Cloth inspection equipment defect inspection method and system based on YOLO algorithm

Abstract

The invention discloses a cloth inspection equipment defect inspection method and system based on a YOLO algorithm, and the method comprises the steps: continuously scanning inspected cloth through a high-resolution line scanning camera, so as to obtain a set of cloth images; the method comprises the following steps: firstly, acquiring a set of cloth images, splicing the set of cloth images by adopting an image recognition and analysis technology based on deep learning to form a complete continuous cloth image, then, capturing the shape, texture and color features of the spliced continuous cloth image by adopting a YOLO algorithm, and respectively carrying out feature space adaptive enhancement processing on the features to form a continuous cloth image set; therefore, whether the detected cloth has flaws or not is automatically judged according to multi-scale cascade coding among the cloth shape enhancement coding feature, the cloth texture enhancement coding feature and the cloth color enhancement coding feature after enhancement processing. Therefore, rapid detection can be realized, and meanwhile, the recognition capability on tiny flaws is improved, so that a more intelligent flaw detection scheme is provided.

Description

Technical Field

[0001] The present application relates to the field of intelligent inspection, and more specifically, to a defect inspection method and system for fabric inspection equipment based on the YOLO algorithm. Background Art

[0002] In the textile industry, fabric inspection is a key step to ensure product quality. The quality of fabric not only affects the aesthetic appearance of the final product, but also affects the durability and functionality of the product. Therefore, the fabric inspection process is of great significance to maintaining brand reputation and customer satisfaction.

[0003] With the development of technology, automated fabric inspection equipment has gradually become mainstream. These devices usually use image processing technology to identify defects on fabrics. However, existing automated fabric inspection equipment faces challenges in efficiency and accuracy when processing high-resolution images. Specifically, at present, most automated fabric inspection equipment uses fixed cameras to capture images, and then uses traditional image processing algorithms to detect defects. Traditional image processing algorithms may have difficulty accurately distinguishing normal features from defects, resulting in a high false alarm rate, and these methods may encounter performance bottlenecks when processing large amounts of data, resulting in slow detection speeds and prone to false alarms. In addition, most automated fabric inspection equipment uses fixed cameras to capture images. This setting may not be able to fully cover all areas of the fabric and is prone to missing potential defects.

[0004] Therefore, an optimized defect inspection solution for fabric inspection equipment is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a defect inspection method and system for a cloth inspection device based on the YOLO algorithm, which continuously scans the inspected cloth using a high-resolution line scan camera to obtain a set of cloth images, and uses image recognition and analysis technology based on deep learning to splice the set of cloth images to form a complete continuous cloth image, then uses the YOLO algorithm to capture the shape, texture and color features of the spliced ​​continuous cloth image, and performs feature space adaptive enhancement processing on these features respectively, so as to automatically judge whether the inspected cloth has defects according to the multi-scale cascade coding between the enhanced cloth shape enhancement coding features, cloth texture enhancement coding features and cloth color enhancement coding features. In this way, using a high-resolution line scan camera to continuously scan the cloth can ensure that all areas of the cloth are covered to avoid missing potential defects. In addition, the YOLO algorithm and deep learning technology can efficiently process high-resolution images, overcome the performance bottleneck of traditional methods when processing large amounts of data, achieve rapid detection, and improve the ability to recognize subtle defects, thereby providing a more intelligent defect inspection solution.

[0006] According to one aspect of the present application, a defect inspection method for a cloth inspection device based on a YOLO algorithm is provided, comprising: continuously scanning the inspected cloth using a high-resolution line scan camera to obtain a set of cloth images; Performing image stitching on the set of cloth images to obtain a continuous cloth image; Capturing multiple features of the continuous cloth images to obtain a cloth shape coding feature map, a cloth texture coding feature map, and a cloth color coding feature map; The cloth shape coding feature map, the cloth texture coding feature map and the cloth color coding feature map are subjected to cloth feature space adaptive enhancement to obtain a cloth shape enhanced coding feature map, a cloth texture enhanced coding feature map and a cloth color enhanced coding feature map, including: performing feature dispersion on the cloth shape coding feature map to obtain a set of cloth shape pixel granularity feature vectors; performing pixel granularity semantic association and spatial attenuation modulation on the set of cloth shape pixel granularity feature vectors to obtain a cloth shape semantic association space soft constraint score topology matrix; performing graph structure space enhancement on the cloth shape semantic association space soft constraint score topology matrix and the set of cloth shape pixel granularity features to obtain the cloth shape enhanced coding feature map; Based on the cloth multi-scale cascade coding features between the cloth shape enhancement coding feature map, the cloth texture enhancement coding feature map and the cloth color enhancement coding feature map, an identification result is obtained, and the identification result is used to indicate whether the inspected cloth has defects.

[0007] According to another aspect of the present application, a defect inspection system for cloth inspection equipment based on the YOLO algorithm is provided, which includes: a cloth image scanning module, which is used to continuously scan the inspected cloth using a high-resolution line scan camera to obtain a set of cloth images; A cloth image stitching module, used for stitching the set of cloth images to obtain a continuous cloth image; A cloth image multi-feature capture module, used for performing multi-feature capture on the continuous cloth image to obtain a cloth shape coding feature map, a cloth texture coding feature map and a cloth color coding feature map; A cloth feature space enhancement module, used for performing cloth feature space adaptive enhancement on the cloth shape coding feature map, the cloth texture coding feature map and the cloth color coding feature map to obtain a cloth shape enhanced coding feature map, a cloth texture enhanced coding feature map and a cloth color enhanced coding feature map, wherein the cloth feature space enhancement module comprises: a cloth feature dispersion unit, used for performing feature dispersion on the cloth shape coding feature map to obtain a set of cloth shape pixel granularity feature vectors; a cloth feature semantic association modulation unit, used for performing pixel granularity semantic association and spatial attenuation modulation on the set of cloth shape pixel granularity feature vectors to obtain a cloth shape semantic association space soft constraint score topology matrix; a cloth feature graph structure enhancement unit, used for performing graph structure space enhancement on the cloth shape semantic association space soft constraint score topology matrix and the set of cloth shape pixel granularity features to obtain the cloth shape enhanced coding feature map; The recognition result generation module is used to obtain a recognition result based on the cloth multi-scale cascade coding features between the cloth shape enhancement coding feature map, the cloth texture enhancement coding feature map and the cloth color enhancement coding feature map, and the recognition result is used to indicate whether the inspected cloth has defects.

[0008] Compared with the prior art, the present application provides a defect inspection method and system for a cloth inspection device based on the YOLO algorithm, which uses a high-resolution line scan camera to continuously scan the inspected cloth to obtain a set of cloth images, and uses image recognition and analysis technology based on deep learning to splice the set of cloth images to form a complete continuous cloth image, then uses the YOLO algorithm to capture the shape, texture and color features of the spliced ​​continuous cloth image, and performs feature space adaptive enhancement processing on these features respectively, so as to automatically judge whether the inspected cloth has defects according to the multi-scale cascade coding between the enhanced cloth shape enhancement coding features, cloth texture enhancement coding features and cloth color enhancement coding features. In this way, using a high-resolution line scan camera to continuously scan the cloth can ensure that all areas of the cloth are covered to avoid missing potential defects. In addition, the YOLO algorithm and deep learning technology can efficiently process high-resolution images, overcome the performance bottleneck of traditional methods when processing large amounts of data, achieve rapid detection, and improve the ability to recognize subtle defects, thereby providing a more intelligent defect inspection solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 It is a flowchart of a defect inspection method for a fabric inspection device based on the YOLO algorithm according to an embodiment of the present application.

[0011] Figure 2 Schematic diagram of data flow of a defect inspection method for a fabric inspection device based on the YOLO algorithm according to an embodiment of the present application.

[0012] Figure 3 The present invention is a flowchart of performing cloth feature space adaptive enhancement on the cloth shape coding feature map, the cloth texture coding feature map and the cloth color coding feature map in the defect inspection method of cloth inspection equipment based on the YOLO algorithm according to an embodiment of the present application to obtain a cloth shape enhanced coding feature map, a cloth texture enhanced coding feature map and a cloth color enhanced coding feature map.

[0013] Figure 4 The present invention is a flowchart of performing pixel granularity semantic association and spatial attenuation modulation on the set of cloth shape pixel granularity feature vectors in the cloth inspection equipment defect inspection method based on the YOLO algorithm according to an embodiment of the present application to obtain a cloth shape semantic association spatial soft constraint score topology matrix.

[0014] Figure 5 4 is a block diagram of a defect inspection system for fabric inspection equipment based on the YOLO algorithm according to an embodiment of the present application. DETAILED DESCRIPTION

[0015] The embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, but rather these embodiments are provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0016] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0017] In the description of the embodiments of the present disclosure, the term "including" and similar terms should be understood as open inclusion, that is, "including but not limited to". The term "based on" should be understood as "based at least in part on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc. may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0018] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0019] In the textile industry, fabric inspection is a key step to ensure product quality. The quality of fabric not only affects the aesthetic appearance of the final product, but also affects the durability and functionality of the product. Therefore, the fabric inspection process is of great significance to maintaining brand reputation and customer satisfaction.

[0020] With the development of technology, automated fabric inspection equipment has gradually become mainstream. These devices usually use image processing technology to identify defects on fabrics. However, existing automated fabric inspection equipment faces challenges in efficiency and accuracy when processing high-resolution images. Specifically, at present, most automated fabric inspection equipment uses fixed cameras to capture images, and then uses traditional image processing algorithms to detect defects. Traditional image processing algorithms may have difficulty accurately distinguishing normal features from defects, resulting in a high false alarm rate, and these methods may encounter performance bottlenecks when processing large amounts of data, resulting in slow detection speeds and prone to false alarms. In addition, most automated fabric inspection equipment uses fixed cameras to capture images. This setting may not be able to fully cover all areas of the fabric and is prone to missing potential defects.

[0021] Therefore, in view of the above technical problems, the present application proposes a defect inspection method for a cloth inspection device based on the YOLO algorithm, which continuously scans the inspected cloth using a high-resolution line scan camera to obtain a set of cloth images, and uses image recognition and analysis technology based on deep learning to splice the set of cloth images to form a complete continuous cloth image, then uses the YOLO algorithm to capture the shape, texture and color features of the spliced ​​continuous cloth image, and performs feature space adaptive enhancement processing on these features respectively, so as to automatically judge whether the inspected cloth has defects according to the multi-scale cascade coding between the enhanced cloth shape enhancement coding features, cloth texture enhancement coding features and cloth color enhancement coding features. In this way, using a high-resolution line scan camera to continuously scan the cloth can ensure that all areas of the cloth are covered to avoid missing potential defects. In addition, the YOLO algorithm and deep learning technology can efficiently process high-resolution images, overcome the performance bottleneck of traditional methods in processing large amounts of data, achieve rapid detection, and improve the ability to recognize subtle defects, thereby providing a more intelligent defect inspection solution.

[0022] Figure 1 It is a flowchart of a defect inspection method for a fabric inspection device based on the YOLO algorithm according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of a defect inspection method for a fabric inspection device based on a YOLO algorithm according to an embodiment of the present application. Figure 1 and Figure 2 As shown, according to the defect inspection method of the cloth inspection equipment based on the YOLO algorithm according to the embodiment of the present application, the method includes: S110, using a high-resolution line scan camera to continuously scan the inspected cloth to obtain a set of cloth images; S120, performing image stitching on the set of cloth images to obtain a continuous cloth image; S130, performing multi-feature capture on the continuous cloth image to obtain a cloth shape coding feature map, a cloth texture coding feature map and a cloth color coding feature map; S140, performing cloth feature space adaptive enhancement on the cloth shape coding feature map, the cloth texture coding feature map and the cloth color coding feature map to obtain a cloth shape enhanced coding feature map, a cloth texture enhanced coding feature map and a cloth color enhanced coding feature map; S150, based on the cloth multi-scale cascade coding features between the cloth shape enhanced coding feature map, the cloth texture enhanced coding feature map and the cloth color enhanced coding feature map, obtaining a recognition result, and the recognition result is used to indicate whether the inspected cloth has defects.

[0023] In step S110, a high-resolution line scan camera is used to continuously scan the inspected cloth to obtain a set of cloth images. It should be understood that the cloth image refers to a photo or image data of the cloth surface taken by a camera. These images contain the color, texture and other visual features of the cloth, which can be used to evaluate the quality of the cloth, find defects or record its appearance attributes. In the textile industry, it is very important to ensure that the cloth is flawless, because any small defect may affect the quality of the final product and customer satisfaction. Based on this, in order to be able to more accurately analyze the features in the cloth image, in the technical solution of the present application, a high-resolution line scan camera is used to continuously scan the inspected cloth to obtain a set of cloth images. It is worth mentioning that the high-resolution line scan camera is a device specially designed to quickly capture detailed images on moving objects. Unlike traditional area array cameras, line scan cameras only capture information of one line or one row of pixels at a time. Specifically, when the object moves in front of the camera, the line scan camera continuously captures a series of linear images, which can be combined to form a complete two-dimensional image. Therefore, high-resolution line scan cameras are often used in application scenarios that require high-speed and high-precision image acquisition, such as quality control on industrial production lines, printed product inspection, agricultural product grading, etc.

[0024] In step S120, the set of cloth images is stitched to obtain a continuous cloth image. Accordingly, considering that the line scan camera can only capture one line or one row of pixels of the image at a time, and a single image can only reflect a small part of the cloth information. Therefore, in order to make the entire cloth surface a unified whole and facilitate the consistency analysis of the entire area, in the technical solution of the present application, the set of cloth images is stitched to obtain a continuous cloth image. In this way, by stitching together multiple such linear images, a complete and continuous cloth image can be obtained, which can fully cover the surface of the entire cloth, thereby providing a complete view, which provides a basis and support for subsequent defect detection.

[0025] In step S130, the continuous image of the cloth is subjected to multi-feature capture to obtain a cloth shape coding feature map, a cloth texture coding feature map and a cloth color coding feature map. Specifically, in an embodiment of the present application, the continuous image of the cloth is subjected to multi-feature capture to obtain a cloth shape coding feature map, a cloth texture coding feature map and a cloth color coding feature map, including: inputting the continuous image of the cloth into a cloth feature extractor based on a YOLO model to obtain the cloth shape coding feature map, the cloth texture coding feature map and the cloth color coding feature map. It should be understood that considering that the continuous image of the cloth contains rich and large amounts of feature information, such as the shape, texture and color of the cloth, and the quality of the cloth does not only depend on a single feature, but is jointly determined by multiple factors. Considering that the YOLO model can extract multiple features of the cloth at the same time, such as shape, texture and color, this is crucial for a comprehensive assessment of the quality of the cloth. Therefore, in order to analyze and process the characteristic information of each cloth more carefully, in the technical solution of the present application, the continuous cloth image is input into the cloth feature extractor based on the YOLO model to obtain the cloth shape coding feature map, the cloth texture coding feature map and the cloth color coding feature map. It can be understood that YOLO, as a real-time target detection algorithm, has obvious advantages in processing speed. At the same time, the deep learning architecture behind the YOLO model (such as the original Darknet or other YOLO-supporting architectures such as ResNet, etc.) usually contains multiple convolutional layers, which can effectively extract multi-level features from the image, such as features including edges, textures, colors and other information, which is crucial for identifying complex patterns and potential defects on cloth.

[0026] In step S140, the cloth shape coding feature map, the cloth texture coding feature map and the cloth color coding feature map are adaptively enhanced in cloth feature space to obtain a cloth shape enhanced coding feature map, a cloth texture enhanced coding feature map and a cloth color enhanced coding feature map. Figure 3 The present invention is a flowchart of performing cloth feature space adaptive enhancement on the cloth shape coding feature map, the cloth texture coding feature map and the cloth color coding feature map in the defect inspection method of the cloth inspection equipment based on the YOLO algorithm according to the embodiment of the present application to obtain the cloth shape enhanced coding feature map, the cloth texture enhanced coding feature map and the cloth color enhanced coding feature map. Specifically, in the embodiment of the present application, Figure 3As shown, the cloth shape coding feature map, the cloth texture coding feature map and the cloth color coding feature map are subjected to cloth feature space adaptive enhancement to obtain a cloth shape enhanced coding feature map, a cloth texture enhanced coding feature map and a cloth color enhanced coding feature map, including: S210, feature dispersion of the cloth shape coding feature map to obtain a set of cloth shape pixel granularity feature vectors; S220, pixel granularity semantic association and spatial attenuation modulation of the set of cloth shape pixel granularity feature vectors to obtain a cloth shape semantic association space soft constraint score topology matrix; S230, graph structure space enhancement of the cloth shape semantic association space soft constraint score topology matrix and the set of cloth shape pixel granularity features to obtain the cloth shape enhanced coding feature map. Accordingly, considering that the cloth shape coding feature map, the cloth texture coding feature map and the cloth color coding feature map contain local details and textures about the cloth, such as edges, small creases, etc., and in order to be able to suppress background noise while enhancing these key features, so as to highlight those features that are crucial for defect detection, in the technical solution of the present application, the cloth shape coding feature map, the cloth texture coding feature map and the cloth color coding feature map are subjected to cloth feature space adaptive enhancement to obtain a cloth shape enhanced coding feature map, a cloth texture enhanced coding feature map and a cloth color enhanced coding feature map. In particular, the processing method of the cloth shape coding feature map is taken as an example for specific description.

[0027] Specifically, first, the cloth shape coding feature map is feature scattered to convert its shape into pixel granularity features to obtain a set of cloth shape pixel granularity feature vectors. In particular, here, the feature scattered of the cloth shape coding feature map is performed along the channel dimension.

[0028] Next, pixel granularity semantic association and spatial attenuation modulation are performed on the set of the cloth shape pixel granularity feature vectors to obtain a cloth shape semantic association space soft constraint score topological matrix. Specifically, first, the set of cloth shape pixel granularity feature vectors is mapped to the Poincare space to obtain a set of mapped cloth shape pixel granularity feature vectors. In particular, the Poincare space is a hyperbolic geometric space, which is usually used to process data with a hierarchical structure or a long-tail distribution, that is, cloth features are usually hierarchical, such as the overall color distribution and local color difference of the cloth, and the Poincare space can better capture the relationship between the data. Afterwards, the Poincare distance between any two mapped pixel granularity feature vectors is calculated as the semantic association score between any two mapped pixel granularity feature vectors to obtain the cloth shape semantic association score topological matrix, so that even a small feature change can produce a larger distance in the mapping, which helps to improve the model's ability to distinguish different feature categories. Subsequently, considering that the association between adjacent regions is often closer than that between far-separated regions, more emphasis should be placed on the mutual influence between adjacent regions when scoring the topological matrix. Therefore, the cloth shape semantic association score topological matrix is ​​further spatially attended to and attenuated modulated based on the spatial distance between any two cloth shape pixel granularity feature vectors to obtain the cloth shape semantic association spatial soft constraint score topological matrix.

[0029] Then, the cloth shape semantic association space soft constraint score topology matrix and the set of cloth shape pixel granularity features are subjected to graph structure space enhancement to obtain the cloth shape enhanced coding feature map. Specifically, in order to capture and extract the complex semantic association relationship between each pixel granularity feature vector, the cloth shape semantic association space soft constraint score topology matrix is ​​convolved by using hole convolution coding to obtain the cloth shape semantic association space soft constraint score topology feature matrix. Afterwards, a graph convolution coding module is used to perform global contextual semantic association on the spatial soft constraint score topology feature matrix and the set of cloth shape pixel granularity features to obtain a set of cloth shape contextual pixel granularity feature vectors. Finally, the set feature shape of the cloth shape contextual pixel granularity feature vector is reshaped to obtain a cloth shape enhanced coding feature map with significant features and key information, so as to further improve the expression ability and detection accuracy of the cloth shape features.

[0030] Specifically, in step S210, the cloth shape coding feature map is feature dispersed to obtain a set of cloth shape pixel granularity feature vectors, including: feature dispersed along the channel dimension of the cloth shape coding feature map to obtain a set of cloth shape pixel granularity feature vectors.

[0031] Figure 4The present invention is a flowchart of performing pixel granularity semantic association and spatial attenuation modulation on the set of the cloth shape pixel granularity feature vectors in the defect inspection method of the cloth inspection equipment based on the YOLO algorithm according to the embodiment of the present application to obtain the cloth shape semantic association space soft constraint score topology matrix. Specifically, in the embodiment of the present application, Figure 4 As shown, pixel granularity semantic association and spatial attenuation modulation are performed on the set of cloth shape pixel granularity feature vectors to obtain the cloth shape semantic association spatial soft constraint score topology matrix, including: S310, calculating the semantic association score between any two cloth shape pixel granularity feature vectors in the set of cloth shape pixel granularity feature vectors to obtain the cloth shape semantic association score topology matrix; S320, spatial attention attenuation modulation is performed on the cloth shape semantic association score topology matrix based on the spatial distance between any two cloth shape pixel granularity feature vectors in the set of cloth shape pixel granularity feature vectors to obtain the cloth shape semantic association spatial soft constraint score topology matrix.

[0032] Specifically, in an embodiment of the present application, in step S310, the semantic association score between any two cloth shape pixel granularity feature vectors in the set of cloth shape pixel granularity feature vectors is calculated to obtain a cloth shape semantic association score topological matrix, including: mapping each cloth shape pixel granularity feature vector in the set of cloth shape pixel granularity feature vectors to the Poincare space to obtain a set of mapped cloth shape pixel granularity feature vectors; calculating the semantic association score between any two mapped cloth shape pixel granularity feature vectors in the set of mapped cloth shape pixel granularity feature vectors to obtain the cloth shape semantic association score topological matrix composed of multiple cloth shape semantic association score values.

[0033] More specifically, in an embodiment of the present application, each cloth shape pixel granularity feature vector in the set of cloth shape pixel granularity feature vectors is mapped to the Poincare space to obtain a set of mapped cloth shape pixel granularity feature vectors, including: matrix-vector multiplication of a first weight matrix and the cloth shape pixel granularity feature vector, and then vector-matrix multiplication of the obtained cloth shape pixel granularity weight vector and a second weight matrix to obtain the mapped cloth shape pixel granularity feature vector.

[0034] More specifically, in the embodiment of the present application, the semantic association score between any two mapped cloth shape pixel granularity feature vectors in the set of mapped cloth shape pixel granularity feature vectors is calculated to obtain the cloth shape semantic association score topological matrix composed of multiple cloth shape semantic association score values, including: extracting the first cloth shape pixel granularity feature vector from the set of mapped cloth shape pixel granularity feature vectors; The pixel granularity feature vector of the cloth shape after mapping and the the pixel granularity feature vector of the cloth shape after mapping; calculating the the mapped cloth shape pixel granularity feature vector and the The position difference between the mapped cloth shape pixel granularity feature vectors is used to obtain the mapped cloth shape pixel granularity differential vector, and the square of the norm of the mapped cloth shape pixel granularity differential vector is calculated to obtain the mapped cloth shape feature differential norm value; the constant one is calculated and the first The difference between the squares of the norms of the pixel granularity feature vectors of the cloth shape after mapping is used to obtain the first The difference of the cloth shape features after mapping; calculating the constant one and the first The difference between the squares of the norms of the pixel granularity feature vectors of the cloth shape after mapping is used to obtain the first The mapped cloth shape feature difference is calculated as follows: The difference between the mapped cloth shape features and the The product of the difference values ​​of the mapped cloth shape features is obtained by multiplying the difference norm value of the mapped cloth shape features by the product value of the mapped cloth shape features, multiplying the obtained value by a constant two and then adding the value to a constant one to obtain a semantic metric value of the mapped cloth shape features, and inputting the semantic metric value of the mapped cloth shape features into an arccosh function to obtain the first the mapped cloth shape pixel granularity feature vector and the The cloth shape semantic association score value between the mapped cloth shape pixel granularity feature vectors.

[0035] Specifically, in an embodiment of the present application, in step S320, the cloth shape semantic association score topology matrix is ​​spatially focused attenuated modulated based on the spatial distance between any two cloth shape pixel granularity feature vectors in the set of cloth shape pixel granularity feature vectors to obtain the cloth shape semantic association space soft constraint score topology matrix, including: calculating the spatial distance between any two cloth shape pixel granularity feature vectors to obtain a set of cloth shape space distance values; calculating an exponential function with the natural constant e as the base and each cloth shape space distance value in the set of cloth shape space distance values ​​as the exponent to obtain a set of cloth shape space distance index values; multiplying the set of cloth shape space distance index values ​​by the attenuation modulation value by position to obtain a set of cloth shape space distance weighted values; dividing each cloth shape semantic association score value in the cloth shape semantic association score topology matrix by the corresponding cloth shape space distance weighted value to obtain the cloth shape semantic association space soft constraint score topology matrix composed of multiple cloth shape semantic association space soft constraint score values.

[0036] Specifically, in an embodiment of the present application, in step S230, the cloth shape semantic association space soft constraint score topological matrix and the set of cloth shape pixel granularity features are subjected to graph structure space enhancement to obtain the cloth shape enhanced coding feature map, including: performing hole convolution coding on the cloth shape semantic association space soft constraint score topological matrix to obtain the cloth shape semantic association space soft constraint score topological feature matrix; inputting the set of cloth shape pixel granularity feature vectors and the cloth shape semantic association space soft constraint score topological feature matrix into a graph convolution coding module to obtain a set of cloth shape context pixel granularity feature vectors; and performing feature shape reshaping on the set of cloth shape context pixel granularity feature vectors to obtain the cloth shape enhanced coding feature map.

[0037] In the embodiment of the present application, specifically, the cloth shape coding feature map is subjected to cloth feature space adaptive feature enhancement, which can be expressed by the formula: ;in, is the cloth shape coding feature map, and , is the set of real numbers, and Respectively represent the width and height of each feature matrix of the cloth shape coding feature map along the channel dimension, represents the number of channels of the cloth shape coding feature map, To reshape the feature map, is the first in the set of cloth shape pixel granularity feature vectors cloth shape pixel granularity feature vectors, and , is the set of pixel granularity feature vectors of the cloth shape, and are the first weight matrix and the second weight matrix respectively, is the set of pixel granularity feature vectors of the mapped cloth shape, are respectively the first, second, ..., first, second, third, fourth, fifth, fifth, sixth, sixth, fifth, sixth, fifth, sixth, seventh ... ,..., ,..., The pixel granularity feature vector of the cloth shape after mapping, is the square of the one-norm of the vector, is the inverse hyperbolic cosine function, For calculation and The cloth shape semantic association score between is the eigenvalue of each position in the cloth shape semantic association score topological matrix, for and The cloth shape space distance value between them, is the attenuation modulation value, is the eigenvalue of each position in the soft constraint score topology matrix of the cloth shape semantic association space, is the soft constraint score topological matrix of the cloth shape semantic association space, is the dilated convolutional coding operation, is the soft constraint score topological feature matrix of the cloth shape semantic association space, is the graph convolutional encoding operation, is the set of context pixel granularity feature vectors of cloth shape, is the cloth shape enhancement coding feature map. Similarly, the processing method of the cloth texture enhancement coding feature map and the cloth color enhancement coding feature map is as shown in the above formula.

[0038] In step S150, based on the cloth multi-scale cascade coding features between the cloth shape enhancement coding feature map, the cloth texture enhancement coding feature map, and the cloth color enhancement coding feature map, a recognition result is obtained, and the recognition result is used to indicate whether the inspected cloth has defects. Specifically, in an embodiment of the present application, based on the cloth multi-scale cascade coding features between the cloth shape enhancement coding feature map, the cloth texture enhancement coding feature map, and the cloth color enhancement coding feature map, the recognition result is used to indicate whether the inspected cloth has defects, including: cascading the cloth shape enhancement coding feature map, the cloth texture enhancement coding feature map, and the cloth color enhancement coding feature map to obtain a cloth multi-scale cascade coding feature map as the cloth multi-scale cascade coding feature; inputting the cloth multi-scale cascade coding feature map into a classifier-based defect equipment module to obtain the recognition result, and the recognition result is used to indicate whether the inspected cloth has defects. It should be understood that, considering that the cloth shape enhancement coding feature map, the cloth texture enhancement coding feature map and the cloth color enhancement coding feature map respectively contain different aspects of visual features, therefore, in order to comprehensively consider multiple visual characteristics, so as to obtain a more comprehensive cloth feature representation, in the technical solution of the present application, the cloth shape enhancement coding feature map, the cloth texture enhancement coding feature map and the cloth color enhancement coding feature map are cascaded to obtain a cloth multi-scale cascade coding feature map. That is, the obtained cloth multi-scale cascade coding feature map can better capture various defects on the cloth, whether it is uneven color, abnormal texture or shape defects, so as to improve the comprehensiveness and accuracy of defect detection. Afterwards, the cloth multi-scale cascade coding features obtained by cascading the cloth shape enhancement coding feature map, the cloth texture enhancement coding feature map and the cloth color enhancement coding feature map are used for classification processing, so as to automatically determine whether the inspected cloth has defects. In this way, using a high-resolution line scan camera to continuously scan the cloth can ensure that all areas of the cloth are covered to avoid missing potential defects. In addition, the YOLO algorithm and deep learning technology can efficiently process high-resolution images, overcome the performance bottleneck of traditional methods when processing large amounts of data, achieve rapid detection, and improve the ability to identify subtle defects, thereby providing a more intelligent defect inspection solution.

[0039] It should be understood that when the cloth shape enhancement coding feature map, the cloth texture enhancement coding feature map and the cloth color enhancement coding feature map respectively represent the image semantic features of the cloth continuous image based on the spatial distribution soft constraint local enhancement, when they are cascaded to obtain the cloth multi-scale cascade coding feature map, the cloth multi-scale cascade coding feature map will also cause the diversity of semantic joint representation due to the differences in the semantic depth expression patterns of different images, thereby affecting the iterative consistency of the classification mapping and reducing the accuracy of the classification results.

[0040] Preferably, inputting the cloth multi-scale cascade coding feature map into a classifier-based defect equipment module to obtain an identification result includes: determining a cloth multi-scale cascade coding probability value corresponding to the cloth multi-scale cascade coding feature map, wherein the cloth multi-scale cascade coding probability value represents the probability that the inspected cloth has a defect; expanding the cloth multi-scale cascade coding feature map into a cloth multi-scale cascade coding feature vector; calculating the product of a first cloth multi-scale cascade coding feature value of the cloth multi-scale cascade coding feature vector and the cloth multi-scale cascade coding probability value to obtain a first cloth multi-scale cascade coding feature probability product; calculating the product of a second cloth multi-scale cascade coding feature value of the cloth multi-scale cascade coding feature vector and the difference between one and the cloth multi-scale cascade coding probability value to obtain a second cloth multi-scale cascade coding feature inverse probability product; calculating the quotient of the first cloth multi-scale cascade coding feature value divided by the difference between one and the cloth multi-scale cascade coding probability value to obtain a first cloth multi-scale cascade coding feature probability product. Coding feature inverse probability quotient; calculating the quotient of the second cloth multi-scale cascade coding feature value and the cloth multi-scale cascade coding probability value to obtain the second cloth multi-scale cascade coding feature probability quotient; adding the first cloth multi-scale cascade coding feature probability product and the second cloth multi-scale cascade coding feature inverse probability product, subtracting the difference between the first cloth multi-scale cascade coding feature inverse probability quotient and the second cloth multi-scale cascade coding feature probability quotient to obtain cloth multi-scale cascade coding correction feature value; for each first cloth multi-scale cascade coding feature value and the second cloth multi-scale cascade coding feature value of the cloth multi-scale cascade coding feature vector, matrix multiplication of the cloth multi-scale cascade coding correction feature matrix composed of the cloth multi-scale cascade coding correction feature value and the cloth multi-scale cascade coding feature vector to obtain a corrected cloth multi-scale cascade coding feature vector; inputting the corrected cloth multi-scale cascade coding feature vector into a classifier-based defect device module to obtain a recognition result.

[0041] Among them, the cloth multi-scale cascade coding feature vector The correction process is expressed as: ;in, is the multi-scale cascade coding feature vector of the cloth, and are the first Position and The eigenvalues ​​at the positions, is the probability that the inspected cloth has defects, represents the first The multi-scale cascade coding correction feature matrix of the cloth at each position, is the cloth multi-scale cascade coding correction feature matrix, represents matrix multiplication, A cloth multi-scale cascade encoding feature vector representing the correction.

[0042] Therefore, the present application aims at the region-boundary integral relationship of the high-dimensional feature manifold of the multi-scale cascade coding feature map of the cloth, and defines the probability distribution of the feature set of the multi-scale cascade coding feature map of the cloth as a manifold constraint boundary to approximate the high-dimensional manifold of the feature set of the multi-scale cascade coding feature map of the cloth, which is composed of a single connected region representation of eigenvalue pairs, so as to avoid the mapping ambiguity of the diversified feature representation of the multi-scale cascade coding feature map of the cloth to the local manifold representation in the high-dimensional feature class convergence space, and improve the execution iteration consistency of each local feature distribution in the mapping task, so as to improve the feature classification convergence effect, and improve the accuracy of the recognition result obtained by the defect device module based on the classifier of the multi-scale cascade coding feature map of the cloth. In this way, the cloth is continuously scanned using a high-resolution line scan camera to ensure that all areas of the cloth are covered and potential defects are avoided. In addition, the YOLO algorithm and deep learning technology can efficiently process high-resolution images, overcome the performance bottleneck of traditional methods in processing large amounts of data, achieve rapid detection, and improve the recognition ability of subtle defects, thereby providing a more intelligent defect inspection solution.

[0043] In summary, the defect inspection method of the cloth inspection equipment based on the YOLO algorithm according to the embodiment of the present application is explained, which uses a high-resolution line scan camera to continuously scan the inspected cloth to obtain a set of cloth images, and uses image recognition and analysis technology based on deep learning to splice the set of cloth images to form a complete continuous cloth image, then, the YOLO algorithm is used to capture the shape, texture and color features of the spliced ​​continuous cloth image, and the feature space adaptive enhancement processing is performed on these features respectively, so as to automatically judge whether the inspected cloth has defects according to the multi-scale cascade coding between the enhanced cloth shape enhancement coding features, cloth texture enhancement coding features and cloth color enhancement coding features. In this way, the cloth is continuously scanned using a high-resolution line scan camera to ensure that all areas of the cloth are covered and potential defects are avoided from being missed. In addition, the YOLO algorithm and deep learning technology can efficiently process high-resolution images, overcome the performance bottleneck of traditional methods in processing large amounts of data, achieve rapid detection, and improve the ability to recognize subtle defects, thereby providing a more intelligent defect inspection solution.

[0044] Figure 5 FIG. 1 is a block diagram of a defect inspection system for a fabric inspection device based on a YOLO algorithm according to an embodiment of the present application. Figure 5As shown, according to the embodiment of the present application, the defect inspection system 100 of the cloth inspection equipment based on the YOLO algorithm includes: a cloth image scanning module 110, which is used to use a high-resolution line scan camera to continuously scan the inspected cloth to obtain a set of cloth images; a cloth image stitching module 120, which is used to stitch the set of cloth images to obtain a continuous cloth image; a cloth image multi-feature capture module 130, which is used to perform multi-feature capture on the continuous cloth image to obtain a cloth shape coding feature map, a cloth texture coding feature map and a cloth color coding feature map; a cloth feature space enhancement module 140, which is used to perform cloth feature space adaptive enhancement on the cloth shape coding feature map, the cloth texture coding feature map and the cloth color coding feature map to obtain a cloth shape enhanced coding feature map, a cloth texture enhanced coding feature map and a cloth color enhanced coding feature map, wherein the cloth feature space enhancement module 140 is used to perform cloth feature space adaptive enhancement on the cloth shape coding feature map, the cloth texture enhanced coding feature map and the cloth color enhanced coding feature map. The strong module includes: a cloth feature dispersion unit, which is used to perform feature dispersion on the cloth shape coding feature map to obtain a set of cloth shape pixel granularity feature vectors; a cloth feature semantic association modulation unit, which is used to perform pixel granularity semantic association and spatial attenuation modulation on the set of cloth shape pixel granularity feature vectors to obtain a cloth shape semantic association space soft constraint score topology matrix; a cloth feature graph structure enhancement unit, which is used to perform graph structure space enhancement on the cloth shape semantic association space soft constraint score topology matrix and the set of cloth shape pixel granularity features to obtain the cloth shape enhanced coding feature map; a recognition result generation module 150, which is used to obtain a recognition result based on the cloth multi-scale cascade coding features between the cloth shape enhanced coding feature map, the cloth texture enhanced coding feature map and the cloth color enhanced coding feature map, and the recognition result is used to indicate whether the inspected cloth has defects.

[0045] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned fabric inspection equipment defect inspection system based on the YOLO algorithm have been referred to above. Figures 1 to 4 The invention has been described in detail in the description of the defect inspection method of the fabric inspection equipment based on the YOLO algorithm, and therefore, its repeated description will be omitted.

[0046] As described above, the fabric inspection equipment defect inspection system 100 based on the YOLO algorithm according to the embodiment of the present disclosure can be implemented in various wireless terminals, such as a server with a fabric inspection equipment defect inspection algorithm based on the YOLO algorithm. In a possible implementation, the fabric inspection equipment defect inspection system 100 based on the YOLO algorithm according to the embodiment of the present disclosure can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the fabric inspection equipment defect inspection system 100 based on the YOLO algorithm can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the fabric inspection equipment defect inspection system 100 based on the YOLO algorithm can also be one of the many hardware modules of the wireless terminal.

[0047] Alternatively, in another example, the fabric inspection equipment defect inspection system 100 based on the YOLO algorithm and the wireless terminal may also be separate devices, and the fabric inspection equipment defect inspection system 100 based on the YOLO algorithm may be connected to the wireless terminal via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0048] The above is merely an example of the principles of the present disclosure, and various modifications may be made by those skilled in the art without departing from the scope of the present disclosure. The above embodiments are presented for the purpose of illustration rather than limitation. The present disclosure may also take many forms other than those explicitly described herein.

Claims

1. A defect inspection method for fabric inspection equipment based on YOLO algorithm, characterized in that: include: Using a high-resolution line scan camera to continuously scan the inspected cloth to obtain a collection of cloth images; Performing image stitching on the set of cloth images to obtain a continuous cloth image; Capturing multiple features of the continuous cloth images to obtain a cloth shape coding feature map, a cloth texture coding feature map, and a cloth color coding feature map; The cloth shape coding feature map, the cloth texture coding feature map and the cloth color coding feature map are subjected to cloth feature space adaptive enhancement to obtain a cloth shape enhanced coding feature map, a cloth texture enhanced coding feature map and a cloth color enhanced coding feature map, including: performing feature dispersion on the cloth shape coding feature map to obtain a set of cloth shape pixel granularity feature vectors; performing pixel granularity semantic association and spatial attenuation modulation on the set of cloth shape pixel granularity feature vectors to obtain a cloth shape semantic association space soft constraint score topology matrix; performing graph structure space enhancement on the cloth shape semantic association space soft constraint score topology matrix and the set of cloth shape pixel granularity features to obtain the cloth shape enhanced coding feature map; Based on the cloth multi-scale cascade coding features between the cloth shape enhancement coding feature map, the cloth texture enhancement coding feature map and the cloth color enhancement coding feature map, an identification result is obtained, and the identification result is used to indicate whether the inspected cloth has defects.

2. The defect inspection method for fabric inspection equipment based on the YOLO algorithm according to claim 1, characterized in that: The method further comprises: performing multi-feature capture on the continuous cloth image to obtain a cloth shape coding feature map, a cloth texture coding feature map and a cloth color coding feature map, comprising: inputting the continuous cloth image into a cloth feature extractor based on a YOLO model to obtain the cloth shape coding feature map, the cloth texture coding feature map and the cloth color coding feature map.

3. The defect inspection method for fabric inspection equipment based on the YOLO algorithm according to claim 2 is characterized in that: Feature dispersion is performed on the cloth shape coding feature map to obtain a set of cloth shape pixel granularity feature vectors, including: feature dispersion is performed on the cloth shape coding feature map along a channel dimension to obtain a set of cloth shape pixel granularity feature vectors.

4. The defect inspection method for fabric inspection equipment based on the YOLO algorithm according to claim 3 is characterized in that: The set of the cloth shape pixel granularity feature vectors is subjected to pixel granularity semantic association and spatial attenuation modulation to obtain a cloth shape semantic association space soft constraint score topology matrix, including: Calculating the semantic association score between any two cloth shape pixel granularity feature vectors in the set of cloth shape pixel granularity feature vectors to obtain a cloth shape semantic association score topological matrix; Based on the spatial distance between any two cloth shape pixel granularity feature vectors in the set of cloth shape pixel granularity feature vectors, the cloth shape semantic association score topological matrix is ​​spatially attended to attenuation modulated to obtain the cloth shape semantic association spatial soft constraint score topological matrix.

5. The defect inspection method for fabric inspection equipment based on the YOLO algorithm according to claim 4 is characterized in that: Calculating the semantic association score between any two cloth shape pixel granularity feature vectors in the set of cloth shape pixel granularity feature vectors to obtain a cloth shape semantic association score topological matrix, including: Mapping each cloth shape pixel granularity feature vector in the set of cloth shape pixel granularity feature vectors to the Poincare space to obtain a set of mapped cloth shape pixel granularity feature vectors; Calculating the semantic association score between any two mapped cloth shape pixel granularity feature vectors in the set of mapped cloth shape pixel granularity feature vectors to obtain the cloth shape semantic association score topological matrix composed of a plurality of cloth shape semantic association score values; Among them, each cloth shape pixel granularity feature vector in the set of cloth shape pixel granularity feature vectors is mapped to the Poincare space to obtain a set of mapped cloth shape pixel granularity feature vectors, including: after matrix-vector multiplication of a first weight matrix and the cloth shape pixel granularity feature vector, vector-matrix multiplication of the obtained cloth shape pixel granularity weight vector and a second weight matrix to obtain the mapped cloth shape pixel granularity feature vector.

6. The defect inspection method for fabric inspection equipment based on the YOLO algorithm according to claim 5, characterized in that: Calculating the semantic association score between any two mapped cloth shape pixel granularity feature vectors in the set of mapped cloth shape pixel granularity feature vectors to obtain the cloth shape semantic association score topological matrix composed of multiple cloth shape semantic association score values, including: Extract the first The pixel granularity feature vector of the cloth shape after mapping and the The pixel granularity feature vector of the cloth shape after mapping; Calculate the the mapped cloth shape pixel granularity feature vector and the Obtaining a mapped cloth shape pixel granularity differential vector according to the position difference between the mapped cloth shape pixel granularity feature vectors, and calculating the square of a norm of the mapped cloth shape pixel granularity differential vector to obtain a mapped cloth shape feature differential norm value; Calculate the constant one and the The difference between the squares of the norms of the pixel granularity feature vectors of the cloth shape after mapping is used to obtain the first The difference of cloth shape features after mapping; Calculate the constant one and the The difference between the squares of the norms of the pixel granularity feature vectors of the cloth shape after mapping is used to obtain the first The difference of cloth shape features after mapping; Calculate the The difference between the mapped cloth shape features and the The product of the mapped cloth shape feature difference values ​​is obtained to obtain the mapped cloth shape feature product value; After dividing the differential norm value of the mapped cloth shape feature by the product value of the mapped cloth shape feature, the obtained value is first multiplied by a constant 2 and then added to a constant 1 to obtain a semantic metric value of the mapped cloth shape feature; The mapped cloth shape feature semantic metric value is input into the arccosh function to obtain the first the mapped cloth shape pixel granularity feature vector and the The cloth shape semantic association score value between the mapped cloth shape pixel granularity feature vectors.

7. The defect inspection method for fabric inspection equipment based on the YOLO algorithm according to claim 6, characterized in that: Based on the spatial distance between any two cloth shape pixel granularity feature vectors in the set of cloth shape pixel granularity feature vectors, the cloth shape semantic association score topology matrix is ​​subjected to spatial attention attenuation modulation to obtain the cloth shape semantic association spatial soft constraint score topology matrix, including: Calculating the spatial distance between any two cloth shape pixel granularity feature vectors to obtain a set of cloth shape spatial distance values; Calculate an exponential function with the natural constant e as the base and each cloth shape space distance value in the set of cloth shape space distance values ​​as the exponent to obtain a set of cloth shape space distance index values; Multiplying the set of cloth shape space distance index values ​​by the attenuation modulation value by position to obtain a set of cloth shape space distance weighted values; Each cloth shape semantic association score value in the cloth shape semantic association score topological matrix is ​​divided by the corresponding cloth shape space distance weighted value to obtain the cloth shape semantic association space soft constraint score topological matrix composed of multiple cloth shape semantic association space soft constraint score values.

8. The defect inspection method for fabric inspection equipment based on the YOLO algorithm according to claim 7, characterized in that: Performing graph structure space enhancement on the set of the cloth shape semantic association space soft constraint score topology matrix and the cloth shape pixel granularity feature to obtain the cloth shape enhanced encoding feature map, including: Performing dilated convolution coding on the cloth shape semantic association space soft constraint score topology matrix to obtain a cloth shape semantic association space soft constraint score topology feature matrix; Inputting the set of cloth shape pixel granularity feature vectors and the cloth shape semantic association space soft constraint score topological feature matrix into a graph convolutional coding module to obtain a set of cloth shape context pixel granularity feature vectors; The set of the cloth shape context pixel granularity feature vectors is reshaped to obtain the cloth shape enhanced encoding feature map.

9. The defect inspection method for fabric inspection equipment based on the YOLO algorithm according to claim 8, characterized in that: Based on the cloth multi-scale cascade coding features between the cloth shape enhancement coding feature map, the cloth texture enhancement coding feature map and the cloth color enhancement coding feature map, the recognition result is used to indicate whether the inspected cloth has defects, including: Cascading the cloth shape enhancement coding feature map, the cloth texture enhancement coding feature map and the cloth color enhancement coding feature map to obtain a cloth multi-scale cascade coding feature map as the cloth multi-scale cascade coding feature; The multi-scale cascade coding feature map of the cloth is input into a classifier-based defect equipment module to obtain the recognition result, and the recognition result is used to indicate whether the inspected cloth has defects.

10. A defect inspection system for fabric inspection equipment based on YOLO algorithm, characterized in that: include: The cloth image scanning module is used to continuously scan the inspected cloth using a high-resolution line scan camera. Get a collection of cloth images; A cloth image stitching module, used for stitching the set of cloth images to obtain a continuous cloth image; A cloth image multi-feature capture module, used for performing multi-feature capture on the continuous cloth image to obtain a cloth shape coding feature map, a cloth texture coding feature map and a cloth color coding feature map; A cloth feature space enhancement module, used for performing cloth feature space adaptive enhancement on the cloth shape coding feature map, the cloth texture coding feature map and the cloth color coding feature map to obtain a cloth shape enhanced coding feature map, a cloth texture enhanced coding feature map and a cloth color enhanced coding feature map, wherein the cloth feature space enhancement module comprises: a cloth feature dispersion unit, used for performing feature dispersion on the cloth shape coding feature map to obtain a set of cloth shape pixel granularity feature vectors; a cloth feature semantic association modulation unit, used for performing pixel granularity semantic association and spatial attenuation modulation on the set of cloth shape pixel granularity feature vectors to obtain a cloth shape semantic association space soft constraint score topology matrix; a cloth feature graph structure enhancement unit, used for performing graph structure space enhancement on the cloth shape semantic association space soft constraint score topology matrix and the set of cloth shape pixel granularity features to obtain the cloth shape enhanced coding feature map; The recognition result generation module is used to obtain a recognition result based on the cloth multi-scale cascade coding features between the cloth shape enhancement coding feature map, the cloth texture enhancement coding feature map and the cloth color enhancement coding feature map, and the recognition result is used to indicate whether the inspected cloth has defects.

Images