniformly set to a pixel value of 255. After the above threshold segmentation operation, the darker areas in the image will be segmented and retained, while areas that are not dark enough will be uniformly set to the maximum grayscale value of 255.
[0050] Subsequently, due to the processing in the previous steps, the color of the stripe patch in the first image after threshold segmentation is darker than that of the background image. In order to facilitate the deep learning model to identify the target recognition region we expect, the pixel values of the first image after threshold segmentation are inverted.
[0051] In some alternative implementations, the specific steps for inverting pixel values are: subtracting the pixel value of each pixel in the image from 255.
[0052] Step S40: Perform multi-size downsampling, median filtering, and image fusion on the second image to obtain a third image with the same size as the initial road surface grayscale image;
[0053] In this step, to ensure that the noise level in the generated pseudo-labels is low enough, the second image is subjected to three processing actions in sequence: multi-size downsampling, median filtering, and image fusion.
[0054] In some alternative implementations, this step specifically includes:
[0055] Step S41: Modify the second image into at least two downsampled images of different sizes, and process each downsampled image with at least two sliding windows of different sizes;
[0056] Step S42: Upsample the downsampled image after median filtering to restore it to the same size as the second image.
[0057] Step S43: Set the non-zero pixel values in the downsampled image after upsampling to the target pixel values, and then fuse the downsampled image after upsampling into a third image with the same size as the initial road surface grayscale image according to their respective preset weights.
[0058] Further and optionally, the target pixel value is 128. In this case, there are only two different pixel values in the image: 0 and 128. 0 belongs to the background, and 128 belongs to the foreground (and the strip-shaped repair portion that serves as the recognition area).
[0059] Further and optionally, the downsampled image includes two downsampled images with sizes of 256*256 and 128*128, wherein the sliding window size corresponding to the 256*256 downsampled image is 7*7, and the sliding window size corresponding to the two 128*128 downsampled images is 3*3.
[0060] Step S50: Determine the coordinates of the pixels within and including the bounding box associated with the stripe patch in the third image;
[0061] Step S60: Set the pixel values of all pixels in the third image except for the pixel coordinates to zero to obtain a semantic segmentation pseudo-label for identifying the road surface patch area.
[0062] In this embodiment, the region in the image that serves as the strip-shaped repair area is marked using bounding boxes. This marking action is completed before step S50. When executing step S50, the previously marked bounding boxes of the third image are obtained, the coordinates of all pixels within the bounding boxes and including the bounding boxes themselves are determined, the pixel values of these pixel coordinates are retained, and the pixel values corresponding to pixels outside these pixel coordinates are set to zero, thus obtaining a semantic segmentation pseudo-label for identifying the strip-shaped repair area of the road.
[0063] Exemplarily, in some specific embodiments, reference is made to Figure 2 The diagram shows the image processing flow for semantic segmentation pseudo-labels used to identify noodle-like repair areas on the road. In the final semantic segmentation pseudo-labels, it can be seen that after a series of processing steps, the areas with noodle-like repairs are clearly distinguished from the background road surface areas.
[0064] In the technical solution provided in this embodiment, semantic segmentation pseudo-labels that clearly distinguish the strip-shaped repair area from the background road surface area are generated to reduce the labeling cost, thereby facilitating the deep learning model to learn the strip-shaped repair features in the semantic segmentation pseudo-labels. Example
[0065] As one implementation, this embodiment provides a semantic segmentation pseudo-label generated based on the semantic segmentation pseudo-label generation method for identifying road surface patch areas described in the first embodiment.
[0066] Exemplarily, in some specific embodiments, reference is made to Figure 3 , Figure 4 and Figure 5 The diagrams show a comparison between semantic segmentation pseudo-labels and the initial road surface grayscale image. The first row in the diagrams shows the semantic segmentation pseudo-labels generated after processing, while the second row shows the initial road surface grayscale image. Example
[0067] As one implementation scheme, this embodiment provides an application of semantic segmentation pseudo-labels generated in the second embodiment in deep learning model training.
[0068] Specifically, the training of a deep learning model follows these steps:
[0069] Step S100: Prepare a real label dataset with the same size as the pseudo-label dataset, with a training set to validation set ratio of 8:1 for both datasets. The images and corresponding labels in the real label dataset are both 512*512 pixels, consistent with the image size in the pseudo-label dataset.
[0070] Step S200: Set training parameters, including batch size, learning rate, number of iterations, etc., and train the semantic segmentation model using two datasets with different labels respectively to obtain fully supervised and weakly supervised strip-patching semantic segmentation models; the specific settings of the training hyperparameters are shown in Table 1 below:
[0071] Table 1. Specific settings for training hyperparameters
[0072]
[0073] Step S300: Construct a separate test set consisting of dataset images and corresponding ground truth labels, and test the segmentation accuracy of the fully supervised model and the weakly supervised model separately. Both datasets are trained using the U-net benchmark model, and then a comparative analysis is performed. The comparison of segmentation accuracy between the weakly supervised and fully supervised datasets trained on U-net is shown in the figure below. Figure 6 As shown.
[0074] Step S400: From the comparison graph of the segmentation accuracy of weakly supervised and fully supervised semantic segmentation models on the test set, it can be seen that, taking U-Net using the VGG16 backbone as the benchmark model, on the same test set with 100 images and real labels, the MIoU, mPA / mRecall, and mPrecision obtained by fully supervised training are 86.51%, 97.16%, and 88.29%, respectively, while the MIoU, mPA / mRecall, and mPrecision obtained by weakly supervised training are 84.62%, 91.66%, and 90.3%, respectively. From the most commonly used metric for measuring the segmentation accuracy of semantic segmentation models, MIoU, the U-Net trained with pseudo-labels under weak supervision is only 1.89% lower than the U-Net trained with real labels under full supervision, a very small difference. Meanwhile, if the segmentation accuracy on the test set is used to measure the quality difference between the pseudo-label and the real label, then the effect of the pseudo-label of the present invention has reached about 98.8% of the real label. It can be basically considered that the quality of the pseudo-label and the real label is almost the same, and it even has the potential to replace the real label of manual annotation, and significantly reduce the annotation time and labor cost.
[0075] As one implementation scheme, Figure 7 This is a schematic diagram of the hardware operating environment of the computer system involved in the embodiments of this application.
[0076] like Figure 7 As shown, the computer system may include: a processor 1001, such as a CPU; a memory 1005; a user interface 1003; a network interface 1004; and a communication bus 1002. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be high-speed RAM or non-volatile memory, such as a disk drive. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.
[0077] Those skilled in the art will understand that Figure 7 The computer system architecture shown does not constitute a limitation on the computer system and may include more or fewer components than shown, or combine certain components, or have different component arrangements.
[0078] like Figure 7 As shown, the memory 1005, as a storage medium, may include an operating system, a network communication module, a user interface module, and computer programs. The operating system is a program that manages and controls the hardware and software resources of the computer system, as well as the operation of the computer programs and other software or programs.
[0079] exist Figure 7 In the computer system shown, the user interface 1003 is mainly used to connect to the terminal and communicate with the terminal; the network interface 1004 is mainly used to communicate with the backend server; and the processor 1001 can be used to call the computer program stored in the memory 1005.
[0080] In this embodiment, the computer system includes: a memory 1005, a processor 1001, and a computer program stored in the memory and executable on the processor, wherein:
[0081] When processor 1001 calls a computer program stored in memory 1005, it performs the following operations:
[0082] Obtain an initial road surface grayscale image, wherein the initial road surface grayscale image includes at least one strip-shaped repair marked with a bounding box, and the image size is a preset size;
[0083] The initial road surface grayscale image is subjected to grayscale mean normalization and median filtering in sequence to obtain the first image;
[0084] The first image is sequentially subjected to threshold segmentation and pixel value inversion to obtain the second image;
[0085] The second image is subjected to multi-size downsampling, median filtering, and image fusion to obtain a third image with the same size as the initial road surface grayscale image.
[0086] Determine the coordinates of the pixels within and including the bounding box associated with the stripe patch in the third image;
[0087] In the third image, the pixel values corresponding to pixels other than the pixel coordinates are set to zero to obtain semantic segmentation pseudo-labels for identifying the noodle-like repair area on the road.
[0088] When processor 1001 calls a computer program stored in memory 1005, it performs the following operations:
[0089] The second image is modified into at least two downsampled images of different sizes, and each downsampled image is processed with a sliding window of at least two different sizes;
[0090] The downsampled images after median filtering are upsampled and restored to the same size as the second image.
[0091] The non-zero pixel values in the downsampled image after upsampling are set as the target pixel values, and then the downsampled images after upsampling are fused into a third image with the same size as the initial road surface grayscale image according to their respective preset weights.
[0092] Furthermore, those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in a computer system to implement the process steps of the embodiments of the above methods.
[0093] Therefore, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the various steps of the semantic segmentation pseudo-label generation method for identifying road surface patch areas as described in the above embodiments.
[0094] The computer-readable storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.
[0095] It should be noted that, since the storage medium provided in the embodiments of this application is the storage medium used to implement the methods of the embodiments of this application, those skilled in the art can understand the specific structure and variations of the storage medium based on the methods described in the embodiments of this application, and therefore will not be repeated here. All storage media used in the methods of the embodiments of this application fall within the scope of protection of this application.
[0096] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0097] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0098] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0099] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0100] It should be noted that any reference signs placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. This application can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.
[0101] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.
[0102] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.
Claims
1. A method for generating semantic segmentation pseudo-labels for identifying noodle-like repair areas on roads, characterized in that, The method includes the following steps: Obtain an initial road surface grayscale image, wherein the initial road surface grayscale image includes at least one strip-shaped repair marked with a bounding box, and the image size is a preset size; The initial road surface grayscale image is subjected to grayscale mean normalization and median filtering in sequence to obtain the first image; The first image is sequentially subjected to threshold segmentation and pixel value inversion to obtain the second image; The second image is subjected to multi-size downsampling, median filtering, and image fusion to obtain a third image with the same size as the initial road surface grayscale image. Determine the coordinates of the pixels within and including the bounding box associated with the stripe patch in the third image; In the third image, the pixel values corresponding to the pixels other than the pixel coordinates are set to zero to obtain semantic segmentation pseudo-labels for identifying the noodle-like repair area on the road. The step of performing multi-size downsampling, median filtering, and image fusion on the second image to obtain a third image with the same size as the initial road surface grayscale image includes: The second image is modified into at least two downsampled images of different sizes, and each downsampled image is processed with a sliding window of at least two different sizes; The downsampled images after median filtering are upsampled and restored to the same size as the second image. The non-zero pixel values in the downsampled image after upsampling are set as the target pixel values, and then the downsampled images after upsampling are fused into a third image with the same size as the initial road surface grayscale image according to their respective preset weights.
2. The method as described in claim 1, characterized in that, The target pixel value is 128.
3. The method as described in claim 1, characterized in that, The downsampled images include two downsampled images with sizes of 256*256 and 128*128. The sliding window size corresponding to the 256*256 downsampled image is 7*7, and the sliding window size corresponding to the 128*128 downsampled image is 3*3.
4. The method as described in claim 1, characterized in that, The first image was obtained by processing a 5x5 sliding window.
5. A semantic segmentation pseudo-label generated using the semantic segmentation pseudo-label generation method for identifying road surface patch areas as described in any one of claims 1 to 4.
6. An application of a semantic segmentation pseudo-label generated by the semantic segmentation pseudo-label generation method for identifying road surface patch regions as described in any one of claims 1 to 4 in deep learning model training.
7. A computer system, characterized in that, The computer system includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, it implements the steps of the semantic segmentation pseudo-label generation method for identifying road surface patch areas as described in any one of claims 1 to 4.
8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the semantic segmentation pseudo-label generation method for identifying road surface patch areas as described in any one of claims 1 to 4.