A surface wear detection system, method and electronic device for curved surface structure

Abstract

The present application provides a surface wear detection system, method and electronic device for curved surface structures, which relates to the field of surface wear detection technology, including: an image acquisition module for multi-angle image acquisition of a target curved surface structure; a curvature determination module for determining a curvature information set; an image combination module for analyzing a curvature information set to perform curvature consistent clustering on a curved surface image set; a model construction module for constructing a wear feature recognition model; a wear recognition module for performing plane conversion on multiple groups of images to be detected with curvature information identifiers and then inputting them into a wear feature recognition channel for wear recognition to generate wear detection results. The present application can solve the technical problem in the prior art that the same wear feature has different forms in the image due to different curvatures in different regions, resulting in limited accuracy of wear detection. By combining curvature information, adaptive processing is performed on images with different curvatures, thereby improving the accuracy of wear detection.

Description

Technical Field

[0001] The present application relates to the technical field of surface wear detection, and in particular to a surface wear detection system, method and electronic equipment for curved surface structures. Background Art

[0002] In practical engineering applications, especially in the fields of automobiles, aviation, mechanical equipment, etc., the surface wear detection of curved structures is of great significance. In the automotive industry, the steering wheel is the main control interface between the driver and the vehicle, and the wear of its surface directly affects the driving experience and safety. Unlike flat structures, curved structures show different characteristics and laws during the wear process, which brings great challenges to wear detection.

[0003] Due to the different curvatures in different areas, the same wear feature will appear in different forms in the image, and this difference will lead to insufficient accuracy and robustness of wear detection. The change in curvature in different areas of the surface will cause the same wear feature in the image to appear in different forms or proportions at different locations. In areas with larger curvatures, the wear feature may be stretched or compressed, while in areas with smaller or flatter curvatures, the wear feature may appear in a more regular form. Due to the difference in curvature, the same wear feature in the image may have different scales and angles in different areas, which will affect the judgment criteria of the detection algorithm and cause misidentification or missed identification.

[0004] In summary, the prior art has a technical problem in that the accuracy of wear detection is limited due to different curvatures in different areas, resulting in different forms of the same wear feature in the image. Summary of the invention

[0005] The purpose of this application is to provide a surface wear detection system, method and electronic equipment for curved structures, so as to solve the technical problem in the prior art that the accuracy of wear detection is limited due to the different curvatures of different areas and the different shapes of the same wear features presented in the image.

[0006] In order to achieve the above objectives, the present application provides a surface wear detection system, method and electronic equipment for curved surface structures.

[0007] In a first aspect, the present application provides a surface wear detection system for a curved surface structure, wherein the surface wear detection system for a curved surface structure comprises: an image acquisition module, the image acquisition module is used to perform multi-angle image acquisition on a target curved surface structure, and obtain a set of curved surface images covering the complete surface of the target curved surface structure; a curvature determination module, the curvature determination module is used to determine a curvature information set of an image acquisition area in the curved surface image set; an image combination module, the image combination module is used to analyze the curvature information set and perform curvature consistent clustering on the curved surface image set to generate multiple groups of images to be detected, and the multiple groups of images to be detected have curvature information identifiers; a model construction module, the model construction module is used to The invention relates to constructing a wear feature recognition model, wherein the wear feature recognition model comprises a plurality of plane wear image conversion channels and a wear feature recognition channel, the plurality of plane wear image conversion channels are connected in parallel, and the output ends of the plurality of plane wear image conversion channels are connected to the input end of the wear feature recognition channel; a wear recognition module, wherein the wear recognition module is used to match a plurality of target plane wear image conversion channels corresponding to the plurality of groups of images to be detected in the plurality of plane wear image conversion channels of the wear feature recognition model with the curvature information as a constraint, convert the plurality of groups of images to be detected from curved surface to plane, and then input the converted images into the wear feature recognition channel for wear recognition, so as to generate a wear detection result.

[0008] Optionally, multiple wear image sample sets corresponding to the same wear feature under different curvatures are collected; similarity analysis of image features is performed on any two wear image sample sets among the multiple wear image sample sets to generate the similarity of any two wear image sample sets; if the similarity of any two wear image sample sets is greater than a preset similarity, the corresponding curvatures are merged to generate multiple curvature consistency judgment intervals; the curvature information set is clustered using the multiple curvature consistency judgment intervals as clustering criteria, and the surface image set is synchronously mapped according to the curvature clustering results to generate the multiple groups of images to be detected; the multiple groups of images to be detected are marked according to the multiple curvature consistency judgment intervals to generate the curvature information identifiers.

[0009] Optionally, several position curvatures of the target surface structure are determined, wherein the several position curvatures are the curvatures of several areas with consistent curvature on the target surface structure; several plane wear image conversion channels and one wear feature recognition channel are constructed for the several position curvatures; the several plane wear image conversion channels are connected in parallel and then connected with the wear feature recognition channel to generate the wear feature recognition model.

[0010] Optionally, using the several position curvatures as index elements, several historical curved surface wear image data sets corresponding to preset wear features are collected; for the several historical curved surface wear image data sets, several historical plane wear image data sets in a plane state are constructed; using the several historical plane wear image data sets and the several historical curved surface wear image data sets, several plane wear image conversion channels are constructed; using the several historical plane wear image data sets and the corresponding historical wear feature training to construct a wear feature recognition channel.

[0011] Optionally, a plurality of network structures are constructed using a generative adversarial network, wherein any network structure includes a generator and a discriminator; a first generator and a first discriminator corresponding to a first network structure among the plurality of network structures are extracted; a first historical surface wear image data set corresponding to a first position curvature is extracted from the plurality of historical surface wear image data sets, and a first historical plane wear image data set is extracted from the plurality of historical plane wear image data sets; the first generator and the first discriminator are alternately trained with the first historical surface wear image data set and the first historical plane wear image data set to generate a first plane wear image conversion channel, and the first plane wear image conversion channel is added to the plurality of plane wear image conversion channels.

[0012] Optionally, a light sensor is connected to the target curved surface structure to collect light information, wherein the light information includes light angles and light intensities at multiple positions; reflection correction is performed on the curved surface image set with reference to the light information; and wear identification is performed on the curved surface image set after reflection correction.

[0013] Optionally, taking the material type of the target curved surface structure as a constraint, surface reflection correction samples are collected, wherein the surface reflection correction samples include curvature samples, surface light samples, distortion image samples and corresponding reflection correction image samples; the curvature samples, the surface light samples and the distortion image samples are used as training inputs, and the reflection correction image samples are used as output supervision to train a reflection correction model; the curvature information set and the surface image set are extracted according to the corresponding relationship, and the light angle and light intensity at the corresponding position of the light information are extracted, and the extracted results are input into the reflection correction model for reflection correction.

[0014] In a second aspect, the present application also provides a surface wear detection method for a curved surface structure, wherein the surface wear detection method for a curved surface structure comprises: performing multi-angle image acquisition on a target curved surface structure to obtain a set of curved surface images covering the complete surface of the target curved surface structure; determining a set of curvature information of an image acquisition area in the set of curved surface images; analyzing the set of curvature information to perform curvature consistent clustering on the set of curved images to generate multiple groups of images to be detected, and the multiple groups of images to be detected have curvature information identifiers; constructing a wear feature recognition model, wherein the wear feature recognition model comprises a plurality of plane wear image conversion channels and a wear feature recognition channel, the plurality of plane wear image conversion channels are connected in parallel, and the output ends of the plurality of plane wear image conversion channels are connected to the input end of the wear feature recognition channel; using the curvature information identifier as a constraint, matching a plurality of target plane wear image conversion channels corresponding to the multiple groups of images to be detected in the plurality of plane wear image conversion channels of the wear feature recognition model, converting the multiple groups of images to be detected from curved surface to flat surface, and inputting the converted images into the wear feature recognition channel for wear recognition to generate a wear detection result.

[0015] In a third aspect, the present application further provides an electronic device, including:

[0016] at least one processor;

[0017] a memory communicatively coupled to the at least one processor;

[0018] Wherein, the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a surface wear detection system for a curved structure as described in any one of the first aspects above.

[0019] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0020] Through an image acquisition module, the image acquisition module is used to perform multi-angle image acquisition on the target curved surface structure to obtain a curved surface image set covering the complete surface of the target curved surface structure; a curvature determination module, the curvature determination module is used to determine the curvature information set of the image acquisition area in the curved surface image set; an image combination module, the image combination module is used to analyze the curvature information set to perform curvature consistent clustering on the curved surface image set to generate multiple groups of images to be detected, and the multiple groups of images to be detected have curvature information identifiers; a model construction module, the model construction module is used to construct a wear feature recognition model, wherein the wear feature recognition model includes The invention comprises a plurality of plane wear image conversion channels and a wear feature recognition channel, wherein the plurality of plane wear image conversion channels are connected in parallel, and the output ends of the plurality of plane wear image conversion channels are connected to the input end of the wear feature recognition channel; and a wear recognition module, wherein the wear recognition module is used to match the plurality of target plane wear image conversion channels corresponding to the plurality of groups of images to be detected in the plurality of plane wear image conversion channels of the wear feature recognition model with the curvature information identifier as a constraint, and then the plurality of groups of images to be detected are converted from curved surface to flat surface and then input into the wear feature recognition channel for wear recognition to generate wear detection results. In other words, by collecting images of the curved surface from multiple angles, grouping the images based on the curvature information, and providing a curvature identifier for each group of images, the plurality of plane wear image conversion channels in the wear feature recognition model can automatically select a suitable channel for image conversion according to the curvature information, and adaptively process images with different curvatures, thereby improving the accuracy of surface wear detection of curved surface structures.

[0021] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented according to the contents of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically cited below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0023] Figure 1 This is a schematic structural diagram of a surface wear detection system for a curved surface structure in the present application;

[0024] Figure 2 A schematic diagram of a flow chart of a surface wear detection method for a curved surface structure according to the present application;

[0025] Figure 3 This is a schematic diagram of the structure of an exemplary electronic device of the present application.

[0026] Explanation of the reference numerals: image acquisition module 11 , curvature determination module 12 , image combination module 13 , model building module 14 , wear identification module 15 , bus 300 , receiver 301 , processor 302 , transmitter 303 , memory 304 , bus interface 305 . DETAILED DESCRIPTION

[0027] The present application provides a surface wear detection system, method and electronic device for curved surface structures, which solves the technical problem in the prior art that the same wear feature has different forms in the image due to different curvatures in different areas, resulting in limited wear detection accuracy. By collecting images of the curved surface from multiple angles, grouping the images based on curvature information, and providing a curvature identifier for each group of images, multiple plane wear image conversion channels in the wear feature recognition model can automatically select a suitable channel for image conversion based on the curvature information, and adaptively process images with different curvatures, thereby improving the accuracy of surface wear detection of curved structures.

[0028] Below, the technical solutions in the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application. It should also be noted that, for the convenience of description, only the parts related to the present application are shown in the accompanying drawings, rather than all of them.

[0029] For example, please refer to the attached Figure 1 The present application provides a surface wear detection system for a curved surface structure, wherein the surface wear detection system for a curved surface structure specifically comprises the following steps:

[0030] The image acquisition module 11 is used to perform multi-angle image acquisition on the target curved surface structure to obtain a set of curved surface images covering the complete surface of the target curved surface structure.

[0031] Specifically, the target surface structure is imaged from multiple different viewing angles or angles to fully capture the detailed information of different areas on the surface and avoid image omissions or errors caused by a single viewing angle. The target surface structure refers to the specific surface structure that needs to be tested for wear, which is usually a complex curved or geometrically shaped surface, such as a car steering wheel. The target surface structure is imaged using a high-resolution camera or laser scanner, and the camera or scanner should cover all important areas of the target surface. For example, a steering wheel usually includes multiple viewing angles of the top, bottom, left, right, and different tilt angles.

[0032] The surface image set refers to the set of all images collected from multiple angles, including different areas and different features of the surface structure. Through multi-angle image acquisition, it can ensure that every part of the surface structure is effectively captured, avoiding image omissions caused by angle limitations.

[0033] The curvature determination module 12 is used to determine a curvature information set of an image acquisition area in the curved surface image set.

[0034] Specifically, each image in a curved image collection covers different areas, and the curvature characteristics of these areas are also different. For example, on a car steering wheel, the curvature of some areas may be larger (for example, the outside of the steering wheel), while the curvature of other areas is smaller (such as the bottom of the steering wheel). Usually, a three-dimensional model of the curved surface structure is constructed, and then the curvature of the image acquisition area is calculated.

[0035] Perform image preprocessing on each surface image in the surface image set, including denoising, enhancement, correction and other methods, to remove noise in the image and improve the image quality. By aligning and fusing the two-dimensional image data from multiple perspectives, the three-dimensional point cloud data of the target surface is obtained. Use corresponding algorithms (such as stereo vision algorithms, image stitching technology, laser scanners, etc.) to convert the two-dimensional image into three-dimensional data. Import the collected point cloud data into the three-dimensional modeling software, and use the software's surface fitting tools, such as NURBS (non-uniform rational B-splines) or Bezier surfaces, to fit the point cloud data and generate a smooth surface. Adjust the control points of the surface as needed to obtain a more accurate three-dimensional model.

[0036] For each three-dimensional point, the Gaussian curvature and the mean curvature are calculated. Gaussian curvature reflects the curvature of the surface at a point, and the calculation method is usually obtained through the second-order derivative; the mean curvature is the average of the principal curvatures and is often used to describe the overall curvature of the surface. For each local area on the three-dimensional model, the least squares fitting method is used to fit a local surface (such as a quadratic surface) to estimate the local curvature. Specifically, after fitting a small-scale surface in each area, the curvature of the area is calculated by the first-order and second-order derivatives of the surface. According to the point cloud data, the key points on the surface are extracted using methods such as the Harris corner detection algorithm, and then the curvature of the point is calculated. For smooth areas, the curvature is small; for areas with larger curvatures, the curvature value is larger.

[0037] Curvature is a quantity that describes the local curvature of a surface, and mainly includes normal curvature, principal curvature, and Gaussian curvature. Normal curvature is the curvature along a certain direction, which is calculated by the second-order derivative and the direction vector; the principal curvature is the extreme value of the normal curvature in the main direction, which is obtained by solving the characteristic value of the quadratic form; Gaussian curvature is the product of two principal curvatures; and mean curvature is the average of the two principal curvatures.

[0038] For each image area, a local curvature estimation method (such as Gaussian curvature, mean curvature, etc.) is used to calculate the curvature value of the area to obtain a curvature information set, including curvature information about each area of ​​the surface extracted from each image, describing the degree of curvature of the surface in each area of ​​the image. Usually, the curvature information set corresponds one-to-one to the surface image set, which is used to analyze the geometric characteristics of different areas.

[0039] The image combination module 13 is used to analyze the curvature information set to perform curvature consistent clustering on the surface image set to generate multiple groups of images to be detected, and the multiple groups of images to be detected have curvature information identifiers.

[0040] Specifically, the curvature information set is usually one-to-one corresponding to the surface image set. The surface image set is divided into multiple groups according to the curvature information set, and each group of image sets has similar curvature information. According to the wear image samples of the same surface wear characteristics under different curvatures, a curvature consistency judgment interval is determined, that is, two samples with similar image features are classified into one category. According to the curvature consistency judgment interval, the curvature information set is clustered according to the similarity of the image features to obtain the curvature clustering result. According to the curvature clustering result, the surface image set is synchronously mapped to obtain multiple groups of images to be detected. The image features (i.e., surface wear features) of each group of images to be detected are similar, but may be in different curvature regions. For example, the images in the top area and the bottom area may be similar in wear type (for example, both have large wear spots), but have different curvatures. Despite the different curvatures, these images can still be classified into the same group due to the similar wear features.

[0041] Add a curvature information mark to each image to be tested to indicate which curvature interval the image belongs to. The curvature information mark can be added by numerical mark, category mark, color coding, etc. By grouping the surface image collection according to the similarity of wear characteristics and curvature information, the wear types of different areas can be accurately identified. The wear characteristics of different curvature areas may be different in morphology and expression, and the impact of curvature changes on wear characteristics can be identified.

[0042] The model building module 14 is used to build a wear feature recognition model, wherein the wear feature recognition model includes a plurality of planar wear image conversion channels and a wear feature recognition channel, the plurality of planar wear image conversion channels are connected in parallel, and the output ends of the plurality of planar wear image conversion channels are connected to the input ends of the wear feature recognition channels.

[0043] Specifically, a wear feature recognition model is constructed, including several plane wear image conversion channels and one wear feature recognition channel. The plane wear image conversion channel is used to process curved surface images. Since the local curvature of curved surface images affects the display effect of wear features in the image, these curved surface images need to be converted into plane images so that the recognition process can be carried out in a standardized image space. Several plane wear image conversion channels are trained using a generative adversarial network. The generator network generates plane images based on curved surface images, while the discriminator is responsible for distinguishing between generated images and real images.

[0044] In order to process wear images at different positions and with different curvatures, the model requires multiple plane wear image conversion channels. Each channel is responsible for processing images at different positions. These channels can work in parallel and select the appropriate channel for processing based on the curvature information of the input image. By designing several plane wear image conversion channels, each channel specializes in processing different types of surface areas (different curvatures), and the outputs of these channels will enter the wear feature recognition channel for recognition, that is, the output ends of several plane wear image conversion channels are connected to the input ends of the wear feature recognition channel.

[0045] The wear feature recognition channel is used to extract wear features from plane wear images, and classify and locate them. Through multiple convolutional layers, pooling layers, and fully connected layers of the convolutional neural network, important information in the image is extracted layer by layer. When training the wear feature recognition channel, historical plane wear image data and its corresponding wear labels are used. Each plane image contains different wear features, such as scratches, indentations, etc., and each feature has a clear location and type label.

[0046] The output ends of several plane wear image conversion channels are connected to the input ends of the wear feature recognition channel to form a complete wear feature recognition model, which automatically processes wear images of different curvature areas and identifies wear features. Multiple plane wear image conversion channels are connected in parallel, and each conversion channel inputs the plane image it processes into the wear feature recognition channel. The wear feature recognition channel learns how to extract wear features from plane images through training, and classifies and locates them according to the image content. By building a wear feature recognition model, it is ensured that the wear features of each area can be accurately identified.

[0047] The wear recognition module 15 is used to use the curvature information as a constraint to match the multiple target plane wear image conversion channels corresponding to the multiple groups of images to be detected in the multiple plane wear image conversion channels of the wear feature recognition model, convert the multiple groups of images to be detected from curved surfaces to planes, and then input them into the wear feature recognition channel for wear recognition to generate wear detection results.

[0048] Specifically, the curvature of each group of images to be detected is different, so the required plane conversion channels are also different. The curvature information identifier provides a relevant curvature value for each area. According to the range of the curvature information identifier, the wear feature recognition model can select a plane wear image conversion channel that matches the curvature range. According to the curvature information identifiers of multiple groups of images to be detected, the corresponding multiple target plane wear image conversion channels are matched in several plane wear image conversion channels of the wear feature recognition model, and the multiple groups of images to be detected are converted into plane images. After the plane conversion, the wear features in the curved surface image can be displayed in a standard form on the plane image, which is convenient for subsequent feature recognition. On the curved surface image, the wear features may present different forms due to different local curvatures. The process of plane conversion is to convert these images affected by curvature into plane images, eliminate the influence of curvature on the presentation of wear features, and enable image features to be unified into a standard space for recognition.

[0049] The converted plane wear images of multiple groups of images to be detected are input into the wear feature recognition channel for wear recognition, and the wear features in the image are identified, classified and located. Finally, the model outputs the wear detection results, including the classification of wear types (such as scratches, fading, indentations, etc.) and the location, area, severity, etc. of these wear features. The wear detection results include a detailed description of each wear area, such as type, location, size, etc. Based on the wear detection results, a complete wear detection report is generated to inform the staff of the status of each wear area and provide a basis for maintenance. By selecting the appropriate conversion channel according to the curvature information, the curvature differences in different areas can be accurately processed, and the wear feature recognition deviation caused by the curvature difference can be reduced, thereby improving the accuracy of recognition, automatically adapting to different types of curved surface structures and different curvature areas, and improving the efficiency and consistency of wear detection.

[0050] Furthermore, the image combination module 13 in the surface wear detection system for a curved surface structure is also used for:

[0051] Collect multiple wear image sample sets corresponding to the same wear feature under different curvatures; perform similarity analysis on image features of any two wear image sample sets among the multiple wear image sample sets to generate the similarity of any two wear image sample sets; if the similarity of any two wear image sample sets is greater than a preset similarity, merge the corresponding curvatures to generate multiple curvature consistency judgment intervals; cluster the curvature information set using the multiple curvature consistency judgment intervals as clustering criteria, and synchronously map the surface image set according to the curvature clustering results to generate the multiple groups of images to be detected; mark the multiple groups of images to be detected according to the multiple curvature consistency judgment intervals to generate the curvature information identifiers.

[0052] Specifically, through multi-angle scanning and image acquisition, the target surface is fully covered, especially the wear sample images are collected in different curvature areas (such as flat areas and curved areas). The same wear features under different curvatures refer to the forms of the same type of wear features (such as surface abrasions, discoloration, peeling, etc.) in different areas on surfaces with different curvatures. Although these wear features may look slightly different in areas with different curvatures, they belong to the same type of wear phenomenon. Multiple wear image sample sets include wear-related image data sets collected in different surface areas (such as different curvature areas), and each wear image sample set contains the wear performance of a specific surface area at different angles and different conditions.

[0053] Perform similarity analysis on image features of any two wear image sample sets in multiple wear image sample sets, calculate similarity, and judge the similarity of two wear image sample sets. Extract key features from each group of wear images. The extracted image features usually exist in the form of multidimensional data, and each feature may contain different values. At this time, it is necessary to organize these features into a unified feature vector, use a similarity function to calculate the similarity between feature vectors, and obtain the similarity between any two wear image sample sets. Commonly used similarity measurement methods include Euclidean distance, cosine similarity, etc.

[0054] Before similarity analysis, it is usually necessary to preprocess the image, including denoising, grayscale, image normalization, etc. Since the wear image may be affected by environmental noise, it is necessary to perform noise removal on the image, such as Gaussian blur, mean filtering, etc. Gaussian blur is a common denoising method that smoothes the image by combining the value of each pixel in the image with the Gaussian weighted average of its neighborhood to remove noise. A Gaussian filter is a linear filter that usually consists of a Gaussian kernel matrix and uses a convolution operation to blur the image. Mean filtering is a simple low-pass filtering method that smoothes the image by replacing the value of each pixel in the image with the average value of all pixels in the neighborhood of the pixel. Grayscale refers to the conversion of a color image into a grayscale image to simplify calculations and focus on processing wear features. Each pixel of a color image usually consists of intensity values ​​of three channels: red, green, and blue (RGB). The grayscale process converts each RGB pixel value into a grayscale value according to a certain weighted average formula. The commonly used weighted average formula is: Gray value = 0.299×R+0.587×G+0.114×B, where R, G, and B are the intensity values ​​of the red, green, and blue channels of each pixel in the image. The three coefficients are set according to the different perceptions of the human eye to the three colors of red, green, and blue. Green is given a higher weight, followed by red, and blue has the smallest weight. Image normalization refers to the normalization of an image to eliminate differences in illumination or contrast between different samples. For each pixel value, it is usually mapped from the original [0,255] range to the [0,1] interval.

[0055] Features are extracted from the preprocessed image. Different types of features are used to describe different aspects of the image, such as edges, textures, shapes, and frequencies. Edge features describe the structural information in the image, such as the outline and boundary of an object. Edges usually represent areas in the image where the grayscale changes significantly, and are used to identify the shape and position of the object. The Canny edge detection algorithm is a classic edge detection method. It uses Gaussian filtering to denoise and reduce noise interference; uses the Sobel operator to calculate the gradient of the image and find the area with large grayscale changes. The Sobel operator includes filters in the horizontal and vertical directions. By calculating the gradients in these two directions, the edge information of each pixel in the image is obtained; the gradient is refined, the local maximum is retained, and the insignificant edges are removed; strong edges and weak edges are screened out by setting two thresholds. Strong edges are directly retained, and weak edges are retained based on their connection with strong edges. Texture features can reflect the structural pattern of the image surface and are often used to describe the surface characteristics of objects, such as roughness, smoothness, regularity, etc. Local binary patterns (LBP) are usually used to generate binary patterns by comparing the grayscale values ​​of each pixel in the image with its neighboring pixels, thereby extracting local texture features of the image. Select the neighborhood around each pixel and compare the pixels in the neighborhood. If the value of the neighborhood pixel is greater than the central pixel, it is marked as 1, otherwise it is 0. The binary number is converted to a decimal number to obtain the LBP value. Shape features are mainly used to describe the geometric form of objects in the image, such as the size, outline, angle, etc. of the object. The outline is extracted through edge detection algorithms (such as the Canny edge detection algorithm) to calculate the area, perimeter, and aspect ratio of the object outline.

[0056] The extracted image features usually exist in the form of multi-dimensional data, and each feature may contain different values. In this case, these features need to be organized into a unified feature vector to facilitate subsequent similarity calculations. The extracted features (such as texture, shape, edge, etc.) are arranged in a certain order into a feature vector.

[0057] After converting the image features into vector space, calculate the Euclidean distance between any two image samples. The smaller the Euclidean distance, the more similar the two images are. Alternatively, the similarity is measured by calculating the angle between the image feature vectors, which is often used to calculate the angular difference between two images. Calculate the similarity between each sample in a wear image sample set, and then use the average similarity or maximum similarity to represent the similarity between the sample set and other sample sets. For example, for two wear image sample sets, calculate the similarity between all samples in each sample set, and finally calculate the overall similarity of the two sample sets. The preset similarity is a predefined threshold used to determine whether two images are similar enough. If the similarity of the images is greater than the preset value, it is considered that the two images represent similar wear features, otherwise they are not considered. Combine the wear features of similar areas to avoid errors caused by differences in curvature in different areas. For example, assuming that the similarity of sample set A is 0.88 with that of sample set B, and is higher than the preset threshold of 0.85, then A and B can be merged to generate a judgment interval with consistent curvature.

[0058] Through similarity analysis and merging, areas with similar curvatures are merged into one interval, and the images in each curvature consistent interval have similar wear characteristics. Multiple curvature consistent judgment intervals include multiple intervals with similar curvatures. Using clustering algorithms such as K-means clustering or DBSCAN, the curvature information set is clustered according to the multiple curvature consistent judgment intervals, and the areas with similar curvature information are divided into the same category to form a curvature clustering result. According to the curvature clustering results, synchronous mapping is performed according to the mapping relationship between the curvature information set and the surface image set to obtain multiple groups of images to be detected. Synchronous mapping refers to the unified mapping and processing of images in different curvature areas according to the results of curvature clustering of the image set, ensuring that images in the same curvature area can be properly processed to avoid errors caused by curvature differences. The wear characteristics of each group of images to be detected are similar.

[0059] For example, wear image sample sets under different curvatures are collected, each sample set contains images with the same wear characteristics, and similarity analysis is performed on any two of the sample sets. The preset similarity is 0.85, and multiple curvature consistency judgment intervals are generated accordingly. Multiple wear image sample sets are shown in Table 1:

[0060] Table 1 Wear image sample set

[0061]

[0062] The similarity of the four wear image sample sets was calculated, and the final similarity between sample set 1 and sample set 2 was 0.45, the final similarity between sample set 1 and sample set 3 was 0.65, the final similarity between sample set 1 and sample set 4 was 0.86, the final similarity between sample set 2 and sample set 3 was 0.21, the final similarity between sample set 2 and sample set 4 was 0.73, and the final similarity between sample set 3 and sample set 4 was 0.15. According to the similarity analysis results, the curvatures corresponding to sequence number 1 and sequence number 4 are merged to form a new curvature consistency judgment interval.

[0063] Add a curvature information mark to each image to be tested to indicate which curvature interval the image belongs to, so as to help the subsequent wear detection model to perform different processing according to the characteristics of different curvature areas. The marking method can be a digital label, color coding or annotation, etc. The curvature information mark refers to the curvature information marked in each group of images to be tested, which is used to indicate the curvature interval or category to which the image belongs. By performing similarity analysis and clustering processing on different curvature areas, it can be ensured that the same wear features in different areas are processed consistently, thereby improving the accuracy and stability of wear detection.

[0064] Furthermore, the model building module 14 in the surface wear detection system for a curved surface structure is also used for:

[0065] Determine several position curvatures of the target surface structure, wherein the several position curvatures are the curvatures of several areas with consistent curvature on the target surface structure; construct several plane wear image conversion channels and one wear feature recognition channel for the several position curvatures; connect the several plane wear image conversion channels in parallel and then connect them with the wear feature recognition channel to generate the wear feature recognition model.

[0066] Specifically, for the target surface (such as a car steering wheel, the surface of a mechanical component, etc.), the three-dimensional point cloud data of the target surface is obtained through calculations based on the surface equation and using a laser scanner or structured light scanning. Curvature analysis is performed based on the point cloud data to determine the curvature at different positions. Position curvature refers to the curvature value measured at different positions, which usually reflects the degree of curvature of each area of ​​the target surface. Several position curvatures refer to the curvatures of several areas with consistent curvature on the target surface structure. In other words, it is equivalent to dividing the target surface structure into several position areas according to the curvature, where the curvature of each position area is consistent, and the curvatures of different position areas are inconsistent.

[0067] According to several position curvatures, multiple plane wear image conversion channels are designed. Each conversion channel converts the wear image of the corresponding curved surface area into a flat image by training a generative adversarial network. The generator of the conversion channel learns how to adjust the image according to the curvature of the curved surface area, and the discriminator is used to distinguish between real images and generated images. The generator and discriminator are trained by using a dataset containing curved surface wear images and pre-labeled flat surface wear images. The generator learns to accurately convert the curved surface image into a flat image through continuous optimization, and the discriminator feeds back its judgment result on the authenticity of the generated image.

[0068] After the plane wear image conversion channel training is completed, the wear feature recognition channel is trained using the historical plane wear image data and the corresponding wear features (such as scratches, fading, etc.). Based on the convolutional neural network (CNN) or other deep learning models, the plane wear images and the corresponding wear feature labels are used to train the wear feature recognition channel.

[0069] Multiple plane wear image conversion channels are connected in parallel, and multiple plane wear image conversion channels work in parallel to process wear images of different curvature areas respectively, and convert the curved surface images into plane images. The converted plane images will be input into the wear feature recognition channel for feature extraction and classification. Multiple plane wear image conversion channels are connected with the wear feature recognition channel to form a complete wear detection model. The curved surface image of each position is processed by the corresponding conversion channel and then passed to the recognition channel for wear detection. By using the adversarial generative network to train multiple plane wear image conversion channels, the curved surface images of different position curvatures can be automatically converted into plane images. The wear feature recognition channel is trained to automatically identify the features in the plane wear image. The wear image of each area is processed by a specific channel, thereby improving the accuracy and consistency of the detection results.

[0070] Furthermore, the model building module 14 in the surface wear detection system for a curved surface structure is also used for:

[0071] Taking the several position curvatures as index elements, several historical curved surface wear image data sets corresponding to preset wear features are collected; several historical plane wear image data sets in a plane state are constructed for the several historical curved surface wear image data sets; several plane wear image conversion channels are constructed with the several historical plane wear image data sets and the several historical curved surface wear image data sets; and a wear feature recognition channel is constructed with the several historical plane wear image data sets and the corresponding historical wear feature training.

[0072] Specifically, several position curvatures refer to the curvature values ​​at different positions on the target surface, and the curvatures at different positions reflect the curvature degree of the surface in the area. Preset wear features refer to specific wear types or forms pre-set in wear detection, including scratches, dents, fading, abrasions, etc. According to several position curvatures and preset wear features, several historical surface wear image data sets of the wear types corresponding to these positions are collected from the historical database. Each image data set includes image data of different surface wear at a position.

[0073] For several historical curved surface wear image data sets, several historical plane wear image data sets of the plane state at several corresponding positions are collected, and each image data includes a plane wear image at one position. Several historical curved surface wear image data sets and several historical plane wear image data sets are one-to-one corresponding, the former is the curved surface wear image data set at the position, and the latter is the plane wear image data set at the position. The historical plane wear image data is mainly used for the subsequent conversion of curved surface images into plane images. These historical plane wear image data are used as references for training the image conversion channel.

[0074] According to several historical plane wear image data sets and several historical curved surface wear image data sets, several plane wear image conversion channels are constructed. Each channel performs conversion processing on images in different curvature regions to ensure that images under different curvatures can be converted to a plane state in a consistent manner. By constructing multiple network structures based on generative adversarial networks, the generator and discriminator in each network structure are alternately trained until the generator generates high-quality plane wear images, so that the input historical curved surface wear image data can generate corresponding plane wear images. The generator in the generative adversarial network converts the curved surface image into a plane image, while the discriminator distinguishes between the real image and the generated image. Through adversarial training, the generator gradually learns how to generate realistic plane wear images. During the training process, the generator and the discriminator are updated alternately to improve the quality of the generated image and the accuracy of the discriminator. Through multiple iterations, the generator is finally able to generate highly realistic plane wear images as part of the conversion channel.

[0075] The wear feature recognition channel is trained using several historical plane wear image data sets and corresponding historical wear features (such as scratches, fading, etc.). The training is usually based on convolutional neural networks (CNN) or other deep learning models. The goal is to extract features in plane images and identify wear types. Each historical plane wear image data corresponds to a label, which annotates the wear features and their corresponding attributes (such as position, size, type, etc.) that appear in the image. Deep learning models such as convolutional neural networks (CNNs) are used as wear feature recognition channels, which contain multiple convolutional layers, pooling layers, and fully connected layers to extract features at different levels from plane wear images. The plane wear image is used as input, and low-order features in the image, such as edges and textures, are extracted through convolution operations. Pooling operations (such as maximum pooling) are used to reduce the size and complexity of the image while retaining important features. The extracted features are synthesized to generate the final wear feature classification results.

[0076] The model is trained using several historical surface wear image datasets and corresponding wear feature labels. The label of each image tells the model which wear features are in the image, as well as the location, size, and other information of these features. By using these labels, the model can learn how to identify these features from the image. During the training process, the network continuously adjusts parameters through the back-propagation algorithm so that the model can accurately identify different types of wear features. By iteratively updating the network parameters, the accuracy of the model gradually improves. The performance of the model is evaluated by using an independent validation set. The validation set contains data that has not been used in the training process to ensure that the model does not overfit. Based on the evaluation results, the model hyperparameters (such as learning rate, number of layers, filter size, etc.) are adjusted, or more advanced models (such as deep residual networks) are used to further improve performance. After training and evaluation, the wear feature recognition channel with higher accuracy is applied to the actual wear detection system to automatically process new surface wear images and identify various wear features in the images.

[0077] The wear feature recognition channel is based on a trained generator, which determines whether an image contains specific wear features by identifying features in a plane wear image. By training the GAN generator, the surface wear image can be accurately converted into a plane wear image. By training the wear feature recognition channel, the wear features in the plane wear image can be accurately identified, and high-quality plane wear images can be automatically generated. The wear feature recognition channel can also be used for efficient identification.

[0078] Furthermore, the model building module 14 in the surface wear detection system for a curved surface structure is also used for:

[0079] Constructing several network structures with a generative adversarial network, wherein any network structure includes a generator and a discriminator; extracting a first generator and a first discriminator corresponding to a first network structure among the several network structures; extracting a first historical surface wear image data set corresponding to a first position curvature from the several historical surface wear image data sets, and extracting a first historical plane wear image data set from the several historical plane wear image data sets; alternately training the first generator and the first discriminator with the first historical surface wear image data set and the first historical plane wear image data set to generate a first plane wear image conversion channel, and adding the first plane wear image conversion channel to the several plane wear image conversion channels.

[0080] Specifically, according to the generative adversarial network, several network structures are constructed, each of which includes a generator and a discriminator. Each network structure will focus on different types of wear data. The generator is responsible for generating realistic surface wear images, while the discriminator is responsible for distinguishing the generated images from real surface wear images. The generative adversarial network (GAN) is a deep learning model composed of two neural networks (generator and discriminator). The generator is responsible for generating data, and the discriminator is responsible for judging the authenticity of the data. The two networks are mutually optimized through adversarial training, and finally the generator can generate samples that approximate real data.

[0081] The generator is a network responsible for generating fake data. Its goal is to generate new samples by learning the distribution of real data so that the discriminator cannot distinguish the difference between the generated data and the real data. The discriminator is a network used to judge whether the data is real. Its task is to distinguish between real data and fake data generated by the generator by learning, and give a probability value (usually 0 or 1). Through adversarial training, the discriminator continuously optimizes its judgment ability. The generator of each network structure will generate a flat wear image from a curved wear image, and the discriminator will judge whether the generated image is real. Through adversarial training, the generator will gradually learn how to generate more realistic flat wear images.

[0082] A network structure is selected from the constructed several network structures as the first network structure, and a first generator and a first discriminator are extracted to generate and identify a first set of historical wear image data, especially for wear identification of a specific curved surface area. A first historical curved surface wear image data set related to the first generator and the discriminator is extracted from several historical curved surface wear image data sets, focusing on a specific area (such as the top of the steering wheel) and including image data of the area under different wear conditions. At the same time, a first historical plane wear image data set corresponding to the first historical curved surface wear image data set is extracted from several historical plane wear image data sets.

[0083] The first generator and the first discriminator are alternately trained using the first historical surface wear image dataset and the first historical plane wear image dataset. The plane wear image is generated by the first generator, and then the first discriminator distinguishes the generated plane wear image from the real image in the first historical plane wear image dataset. According to the feedback from the discriminator, the generator continuously adjusts its generation process to produce more realistic plane wear images. For each training batch, the parameters of the discriminator are first updated, and then the parameters of the generator are updated. A loss function (such as binary cross entropy loss) is used to measure the performance of the generator and the discriminator. The goal of the generator is to generate realistic plane images, while the goal of the discriminator is to identify the generated fake images. Through this adversarial training, the two networks are continuously optimized, and the generator will gradually learn how to generate more realistic plane wear images.

[0084] After the training is completed, the first generator will be able to generate a plane wear image corresponding to the input surface image. The generated plane image together with the historical plane wear image data constitutes a new plane wear image conversion channel. The generated first plane wear image conversion channel is added to several plane wear image conversion channels, and they act together on the surface wear images of different regions to generate corresponding plane images and provide input data for the wear feature recognition model. Through the adversarial training of the generative adversarial network, the generator can generate realistic plane wear images. The continuous feedback of the discriminator promotes the continuous optimization of the generator, and the plane image generated is close to the real image. Through the training of the generative adversarial network, the wear image generation and recognition process can be automatically executed. Wear images of different regions are generated by the corresponding generators, which improves the adaptability and accuracy of the model.

[0085] Furthermore, the surface wear detection system for a curved surface structure further includes a reflection correction module, and the reflection correction module is used to:

[0086] A light sensor is connected to the target curved surface structure to collect light information, wherein the light information includes light angles and light intensities at multiple positions; reflection correction is performed on the curved surface image set with reference to the light information; and wear identification is performed on the curved surface image set after reflection correction.

[0087] Specifically, light sensors are connected to the target curved surface structure to collect light information. Light sensors are arranged at multiple locations around the curved surface to capture the angle and intensity of reflected light in real time. A light sensor is a device used to detect light intensity and light angle. It provides data on the direction and intensity of light propagation by sensing changes in light at different locations. In the wear detection of curved surface structures, light sensors are mainly used to capture light information reflected from the curved surface to help analyze how light is affected by the surface morphology and wear. Light information refers to the data collected by the light sensor about the light angle and light intensity of the target curved surface structure. The light angle refers to the directional information of light when it is reflected on the surface, while the light intensity indicates the intensity of light reflected back from different areas.

[0088] According to the corresponding relationship between the curvature information set and the surface image set, the distorted image and the curvature information of the image are extracted, and the light angle and light intensity of the corresponding position are extracted from the light information. The distorted image, curvature information, light angle and light intensity are input into the trained reflection correction model to correct the image, remove the reflection error caused by different lighting and surface shapes, and accurately display the wear characteristics of the surface. According to the reflection model, each image in the surface image set is reflected and corrected. By analyzing the lighting conditions and reflection angle, the reflection part in the image is corrected to make the image closer to the actual surface state. The corrected image should remove the distortion caused by lighting and reflection angle, making the wear characteristics in the image clearer and more visible.

[0089] After completing the reflection correction, the image of the target surface is subjected to wear recognition to accurately identify the area and degree of surface wear, that is, it is input into the wear feature recognition model for wear recognition. By using the light information collected by the light sensor for reflection correction, errors caused by factors such as changes in lighting angle and reflection can be eliminated, thereby obtaining a more accurate image and improving the accuracy of wear detection.

[0090] Furthermore, the reflection correction module in the surface wear detection system for a curved surface structure is also used for:

[0091] Taking the material type of the target curved surface structure as a constraint, surface reflection correction samples are collected, wherein the surface reflection correction samples include curvature samples, surface light samples, distortion image samples and corresponding reflection correction image samples; taking the curvature samples, the surface light samples and the distortion image samples as training inputs, and taking the reflection correction image samples as output supervision, a reflection correction model is trained; extracting the curvature information set and the surface image set according to the corresponding relationship, and at the same time extracting the light angle and light intensity at the corresponding position in the light information, and inputting the extracted results into the reflection correction model for reflection correction.

[0092] Specifically, the material type of the target surface structure is the type of substance or synthetic material that constitutes the surface. For example, a car steering wheel may be made of a variety of materials such as plastic, leather, metal, wood, or composite materials. Different materials have different light reflection properties, so their performance during wear detection will be different. With the material type of the target surface structure as a constraint, surface reflection correction samples are collected, including curvature samples, surface light samples, distortion image samples, and corresponding reflection correction image samples.

[0093] Curvature samples refer to the curvature data of the target surface structure at different positions (i.e., the degree of curvature of the surface at certain points). Curvature is a quantity that describes the degree of curvature of the surface and affects how light is reflected on the surface. For different materials and surface shapes, the curvature value is different; surface light samples reflect the angle, intensity and other information of light reflection on each part of the surface, such as the angle of incidence, angle of reflection and light intensity; distorted image samples refer to image deformation caused by surface shape, light angle changes or material properties, usually manifested as illumination changes, reflection errors, geometric distortions, etc. in the image; reflection corrected image samples refer to image distortions caused by illumination changes and surface reflections after reflection correction, thereby more accurately reflecting the actual surface state.

[0094] The model is trained using surface reflection correction samples. The curvature samples, surface light samples, and distorted image samples are used as input data, and the reflection correction image samples are used as output supervision signals. The model is trained using a supervised learning method so that the model can learn how to correct the reflection distortion in the image based on the input information.

[0095] The surface reflection correction samples are preprocessed and divided into a training set and a validation set, usually 80% training set and 20% validation set. Using a convolutional neural network, features are extracted through multiple convolutional layers, and then the output image is generated through a fully connected layer. The training is performed based on the curvature, light, and distortion images in the input training set to learn how to perform reflection correction. The model is trained using the collected samples (curvature, light information, distorted images, and their corresponding corrected images). The parameters of the model are gradually optimized by minimizing the difference between the model output and the actual corrected image (such as using the mean square error loss function). The reflection correction model is trained using the input training data (curvature samples, light samples, distorted images) and supervision data (reflection correction image samples) through the back propagation algorithm.

[0096] Each set of data (curvature, light, and distorted image) in the validation set is used as the input of the model. The reflection correction image output by the reflection correction model is compared with the actual reflection correction image sample in the data set. The error between the two is calculated by the mean square error, and the model parameters are adjusted to gradually reduce the error. This process requires multiple rounds of training so that the model can learn the optimal reflection correction strategy.

[0097] The curvature information set and the surface image set are extracted according to their corresponding relationship, and the light information of the corresponding position is collected at the same time, including the light angle and light intensity. The curvature information, light information and distorted surface image information are input into the reflection correction model, and the reflection correction model outputs the corrected image, removing the reflection error caused by different lighting and surface shapes. The output image should reflect the surface wear characteristics more accurately than the original image, removing the distortion caused by reflection, lighting and other factors. Through reflection correction, the influence of uneven lighting and surface distortion is removed, making the final image more realistic and able to more accurately identify and locate the wear area.

[0098] In summary, the surface wear detection system for curved surface structure provided by the present application has the following technical effects:

[0099] Through an image acquisition module, the image acquisition module is used to perform multi-angle image acquisition on the target curved surface structure to obtain a curved surface image set covering the complete surface of the target curved surface structure; a curvature determination module, the curvature determination module is used to determine the curvature information set of the image acquisition area in the curved surface image set; an image combination module, the image combination module is used to analyze the curvature information set to perform curvature consistent clustering on the curved surface image set to generate multiple groups of images to be detected, and the multiple groups of images to be detected have curvature information identifiers; a model construction module, the model construction module is used to construct a wear feature recognition model, wherein the wear feature recognition model includes The invention comprises a plurality of plane wear image conversion channels and a wear feature recognition channel, wherein the plurality of plane wear image conversion channels are connected in parallel, and the output ends of the plurality of plane wear image conversion channels are connected to the input end of the wear feature recognition channel; and a wear recognition module, wherein the wear recognition module is used to match the plurality of target plane wear image conversion channels corresponding to the plurality of groups of images to be detected in the plurality of plane wear image conversion channels of the wear feature recognition model with the curvature information identifier as a constraint, and then the plurality of groups of images to be detected are converted from curved surface to flat surface and then input into the wear feature recognition channel for wear recognition to generate wear detection results. In other words, by collecting images of the curved surface from multiple angles, grouping the images based on the curvature information, and providing a curvature identifier for each group of images, the plurality of plane wear image conversion channels in the wear feature recognition model can automatically select a suitable channel for image conversion according to the curvature information, and adaptively process images with different curvatures, thereby improving the accuracy of surface wear detection of curved surface structures.

[0100] Embodiment 2: Based on the same inventive concept as the surface wear detection system for curved surface structure in the aforementioned embodiment 1, the present application also provides a surface wear detection method for curved surface structure, please refer to the attached Figure 2, the surface wear detection method for a curved surface structure comprises:

[0101] Perform multi-angle image acquisition on the target curved surface structure to obtain a set of curved surface images covering the complete surface of the target curved surface structure; determine the curvature information set of the image acquisition area in the curved surface image set; analyze the curvature information set to perform curvature consistent clustering on the curved surface image set to generate multiple groups of images to be detected, and the multiple groups of images to be detected have curvature information identification; construct a wear feature recognition model, wherein the wear feature recognition model includes a plurality of plane wear image conversion channels and a wear feature recognition channel, the plurality of plane wear image conversion channels are connected in parallel, and the output ends of the plurality of plane wear image conversion channels are connected to the input end of the wear feature recognition channel; using the curvature information identification as a constraint, match the plurality of target plane wear image conversion channels corresponding to the plurality of groups of images to be detected in the plurality of plane wear image conversion channels of the wear feature recognition model, convert the plurality of groups of images to be detected from curved surface to flat surface, and then input them into the wear feature recognition channel for wear recognition to generate wear detection results.

[0102] Further, the analyzing the curvature information set to perform curvature consistent clustering on the curved surface image set to generate multiple groups of images to be detected, and the multiple groups of images to be detected have curvature information identifiers, including:

[0103] Collect multiple wear image sample sets corresponding to the same wear feature under different curvatures; perform similarity analysis on image features of any two wear image sample sets among the multiple wear image sample sets to generate the similarity of any two wear image sample sets; if the similarity of any two wear image sample sets is greater than a preset similarity, merge the corresponding curvatures to generate multiple curvature consistency judgment intervals; cluster the curvature information set using the multiple curvature consistency judgment intervals as clustering criteria, and synchronously map the surface image set according to the curvature clustering results to generate the multiple groups of images to be detected; mark the multiple groups of images to be detected according to the multiple curvature consistency judgment intervals to generate the curvature information identifiers.

[0104] Furthermore, the construction of the wear feature recognition model includes:

[0105] Determine several position curvatures of the target surface structure, wherein the several position curvatures are the curvatures of several areas with consistent curvature on the target surface structure; construct several plane wear image conversion channels and one wear feature recognition channel for the several position curvatures; connect the several plane wear image conversion channels in parallel and then connect them with the wear feature recognition channel to generate the wear feature recognition model.

[0106] Furthermore, the construction of a plurality of plane wear image conversion channels and a wear feature recognition channel for the plurality of position curvatures includes:

[0107] Taking the several position curvatures as index elements, several historical curved surface wear image data sets corresponding to preset wear features are collected; several historical plane wear image data sets in a plane state are constructed for the several historical curved surface wear image data sets; several plane wear image conversion channels are constructed with the several historical plane wear image data sets and the several historical curved surface wear image data sets; and a wear feature recognition channel is constructed with the several historical plane wear image data sets and the corresponding historical wear feature training.

[0108] Further, the constructing of a plurality of plane wear image conversion channels using the historical plane wear image data and the plurality of historical curved surface wear image data sets includes:

[0109] Constructing several network structures with a generative adversarial network, wherein any network structure includes a generator and a discriminator; extracting a first generator and a first discriminator corresponding to a first network structure among the several network structures; extracting a first historical surface wear image data set corresponding to a first position curvature from the several historical surface wear image data sets, and extracting a first historical plane wear image data set from the several historical plane wear image data sets; alternately training the first generator and the first discriminator with the first historical surface wear image data set and the first historical plane wear image data set to generate a first plane wear image conversion channel, and adding the first plane wear image conversion channel to the several plane wear image conversion channels.

[0110] Furthermore, the surface wear detection for curved surface structures also includes:

[0111] A light sensor is connected to the target curved surface structure to collect light information, wherein the light information includes light angles and light intensities at multiple positions; reflection correction is performed on the curved surface image set with reference to the light information; and wear identification is performed on the curved surface image set after reflection correction.

[0112] Furthermore, the surface wear detection for curved surface structures also includes:

[0113] Taking the material type of the target curved surface structure as a constraint, surface reflection correction samples are collected, wherein the surface reflection correction samples include curvature samples, surface light samples, distortion image samples and corresponding reflection correction image samples; taking the curvature samples, the surface light samples and the distortion image samples as training inputs, and taking the reflection correction image samples as output supervision, a reflection correction model is trained; extracting the curvature information set and the surface image set according to the corresponding relationship, and at the same time extracting the light angle and light intensity at the corresponding position in the light information, and inputting the extracted results into the reflection correction model for reflection correction.

[0114] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. Figure 1 The surface wear detection system for curved surface structure and the specific examples in the first embodiment are also applicable to the surface wear detection method for curved surface structure in the present embodiment. Through the above detailed description of the surface wear detection system for curved surface structure, those skilled in the art can clearly know the surface wear detection method for curved surface structure in the present embodiment, so for the sake of brevity of the specification, it will not be described in detail here. As for the method disclosed in the embodiment, since it corresponds to the system disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the system part.

[0115] Embodiment three, based on the inventive concept of a surface wear detection system for a curved structure in the aforementioned embodiment one, the present application also provides an electronic device, comprising: at least one processor; a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute any one of the surface wear detection systems for a curved structure described in the aforementioned embodiment one.

[0116] Attached Figure 3 This is a schematic diagram of the structure of an exemplary electronic device of this application. Figure 3In the embodiment, the bus architecture is represented by bus 300, which may include any number of interconnected buses and bridges, and bus 300 connects various circuits including one or more processors represented by processor 302 and memory represented by memory 304. Bus 300 may also connect various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. Bus interface 305 provides an interface between bus 300 and receiver 301 and transmitter 303. Receiver 301 and transmitter 303 may be the same element, namely a transceiver, providing a unit for communicating with various other devices over a transmission medium. Processor 302 is responsible for managing bus 300 and general processing, while memory 304 may be used to store data used by processor 302 when performing operations.

[0117] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0118] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is also intended to include these modifications and variations.

Claims

1. A surface wear detection system for a curved surface structure, characterized in that: include: An image acquisition module, wherein the image acquisition module is used to perform multi-angle image acquisition on the target curved surface structure to obtain a set of curved surface images covering the complete surface of the target curved surface structure; A curvature determination module, the curvature determination module is used to determine a curvature information set of an image acquisition area in the curved surface image set; An image combination module, the image combination module is used to analyze the curvature information set to perform curvature consistent clustering on the surface image set to generate multiple groups of images to be detected, and the multiple groups of images to be detected have curvature information identifiers; A model building module, wherein the model building module is used to build a wear feature recognition model, wherein the wear feature recognition model includes a plurality of plane wear image conversion channels and a wear feature recognition channel, wherein the plurality of plane wear image conversion channels are connected in parallel, and output ends of the plurality of plane wear image conversion channels are connected to input ends of the wear feature recognition channel; A wear recognition module is used to use the curvature information as a constraint to match multiple target plane wear image conversion channels corresponding to the multiple groups of images to be detected in several plane wear image conversion channels of the wear feature recognition model, convert the multiple groups of images to be detected from curved surfaces to planes, and then input them into the wear feature recognition channel for wear recognition to generate wear detection results.

2. A surface wear detection system for a curved surface structure as claimed in claim 1, characterized in that: The model building module is specifically used for: Determine a plurality of position curvatures of the target curved surface structure, wherein the plurality of position curvatures are curvatures of a plurality of regions with consistent curvatures on the target curved surface structure; Constructing a plurality of plane wear image conversion channels and a wear feature recognition channel according to the plurality of position curvatures; The plurality of plane wear image conversion channels are connected in parallel and then connected with the wear feature recognition channel to generate the wear feature recognition model.

3. A surface wear detection system for a curved surface structure as claimed in claim 2, characterized in that: The model building module is specifically used for: Using the plurality of position curvatures as index elements, collecting a plurality of historical curved surface wear image data sets corresponding to preset wear features; Constructing a plurality of historical plane wear image data sets in a plane state according to the plurality of historical curved surface wear image data sets; Constructing a plurality of plane wear image conversion channels using the plurality of historical plane wear image data sets and the plurality of historical curved surface wear image data sets; A wear feature recognition channel is constructed by training the plurality of historical plane wear image data sets and corresponding historical wear features.

4. A surface wear detection system for a curved surface structure as claimed in claim 3, characterized in that: The model building module is specifically used for: Constructing several network structures using generative adversarial networks, where any network structure includes a generator and a discriminator; Extracting a first generator and a first discriminator corresponding to a first network structure from among the plurality of network structures; Extracting a first historical curved surface wear image data set corresponding to a first position curvature from the plurality of historical curved surface wear image data sets, and extracting a first historical plane wear image data set from the plurality of historical plane wear image data sets; The first generator and the first discriminator are alternately trained with the first historical curved surface wear image data set and the first historical flat surface wear image data set to generate a first flat surface wear image conversion channel, and the first flat surface wear image conversion channels are added.

5. A surface wear detection system for a curved surface structure as claimed in claim 1, characterized in that: The image combination module is specifically used for: Collect multiple wear image sample sets corresponding to the same wear features under different curvatures; Performing similarity analysis of image features on any two wear image sample sets among the multiple wear image sample sets to generate similarity between the any two wear image sample sets; If the similarity between any two wear image sample sets is greater than a preset similarity, the corresponding curvatures are merged to generate multiple curvature consistency judgment intervals; The curvature information set is clustered using the multiple curvature consistency judgment intervals as clustering criteria, and the surface image set is synchronously mapped according to the curvature clustering result to generate the multiple groups of images to be detected; The plurality of groups of images to be detected are marked according to the plurality of curvature consistency determination intervals to generate the curvature information identifier.

6. A surface wear detection system for a curved surface structure as claimed in claim 1, characterized in that: The surface wear detection system for a curved surface structure further includes a reflection correction module, which is used to: Connecting a light sensor on the target curved surface structure to collect light information, wherein the light information includes light angles and light intensities at multiple positions; Taking the light information as a reference, performing reflection correction on the curved surface image set; Wear identification is performed using a set of reflection-corrected surface images.

7. A surface wear detection system for a curved surface structure as claimed in claim 6, characterized in that: The reflection correction module is specifically used for: Taking the material type of the target curved surface structure as a constraint, collecting curved surface reflection correction samples, wherein the curved surface reflection correction samples include curvature samples, curved surface light samples, distortion image samples and corresponding reflection correction image samples; Using the curvature samples, the curved surface light samples and the distorted image samples as training inputs and the reflection correction image samples as output supervision, training a reflection correction model; The curvature information set and the curved surface image set are extracted according to the corresponding relationship, and the light angle and light intensity of the corresponding position are extracted at the same time, and the extraction results are input into the reflection correction model for reflection correction.

8. A surface wear detection method for a curved surface structure, characterized in that: The method for detecting surface wear of a curved surface structure is performed by a surface wear detection system for a curved surface structure according to any one of claims 1 to 7, wherein the surface wear detection method for a curved surface structure comprises: Capturing images of the target curved surface structure from multiple angles to obtain a set of curved surface images covering the entire surface of the target curved surface structure; Determine a curvature information set of an image acquisition area in the curved surface image set; Analyzing the curvature information set to perform curvature consistent clustering on the curved surface image set to generate multiple groups of images to be detected, wherein the multiple groups of images to be detected have curvature information identifiers; Constructing a wear feature recognition model, wherein the wear feature recognition model includes a plurality of plane wear image conversion channels and a wear feature recognition channel, the plurality of plane wear image conversion channels are connected in parallel, and the output ends of the plurality of plane wear image conversion channels are connected to the input end of the wear feature recognition channel; Taking the curvature information identifier as a constraint, multiple target plane wear image conversion channels corresponding to the multiple groups of images to be detected are matched in the multiple plane wear image conversion channels of the wear feature recognition model, and the multiple groups of images to be detected are converted from curved surfaces to planes and then input into the wear feature recognition channel for wear recognition to generate wear detection results.

9. An electronic device, comprising: at least one processor; a memory communicatively coupled to the at least one processor; Wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a surface wear detection system for a curved structure as described in any one of claims 1 to 7.

Images