Textile cloth cover defect intelligent detection method based on machine vision

Abstract

The invention relates to the field of image processing, in particular to a textile cloth surface defect intelligent detection method based on machine vision, and the method comprises the steps: obtaining an original gray level image of a textile cloth surface at the current moment; calculating a texture confidence index based on gray similarity distribution of pixel points and neighborhood pixels in a search radius area of the pixel points, and adaptively inhibiting normal textures by using a ratio relation between a maximum value and a mean value of similarity weights; calculating a form confidence index based on a local brightness contrast characteristic and an isolation characteristic of a pixel point, and filtering fluff interference by integrating brightness, isolation and abnormal characteristics; calculating a gradient sharpness index based on a high-order power operation result of the gray difference of the pixel points in the horizontal and vertical directions; and finally, generating a defect residual image through multi-dimensional feature fusion, and carrying out segmentation identification. According to the method, normal textures, villus, shadows and real defects can be effectively distinguished, and the detection accuracy under a complex background is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of image processing. More particularly, the present application relates to a method for intelligent detection of defects on the surface of a textile based on machine vision. BACKGROUND

[0002] In the large-scale industrial production of textiles, due to factors such as fluctuations in raw material quality, unstable operation of weaving equipment, drift of textile process parameters, and human operation errors, the output fabric surface will inevitably have various types of physical defects such as broken warp, broken weft, holes, and oil stains. The presence of these defects can seriously reduce the quality grade and commercial value of the final product. For a long time, the textile industry has mainly relied on manual visual inspection for quality control. However, this method has strong subjectivity, non-uniform detection standards, and high missed detection rate due to fatigue during long-term operation, which cannot match the high-speed, wide-width, and high-precision production rhythm of modern textile industry.

[0003] To solve the drawbacks of manual detection, using machine vision technology to replace manual detection for automatic detection has become an industry trend. Existing machine vision detection technology usually uses connected component analysis or simple gray threshold segmentation algorithm. However, due to the strong periodic two-dimensional texture structure of the surface of the textile, the traditional detection algorithm is prone to misjudging normal yarn texture fluctuations, fabric surface fluff characteristics, or shadows caused by uneven lighting as defects when processing such images, resulting in a high false positive rate.

[0004] When the existing technical solutions are difficult to balance the contradiction between suppressing background texture and preserving weak defects, they often face a dilemma: if the detection threshold is increased to reduce false alarms, weak defects similar to features and textures, such as fine weft defects or light-colored oil stains, will be missed; conversely, if the threshold is lowered, normal texture features will be falsely reported as defects. Therefore, how to accurately distinguish and suppress normal fabric texture, fluff, and shadow features while completely preserving real defect information is a technical problem that needs to be solved in the current field of intelligent textile detection. SUMMARY

[0005] To solve the problem of difficulty in balancing normal feature suppression and defect feature preservation in a strong texture background, resulting in high false detection rate and missed detection rate, the present application proposes a method for intelligent detection of defects on the surface of a textile based on machine vision, which includes the following steps: Obtain the original gray-scale image of the surface of the textile at the current time; A texture confidence index is calculated based on the gray scale similarity distribution of the pixel point and the neighborhood pixels in the search radius region, the texture confidence index is positively correlated with the maximum value of the similarity weight in the search radius region, and is negatively correlated with the average value of the similarity weight, and the texture confidence index represents the probability of belonging to the normal periodic fabric texture; A morphology confidence index is calculated based on the local brightness contrast feature of the pixel point and the isolated characteristics in the neighborhood, and the morphology confidence index represents the probability of belonging to the fabric surface interference fluff; A gradient sharpness index is calculated based on the gray scale change rate of the pixel point in the horizontal and vertical directions and the high-order energy mapping, and the gradient sharpness index represents the probability of belonging to the non-physical light shadow; According to the texture confidence index, the morphology confidence index and the gradient sharpness index, a defect residual index is calculated through a weighted suppression model, a defect residual map is generated, and the defect residual map is threshold segmented to identify the fabric defects.

[0006] The application realizes the decoupling and suppression of the normal texture, the interference fluff and the light shadow in the fabric background by constructing the confidence indexes of the texture, the morphology and the gradient, and meanwhile, the defect features are completely retained, so that the accuracy and stability of the detection under the complex background are significantly improved.

[0007] Preferably, the original gray scale image of the textile fabric surface at the current time is obtained, and specifically includes: The uniform motion of the fabric is scanned row by row by the industrial line array camera cooperating with the bar light source to obtain the fabric color image; The fabric color image is grayed by using the weighted average method to obtain the original gray scale image with a bit depth of 8.

[0008] Preferably, any pixel point is recorded as a target pixel point, and a texture unit region and a search radius region are defined with the target pixel point as the center; The similarity weight between the target pixel point and each pixel point in the search radius region is calculated; The maximum value and the average value of the similarity weight of the pixel points other than the target pixel point in the search radius region are found; The texture confidence index is calculated based on the maximum value and the average value, and the calculation relationship is:

[0009] Wherein, The texture confidence index of the target pixel point is represented by T, The texture gain coefficient is represented by K; The total number of the pixel points other than the target pixel point in the search radius region is represented by N; ​a similarity weight between a texture unit region centered at a target pixel point x and a texture unit region centered at a pixel point i; for preventing a very small positive number of a denominator from being zero; denotes a maximum value function.

[0010] The application utilizes the periodic recurrence feature of the fabric, and by comparing the maximum similarity and the average similarity of the local texture in the search radius region, the normal texture region can be adaptively identified and marked.

[0011] Preferably, the similarity weight is calculated by: calculating a normalized gray difference distance between the target pixel point and each pixel in the search radius region; mapping the normalized gray difference distance into a similarity weight based on a negative exponential function, and the similarity weight is negatively correlated with the normalized gray difference distance.

[0012] Preferably, the morphological confidence index is calculated by: calculating a bright confidence index of the pixel point, the bright confidence index representing the brightness of the pixel point relative to the neighborhood background; calculating an isolated confidence index of the pixel point, the isolated confidence index representing the spatial isolation degree of the pixel point in the neighborhood; multiplying the product of the bright confidence index, the isolated confidence index and the abnormal saliency index of the pixel point as the morphological confidence index.

[0013] The application constructs a description index for the pile feature, and only the noise points satisfying the high brightness and high isolation degree are determined as the pile and inhibited, thereby effectively reducing the false alarm caused by the rough surface of the fabric.

[0014] Preferably, the bright confidence index of the pixel point is calculated by: calculating the difference between the gray value of the pixel point and the median in the neighborhood as a reference residual; obtaining a dynamic local contrast in the neighborhood; mapping the ratio of the reference residual and the dynamic local contrast by a hyperbolic tangent function to obtain the bright confidence index, and the bright confidence index is positively correlated with the reference residual.

[0015] The application normalizes the local contrast, so that the judgment of the brightness degree is not affected by the overall illumination change, and the bright spot relative to the background is accurately identified.

[0016] Preferably, the isolated confidence index of the pixel point is calculated by: calculate an abnormality saliency index of the pixel point based on the texture confidence index, the abnormality saliency index being negatively correlated with the texture confidence index; calculate a mean value of abnormality saliency indexes of other pixel points in the neighborhood; map the mean value to an isolation confidence index based on a negative exponential function, the isolation confidence index being negatively correlated with the mean value.

[0017] The application effectively distinguishes isolated fluff from connected defects by judging the density of abnormal points in the neighborhood.

[0018] Preferably, the gradient sharpness index is calculated by: respectively calculating a gray value difference of the pixel point in a horizontal direction and a gray value difference of the pixel point in a vertical direction; performing a cubic power operation on the gray value differences to suppress low-amplitude shadow edge gradients and enhance high-amplitude defect edge gradients; mapping the results of the cubic power operation to the gradient sharpness index through a hyperbolic tangent function.

[0019] Preferably, the defect residual index is calculated by a weighted suppression model, and the calculation relationship is:

[0020] wherein, represents the defect residual index of the target pixel point; represents the texture confidence index of the target pixel point; represents the gradient sharpness index of the target pixel point; represents the shape confidence index of the target pixel point; , , respectively represent a texture suppression index, a shadow suppression index, and a fluff suppression index.

[0021] Preferably, the defect residual image is threshold segmented to identify fabric defects, including: performing binaryzation processing on the defect residual image by using an OTSU adaptive threshold algorithm to separate a high-score area from a background; applying a morphological closing operation to fill internal cavities of connected domains and connect broken edges; extracting geometric features of the connected domains, and matching the geometric features with a preset defect physical feature library to output a defect identification result.

[0022] The application has the following beneficial effects: The application realizes adaptive perception of fabric backgrounds of different density and texture types by constructing a texture confidence index and using the self-similarity principle of the fabric. This overcomes the defects of traditional methods that are heavily dependent on fixed empirical parameters and cannot adapt to texture stretching and deformation on the production line, which easily leads to missed detection or false positives.

[0023] Further, a morphology confidence index is introduced, a special filter is constructed using the physical characteristics of bright and isolated villi, which can accurately distinguish and ignore harmless villi on the fabric surface, and significantly reduce false alarms.

[0024] Further, by calculating the gradient sharpness index, the numerical difference between shadows and defects is widened at the physical level by using high-order power operation, effectively suppressing the shadow interference caused by uneven fabric surface or uneven lighting.

[0025] Further, a high signal-to-noise ratio defect residual map is generated by the above multi-dimensional local analysis, which clearly separates defects and normal areas. This makes subsequent simple threshold segmentation sufficient for accurate recognition, thereby significantly improving the accuracy and stability of the entire detection system. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is a step flow chart of a textile fabric surface defect intelligent detection method based on machine vision provided by an embodiment of the application; Figure 2 is a raw gray image provided by an embodiment of the application; Figure 3 is a defect detection image of a traditional method provided by an embodiment of the application; Figure 4 is a residual image provided by an embodiment of the application; Figure 5 is a final defect detection image provided by an embodiment of the application. DETAILED DESCRIPTION

[0027] Please refer to Figure 1 which shows a step flow chart of a textile fabric surface defect intelligent detection method based on machine vision provided by an embodiment 1, the method comprising the following steps: S1: obtaining a raw gray image of the textile fabric surface at the current time.

[0028] It should be noted that, due to the high-speed motion of the production line and the rich details on the fabric surface, ordinary imaging methods are prone to produce trailing, and the color information of the fabric itself often interferes with the judgment of structural defects. Therefore, a clear image is first obtained through high-performance hardware, and then irrelevant color information is removed through preprocessing, and the core brightness and texture structure information is retained.

[0029] Specifically, the specific process of acquiring the high-resolution original gray scale image of the textile fabric surface at the current time includes: First, when the fabric moves at a constant speed through the detection station, control the industrial line array camera to cooperate with the high-brightness strip light source to continuously scan the fabric surface line by line to obtain the fabric color image.

[0030] It can be understood that the line array camera can provide extremely high lateral resolution, and cooperate with the high-brightness light source to suppress environmental light interference and shorten the exposure time, thereby obtaining a clear fabric color image.

[0031] Next, the collected fabric color image is grayed using a weighted average method, converted into an 8-bit gray scale image, and recorded as an original gray scale image. Figure 2 The original gray scale image is shown in the image, which contains shadow interference, several bright and obvious fluff, and an oil stain defect.

[0032] Figure 3 The traditional method defect detection image is shown in the image, which shows that the traditional method is interfered by fluff and shadow, resulting in a large number of interference areas in the detection result in addition to the defect.

[0033] S2: Based on the gray scale similarity distribution of the pixel points in the original gray scale image and the neighborhood pixels in the search radius region, the texture confidence index is calculated.

[0034] It should be noted that since the fabric texture has the characteristic of periodic repetition, and the defect does not have this characteristic, the similarity between the textures can be used to distinguish the fabric texture and the defect texture.

[0035] Preferably, as an example, based on the gray scale similarity distribution of the pixel points in the original gray scale image and the neighborhood pixels in the search range, the texture confidence index is calculated, specifically including: First, in the original gray scale image, any one pixel point is recorded as a target pixel point, and a Rectangular region is recorded as a texture unit region, and a Rectangular region is recorded as a search radius region.

[0036] Next, the gray scale difference between the texture unit region and the texture unit region at other positions in the search radius region is calculated, and the specific calculation method is:

[0037] Where x is the target pixel point, i is any pixel point in the search radius region obtained from the target pixel point, the gray difference distance between the texture unit region centered at the target pixel point x and the texture unit region centered at the pixel point i; the gray value of the u-th pixel point in the texture unit region centered at the target pixel point x, the gray value of the u-th pixel point in the texture unit region centered at the target pixel point i; the total number of pixel points in the texture unit region, and 255 is the maximum gray level, for normalizing the gray difference distance.

[0038] It can be understood that the numerator part reflects the difference degree of the two texture unit regions in structure and gray by calculating the sum of the absolute difference of the two texture unit regions pixel by pixel; the denominator part normalizes the difference to an interval, eliminating the influence of different light intensities.

[0039] Then, the similarity weight is calculated, and the specific calculation method is:

[0040] wherein, the similarity weight between the texture unit region centered at the target pixel point x and the texture unit region centered at the pixel point i; the similarity tolerance, which is exemplarily, 0.1, and exp() represents the exponential function with the natural constant as the base.

[0041] It can be understood that the difference distance is mapped to the weight by the negative exponential function, and the smaller the difference is, the closer the weight is to 1, which accurately describes the similarity degree between the texture unit regions.

[0042] Finally, the texture confidence index is calculated, and the specific calculation method is:

[0043] wherein, the texture confidence index of the target pixel point, the texture gain coefficient; the total number of pixel points in the search radius region except the target pixel point; a very small positive number to prevent the denominator from being zero; represents the maximum function.

[0044] It can be understood that since the fabric texture has the characteristic of periodic repetition, if the target pixel is a fabric texture, there will be a corresponding repeated texture in other regions, reflects the similarity of the texture unit region of the target pixel point with the most similar texture, and the greater the value, the greater the possibility of repeated texture in the texture unit region of the target pixel, and thus the greater the possibility of the target pixel being fabric; reflects the average similarity situation, and is used as a reference object for comparative analysis of ).

[0045] S3: Calculate a morphological confidence index based on the local brightness contrast feature of the pixel point in the original gray image and the isolated characteristic in the neighborhood.

[0046] It should be noted that there are some fluff on the surface of the textile fabric, and these fluff also do not have periodic repetition characteristics, so the fluff will be misjudged as defects through the above analysis, and thus in order to improve the accuracy of defect detection, the features of the fluff are needed to distinguish the defects and the fluff.

[0047] It should be further pointed out that the fluff has the physical characteristics of being bright and isolated, and thus the fluff interference can be removed based on this feature.

[0048] First, the bright confidence index is calculated, and the specific calculation method is:

[0049] wherein a 5x5 region centered on the target pixel is set as an analysis region, is the reference residual of the target pixel point, which is the difference between the gray value of the target pixel point and the median of the gray values of the analysis region; is the dynamic local contrast of the target pixel point, which is the standard deviation of the gray values of the analysis region; is the polarity gain coefficient, and the exemplary is 2.0, and tanh() represents the hyperbolic tangent function for normalization, the bright confidence index of the target pixel point, is a preset zero prevention parameter for preventing the denominator from being 0.

[0050] It can be understood that reflects the brightness of the target pixel point relative to the background, reflects the flatness of the background, and when is large, is small, indicating that the brightness of the target pixel point is relatively conspicuous relative to the flat background, and thus the target pixel point is more likely to be fluff.

[0051] Next, the isolated confidence index is calculated, and the specific calculation method is:

[0052]

[0053] wherein, is the abnormal saliency of the target pixel point, is the abnormal saliency of the i-th pixel point in the analysis region except the target pixel point, represents the number of pixel points in the analysis region except the target pixel point; is the isolation gain coefficient, exemplary, takes 5.0. is the isolation confidence index of the target pixel point, is the local neighborhood region centered on the target pixel point, is a preset zero prevention parameter, used to prevent the denominator from being 0.

[0054] It can be understood that, reflects the abnormal isolation of the target pixel point relative to its neighborhood pixels, and the smaller the value, the more isolated the abnormal target pixel point. Finally, the shape confidence index is calculated, and the specific calculation method is:

[0055] wherein, is the shape confidence index of the target pixel point.

[0056] It can be understood that the shape confidence index reflects the abnormal, bright, and isolated conditions of the target pixel point, and only when the target pixel point meets the three conditions of abnormality, brightness, and isolation at the same time, the probability of being a villus is greater.

[0057] S4: Based on the gray scale change rate and high-order energy mapping of the pixel points in the original gray scale image in the horizontal and vertical directions, a gradient sharpness index is calculated.

[0058] It should be noted that shadows also interfere with defect region recognition, and therefore in order to achieve accurate defect recognition, it is also necessary to exclude shadow interference on the basis of the above analysis.

[0059] It needs to be further explained that the edge of the shadow is blurred, and the edge of the defect is relatively sharp. Therefore, the defect and the shadow can be distinguished based on this feature.

[0060] The gradient sharpness index is calculated, and the specific calculation method is:

[0061] wherein, is the high-order energy coefficient, exemplary, takes 0.001; , , respectively are left, right, top and bottom adjacent pixel points of the target pixel point; is a cubic power operation; is a gray value of the pixel point, is a gradient sharpness index of the target pixel point.

[0062] It can be understood that the small gradient of the shadow is smaller after the cubic operation, and the large gradient of the defect is greatly increased after the cubic operation, thereby effectively narrowing the numerical gap between the two, and realizing shadow suppression.

[0063] S5: According to the texture confidence index, the shape confidence index and the gradient sharpness index, a defect residual index is calculated through a weighted suppression model, a defect residual image is generated, and the defect residual image is threshold segmented to identify the fabric defects.

[0064] Firstly, the defect residual index is calculated, and the specific calculation method is:

[0065] wherein, , , respectively are a texture suppression index, a shadow suppression index and a pile suppression index, and exemplary, , , respectively are 2, 1 and 3, is a gradient sharpness index of the target pixel point.

[0066] It can be understood that, reflects that the target pixel point does not belong to the fabric texture, reflects the gradient of the target pixel point, and the target pixel point does not belong to the shadow texture through the gradient, reflects that the target pixel point area does not belong to the pile, and only when the target pixel point does not belong to the fabric texture, the shadow texture and the pile, the greater the possibility of the target pixel point being a defect, is a defect residual index of the target pixel point.

[0067] An image composed of the defect residual indexes of all the pixel points is taken as a residual image; Figure 4 is a residual image, and it can be seen from the image that the scheme can accurately locate the oil stain connected domain.

[0068] Then, the OTSU algorithm is used to obtain a segmentation threshold of the residual image, the gray value of the pixel point greater than the segmentation threshold is set to 1, and the gray value of the pixel point less than the segmentation threshold is set to 0, to obtain a binary image.

[0069] The binary image is subjected to morphological closing operation and connected processing to obtain a plurality of connected domains; Geometric features of each connected domain are obtained, and the geometric features are exemplarily area, aspect ratio and compactness. The geometric features are subjected to logical matching with a preset defect physical feature library, and a defect category is output. For example, if the aspect ratio of the connected domain conforms to the broken warp index in the defect physical feature library, the connected domain is identified as a broken warp defect. If the compactness and area of the connected domain conform to the oil stain index in the defect physical feature library, the connected domain is identified as an oil stain defect. Figure 5 The final defect detection image is obtained, and it can be seen from the image that the defect area can be accurately identified.

[0070] Thus, the embodiment is completed.

[0071] The above merely provides the preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. within the principle of the present application shall be included in the protection scope of the present application.

Claims

1. A machine vision-based intelligent detection method for textile fabric surface defects, characterized in that, include: Obtain the original grayscale image of the textile fabric at the current moment; Based on the gray-level similarity distribution of a pixel and its neighboring pixels within the search radius, a texture confidence index is calculated. The texture confidence index is positively correlated with the maximum value of the similarity weight within the search radius and negatively correlated with the mean value of the similarity weight. The texture confidence index represents the probability of belonging to a normal periodic fabric texture. Based on the local brightness contrast features of pixels and their isolation characteristics in the neighborhood, a morphological confidence index is calculated, which represents the probability of belonging to the interfering fibers on the fabric surface. The gradient sharpness index is calculated based on the grayscale change rate of pixels in the horizontal and vertical directions and the higher-order energy mapping. The gradient sharpness index represents the probability of belonging to non-physical lighting shadows. Based on the texture confidence index, morphology confidence index, and gradient sharpness index, a defect residual index is calculated using a weighted suppression model to generate a defect residual map. The defect residual map is then subjected to threshold segmentation to identify fabric defects.

2. The intelligent detection method for textile surface defects based on machine vision according to claim 1, characterized in that, The acquisition of the original grayscale image of the textile fabric at the current moment specifically includes: The industrial line scan camera, in conjunction with a bar light source, scans the fabric moving at a constant speed line by line to acquire a color image of the fabric surface; The color image of the fabric surface is converted to grayscale using a weighted average method to obtain an original grayscale image with a bit depth of 8 bits.

3. The intelligent detection method for textile surface defects based on machine vision according to claim 1, characterized in that, Define any pixel as the target pixel, and define the texture unit area and search radius area with the target pixel as the center. Calculate the similarity weights between the target pixel and each pixel within the search radius; Within the search radius, find the maximum and average similarity weights of pixels other than the target pixel. The texture confidence index is calculated based on the maximum value and the mean value, and the calculation formula is as follows: in, A texture confidence index representing the target pixel. This is the texture gain coefficient; Search radius area The total number of pixels excluding the target pixel. The similarity weight between the texture unit region centered on the target pixel x and the texture unit region centered on pixel i; To prevent extremely small positive numbers with a denominator of zero; This represents the function that takes the maximum value.

4. The intelligent detection method for textile surface defects based on machine vision according to claim 3, characterized in that, The similarity weight is calculated as follows: Calculate the normalized grayscale difference distance between the target pixel and the texture unit region corresponding to each pixel within the search radius; The normalized grayscale difference distance is mapped to a similarity weight based on a negative exponential function, and the similarity weight is negatively correlated with the normalized grayscale difference distance.

5. The intelligent detection method for textile surface defects based on machine vision according to claim 1, characterized in that, The calculated morphological confidence index includes: Calculate the brightness confidence index of a pixel, wherein the brightness confidence index characterizes the brightness of a pixel relative to its neighboring background; Calculate the isolation confidence index of a pixel, wherein the isolation confidence index characterizes the degree of spatial isolation of a pixel within its neighborhood; The product of the brightness confidence index, the isolation confidence index, and the pixel anomaly significance index is used as the morphological confidence index.

6. The intelligent detection method for textile surface defects based on machine vision according to claim 5, characterized in that, The brightness confidence index for calculating pixels is specifically as follows: The difference between the gray value of a pixel and the median of its neighborhood is calculated as the baseline residual. Obtain dynamic local contrast within the neighborhood; The ratio of the baseline residual to the dynamic local contrast is mapped using a hyperbolic tangent function to obtain a brightness confidence index, which is positively correlated with the baseline residual.

7. The intelligent detection method for textile surface defects based on machine vision according to claim 5, characterized in that, The calculation of the isolated confidence index of a pixel is specifically as follows: The anomaly significance index of a pixel is calculated based on the texture confidence index, and the anomaly significance index is negatively correlated with the texture confidence index. Calculate the mean of the anomaly significance index of other pixels in the neighborhood; The mean is mapped to an isolated confidence index based on a negative exponential function, and the isolated confidence index is negatively correlated with the mean.

8. The intelligent detection method for textile surface defects based on machine vision according to claim 1, characterized in that, The gradient sharpness index is calculated as follows: Calculate the grayscale difference between adjacent pixels in the horizontal direction and the grayscale difference between adjacent pixels in the vertical direction. The absolute value of the grayscale difference is raised to the power of three to suppress low-amplitude shadow edge gradients and enhance high-amplitude defect edge gradients. The gradient sharpness index is obtained by mapping the result of the cube power operation to the hyperbolic tangent function.

9. The intelligent detection method for textile surface defects based on machine vision according to claim 1, characterized in that, The formula for calculating the defect residual index using the weighted suppression model is as follows: in, The defect residual index represents the target pixel. The texture confidence index for the target pixel; The gradient sharpness index for the target pixel; A morphological confidence index for target pixels; , , These are the texture suppression index, shadow suppression index, and fuzz suppression index, respectively.

10. The intelligent detection method for textile surface defects based on machine vision according to claim 1, characterized in that, The step of thresholding the defect residual map to identify fabric defects includes: The defect residual map is binarized using the OTSU adaptive thresholding algorithm to separate high-resolution regions from the background. Apply morphological closing operations to fill the voids inside connected components and connect broken edges; Extract the geometric features of the connected components, match the geometric features with a preset defect physical feature library, and output the defect identification results.

Images