Industrial part defect detection method

Abstract

The invention provides an industrial part defect detection method, which comprises the following steps: acquiring an input industrial part burr defect image, extracting the shape, size and distribution position characteristics of burrs in the image to obtain a burr characteristic pattern, segmenting a burr defect region according to the edge roughness and color difference in the burr characteristic pattern, and generating a binary burr defect mask; performing morphological operation on the binarized burr defect mask, eliminating isolated noisy points through expansion and corrosion, connecting adjacent burr areas, segmenting the processed burr mask, and extracting independent burr defect areas according to the three-dimensional contour and surface texture features of burrs; performing image enhancement processing on the preliminary verification code character matrix, adjusting the color contrast of characters to highlight the thickness of character strokes, and generating a verification code image; and carrying out compression and format conversion on the generated verification code image to obtain a final industrial part burr defect verification code which is used for defect detection of workpiece parts.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to an industrial parts defect detection method. Background Art

[0002] In the field of burr defect detection of industrial parts, there is a technical problem that needs to be solved urgently. At present, burr defect images are highly complex and diverse, which poses a huge challenge to accurate identification. In order to improve the accuracy and reliability of detection, researchers have proposed a new method based on verification code recognition. However, how to effectively convert complex and changeable burr defect images into verification code form while retaining key feature information has become a thorny problem. There are significant differences in the size, shape and distribution of these defects. In this case, how to extract these complex image features and convert them into concise and clear verification code form is a huge challenge. At the same time, it is also necessary to ensure that key defect information is not lost during the conversion process. This requires the algorithm to simplify the image while retaining enough detail information to support subsequent accurate identification and classification. Another difficulty is how to ensure that the converted verification code has sufficient recognition and anti-interference ability. It is necessary to enhance the prominence of key features while suppressing the interference of irrelevant information during the verification code generation process. In short, how to efficiently convert complex and changeable burr defect images into concise and clear verification code forms while retaining key features, while ensuring that the converted verification codes have good recognition and anti-interference capabilities, is the core challenge facing this technology. Summary of the invention

[0003] The present invention provides an industrial parts defect detection method, which mainly includes:

[0004] Obtain an input burr defect image of an industrial part, extract the shape, size, and distribution position features of the burrs in the image, obtain a burr feature map, segment the burr defect area according to the edge roughness and color difference in the burr feature map, and generate a binary burr defect mask;

[0005] Perform morphological operations on the binary burr defect mask, eliminate isolated noise points through dilation and corrosion, connect adjacent burr areas, segment the processed burr mask, and extract independent burr defect areas based on the 3D contour and surface texture features of the burr;

[0006] For each independent burr defect area, its geometric features are calculated, a burr defect feature vector is constructed, and the semantic features of the burr defect feature vector are extracted to obtain the feature representation of the burr defect. Different types of burr defects are grouped based on the feature similarity of the burr defects, and the shape and size features of the most representative samples in each group of burr defects are selected as the representative features of the burr defects in this group.

[0007] Establishing a mapping relationship between each group of burr defect representatives and verification code characters, including mapping the burr shape and size attributes to the shape and size characteristics of the character, dividing the original burr defect image into a number of grids, and generating a preliminary verification code character matrix according to the grid position and the verification code character mapping results;

[0008] Perform image enhancement processing on the preliminary verification code character matrix, adjust the color contrast of the characters to highlight the thickness of the character strokes, and generate a verification code image;

[0009] The generated verification code image is compressed and format converted to obtain the final industrial parts burr defect verification code for defect detection of workpiece parts.

[0010] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0011] The invention discloses a method for detecting defects in industrial parts. The method extracts features such as the shape, size, and distribution position of the burrs, segments a burr defect area, and generates a binary mask. After morphological processing and segmentation, the geometric features and semantic features of the independent burr defect area are extracted to construct a feature vector. The burr defects are grouped based on feature similarity, and representative samples are selected to establish a mapping relationship with verification code characters. The original image is divided into grids, and a preliminary character matrix is ​​generated according to the mapping result, and the final verification code is obtained after image enhancement processing. The present invention realizes the conversion of the burr defect features of industrial parts into verification codes, which can be used for defect detection and improve the security of the verification code, and has important practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 The present invention is a flow chart of an industrial parts defect detection method.

[0013] Figure 2 It is a schematic diagram of an industrial parts defect detection method of the present invention.

[0014] Figure 3 It is another schematic diagram of an industrial parts defect detection method of the present invention. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0016] like Figure 1 -3. In this embodiment, a method for detecting defects in industrial parts may specifically include:

[0017] Step S101, obtain an input industrial part burr defect image, extract the shape, size, and distribution position features of the burrs in the image, and obtain a burr feature map. According to the edge roughness and color difference in the burr feature map, segment the burr defect area and generate a binary burr defect mask.

[0018] The acquired industrial parts burr defect image is preprocessed by linear grayscale transformation and adaptive histogram equalization to obtain a preprocessed image with enhanced contrast and prominent burr features; the shape, size and distribution position of the burr are extracted according to the preprocessed image to generate a burr feature map containing burr contour, area and coordinates; the Canny operator is used to perform edge detection on the burr feature map to obtain edge information in the burr feature map and calculate the gradient amplitude and direction of edge pixel points; if the edge information meets the high and low thresholds automatically determined by the OTSU algorithm, the edge is binarized to obtain a preliminary burr contour; the preliminary burr contour is segmented by a regional growing algorithm based on edge and grayscale gradient, starting with the seed point obtained by the edge detection, and gradually expanding the segmented area; if the grayscale value difference and gradient direction difference of adjacent pixel points are both less than the set threshold, they are included in the same area until they cannot be further expanded; the segmented area is evaluated and optimized, and the rationality of the segmented area is evaluated by shape features and texture features; if the segmented area is unreasonable, it is merged or split.

[0019] Exemplarily, the acquired burr defect image of industrial parts is preprocessed by linear grayscale transformation and adaptive histogram equalization to enhance the image contrast and highlight the burr features. The shape, size, and distribution position of the burr are extracted from the preprocessed image to generate a burr feature map containing the burr contour, area, and coordinates. The Canny operator is used to detect the edge of the burr feature map, obtain the edge information in the image, calculate the gradient amplitude and direction of the edge pixel points, and binarize the edge according to the high and low thresholds automatically determined by the OTSU algorithm to obtain a preliminary burr contour. The preliminary burr contour is segmented by using the regional growing algorithm based on the edge and grayscale gradient. Starting from the seed point obtained by edge detection, the segmented area is gradually expanded. When the grayscale value difference and gradient direction difference of adjacent pixels are less than the set threshold, they are included in the same area until they can no longer be expanded. The segmentation results are evaluated and optimized, and the rationality of the segmented area is evaluated using shape features and texture features, and unreasonable areas are merged or split. The optimized area is subjected to morphological operation. A 3x3 rectangular structural element is used to perform an opening operation once to remove small noise, and a closing operation is performed twice to fill the internal holes to obtain a complete burr defect area. The defect area is marked as the foreground and the non-defect area is marked as the background to generate a binary burr defect mask image. After the acquired industrial parts burr defect image is subjected to linear grayscale transformation, the pixel value is mapped from 0-255 to the 0-1 interval to enhance the image contrast. Then, adaptive histogram equalization is applied to divide the image into 8x8 local blocks, and histogram equalization is performed in each block to highlight local details. The preprocessed image is subjected to morphological gradient operation to extract the burr contour, and the 3x3 structural element is used to perform dilation and corrosion operations to obtain the burr edge. The burr area and centroid coordinates are calculated to generate a burr feature map containing contours, areas, and coordinates. Canny edge detection was applied to the feature map to calculate the image gradient. The x-direction gradient Gx and the y-direction gradient Gy were obtained by the Sobel operator. The edge strength M = sqrt(Gx^2+Gy^2) and the edge direction θ = arctan(Gy / Gx) were obtained. The OTSU algorithm was used to automatically determine the high and low thresholds. The high threshold was 128 and the low threshold was 0.4 times the high threshold, i.e., 51.2. The edges were binarized. The region growing algorithm based on edge and grayscale gradient took edge pixels as seed points. When the grayscale value difference between adjacent pixels was less than 20 and the gradient direction difference was less than 30 degrees, they were included in the same region. The segmentation results were evaluated and the perimeter / area ratio of each region and the energy characteristics of the grayscale co-occurrence matrix were calculated. If the ratio was greater than 0.1 and the energy characteristic was less than 0.5, the segmentation was considered reasonable. Otherwise, the regions were merged or split.Finally, morphological operations are performed on the optimized area, using a 3x3 rectangular structure element to perform one opening operation and two closing operations to obtain a complete burr defect area. The pixel value of the defect area is set to 255, and the pixel value of the non-defect area is set to 0 to generate a binary burr defect mask image.

[0020] Step S102, performing morphological operations on the binary burr defect mask, eliminating isolated noise points by dilation and corrosion, and connecting adjacent burr areas; segmenting the processed burr mask, and extracting independent burr defect areas according to the three-dimensional contour and surface texture features of the burr.

[0021] The binary burr defect mask is subjected to a dilation operation, and the structuring element is used for iteration to expand the boundary of the burr area, and then an erosion operation is performed to remove isolated noise points to obtain an optimized burr mask image; the contour information of the optimized burr mask is extracted using the Sobel operator, and the curvature change and direction consistency are calculated according to the contour information to determine the connectivity of adjacent burr areas; the two-dimensional burr mask image is registered with the three-dimensional point cloud data collected by the depth camera using a rigid body transformation algorithm, and the two-dimensional image coordinate system is converted to a three-dimensional point cloud coordinate system; the point cloud data is subjected to bilateral filtering for denoising, and the surface is reconstructed and smoothed using the moving least squares method, and the depth value and normal vector information of each pixel are extracted. Combined with the grayscale value of the two-dimensional image, a feature vector containing the depth value, the component of the normal vector and the grayscale value is constructed; the feature vector is segmented using a regional growing algorithm, and the edge points of the burr area are selected as seed points, and the area is gradually expanded. When the Euclidean distance of the feature vectors of adjacent points is less than a preset threshold, they are included in the same area until they cannot be further expanded to obtain an independent burr defect area.

[0022] Exemplarily, a dilation operation is applied to the binary burr defect mask, and a 3x3 structure element is used for three iterations to expand the burr area boundary and fill small holes and gaps. Then an erosion operation is performed, and the same structure element is used for two iterations to remove isolated noise points and retain the main burr structure to obtain an optimized burr mask image. The Sobel operator is used to extract the contour information of the optimized burr mask, and the contour curvature and direction are calculated. The connectivity of adjacent burr areas is determined based on the curvature change and direction consistency. If the curvature change is less than 0.1 and the direction angle is less than 15 degrees, the adjacent areas are merged into a complete burr defect area. The rigid body transformation algorithm is used to align the two-dimensional burr mask image with the three-dimensional point cloud data collected by the depth camera, and the two-dimensional image coordinate system is converted to the three-dimensional point cloud coordinate system. The point cloud data is subjected to bilateral filtering for noise reduction, and then the moving least squares method is used for surface reconstruction and smoothing. The depth value and normal vector information of each pixel are extracted. Combined with the grayscale value of the two-dimensional image, a 6-dimensional feature vector containing the depth value, the x, y, z components of the normal vector and the grayscale value is constructed. The constructed feature vector is segmented using the region growing algorithm. The edge points of the burr area are selected as seed points, and the area is gradually expanded. When the Euclidean distance of the feature vectors of adjacent points is less than 0.5, they are included in the same area until they can no longer be expanded, and finally an independent burr defect area is obtained. For the binary burr defect mask, the 3x3 structure element is first applied for three dilation operations to expand the area with the original pixel value of 1 outward by 3 pixels and fill the holes with a diameter of less than 6 pixels. Then the same structure element is used for two erosion operations to remove protrusions and isolated noise points less than 4 pixels at the edge. The optimized mask image is edge detected by the Sobel operator, and the gradients in the x and y directions are calculated to obtain the edge strength and direction. The curvature is calculated along the edge, and a sample is taken every 10 pixels. The curvature is calculated by fitting a circle with three adjacent points. If the curvature change of adjacent areas is less than 0.1 and the direction angle is less than 15 degrees, these areas are merged. The iterative nearest point algorithm is used to align the two-dimensional mask image with the three-dimensional point cloud data collected by the depth camera. The maximum number of iterations is set to 50 and the convergence threshold is set to 0.001mm. The aligned point cloud data is denoised using a bilateral filter with a radius of 5mm, the standard deviation of the spatial domain is set to 2.5mm, and the standard deviation of the value domain is set to 10mm. Then the moving least squares method is applied to reconstruct the surface, and 20 nearest neighbor points are selected. The local surface is fitted using a quadratic polynomial. For the reconstructed surface, the depth value and normal vector of each point are extracted, and the 6-dimensional feature vector is constructed by combining the grayscale value of the original two-dimensional image. Finally, the region growing algorithm is used for segmentation, and the edge points are selected as seed points. When the Euclidean distance of the 6-dimensional feature vectors of adjacent points is less than 0.5, they are included in the same region. Continue to expand until no adjacent points that meet the conditions can be found, and finally an independent burr defect area is obtained.

[0023] Step S103, for each independent burr defect area, calculate its geometric features and construct a burr defect feature vector; extract the semantic features of the burr defect feature vector to obtain the feature representation of the burr defect, group different types of burr defects based on the feature similarity of the burr defects, and select the shape and size features of the most representative samples in each group of burr defects as the representative features of the group of burr defects.

[0024] The geometric features and texture features of the burr defect area are obtained to construct a multi-dimensional burr defect feature vector; wherein the geometric features include area, perimeter, circularity, rectangularity, aspect ratio, and direction angle. The multi-dimensional burr defect feature vector is reduced in dimension by an autoencoder with three layers of encoders and three layers of decoders; if the number of nodes in the middle layer is 1 / 4 of the dimension of the original feature vector, a low-dimensional burr defect feature representation is obtained through compression and reconstruction. The feature vector output by the autoencoder is Z-score standardized, and the cosine similarity between different burr defect samples is calculated. A similarity matrix is ​​constructed based on the cosine similarity, and the similarity matrix is ​​analyzed using a density-based clustering algorithm; if the neighborhood radius is 0.5 and the minimum number of samples is 4, burr defects with similar features are automatically grouped. For each group of burr defects, the average feature vector and variance of the samples in the group are calculated; if the Euclidean distance weight between the sample and the average feature vector in the group is 0.7 and the variance weight of each dimension of the sample feature is 0.3, the sample with the smallest weighted sum is selected as the representative sample of the group.

[0025] Exemplarily, for each independent burr defect area, its area, perimeter, circularity = 4π * area / perimeter ^2, rectangularity = area / minimum circumscribed rectangle area, aspect ratio, and direction angle are calculated, and the grayscale distribution and texture features of the defect area are combined to construct a multi-dimensional burr defect feature vector. The burr defect feature vector is reduced in dimension by using an autoencoder with three layers of encoders and three layers of decoders. The number of nodes in the middle layer is set to 1 / 4 of the dimension of the original feature vector. The mean square error is used as the loss function, and the Adam optimizer is used for training to extract the key semantic features. The low-dimensional burr defect feature representation is obtained through compression and reconstruction, and the main information of the original feature is retained. The feature vector output by the autoencoder is Z-score standardized so that the features of each dimension have the same scale. Based on the standardized burr defect feature representation, the cosine similarity between different burr defect samples is calculated, and a similarity matrix is ​​constructed. The similarity matrix is ​​analyzed using a density-based clustering algorithm, and the neighborhood radius is set to 0.5 and the minimum number of samples is 4. Burr defects with similar features are automatically grouped. For each group of burr defects, the average feature vector and variance of the samples in the group are calculated, the Euclidean distance between each sample and the average feature vector in the group is calculated, and the variance of each dimension of the sample feature is calculated. The two indicators are weighted and summed, with weights of 0.7 and 0.3 respectively. The sample with the smallest weighted sum is selected as the representative sample of the group, and the shape contour and size parameters of the sample are extracted as the representative features of the burr defects in the group. The geometric features of each independent burr defect area are calculated. First, the area is obtained by pixel counting, and the perimeter is obtained by the boundary tracking algorithm. The circularity is calculated by 4π area / perimeter ^2, and the value between 0.8 and 0.95 is obtained. The rectangularity is calculated by dividing the area by the area of ​​the minimum circumscribed rectangle, which is usually in the range of 0.6-0.9. The aspect ratio is obtained by dividing the length in the main axis direction by the width in the vertical direction, and the value range is between 1.2 and 3. The direction angle is determined by the angle between the main axis and the horizontal line, ranging from 0 to 180 degrees. Grayscale distribution features are described by mean, standard deviation, skewness and kurtosis, and texture features are represented by energy, contrast and correlation of grayscale co-occurrence matrix. The dimension of the constructed feature vector is 20. The autoencoder structure is 20-10-5-10-20, using ReLU activation function, learning rate is set to 0.001, batch size is 32, and training is 1000 rounds. After dimensionality reduction, a 5-dimensional feature vector is obtained. Z-score normalization subtracts the mean from each feature and divides it by the standard deviation, so that the mean of all features is 0 and the variance is 1. The cosine similarity between the standardized feature vectors is calculated to obtain the similarity matrix. The DBSCAN algorithm is used with neighborhood radius ε=0.5 and minimum number of samples MinPts=4 to cluster similar samples into one category. The average feature vector μ and covariance matrix Σ are calculated for each group, and the Mahalanobis distance d=(x-μ)^TΣ^-1(x-μ) from each sample x to μ is calculated, and the variance v of each dimension of x is calculated at the same time.Let s = 0.7d + 0.3v, select the sample with the smallest s as the representative, and extract its contour coordinate sequence and length, width, and area size parameters as the representative features of the group.

[0026] Step S104, establishing a mapping relationship between each group of burr defect representatives and verification code characters, including mapping the burr shape and size attributes to the shape and size characteristics of the character; dividing the original burr defect image into a number of grids, and generating a preliminary verification code character matrix according to the grid positions and the verification code character mapping results.

[0027] The representative features of burr defects and the features of verification code characters are obtained, and a database of corresponding relationships between the representative features of burr defects and the features of verification code characters is established; the Otsu algorithm is applied to the original burr defect image according to the corresponding relationship database, and adaptive threshold segmentation is performed to extract independent burr areas in the burr defect image; for the independent burr areas, the corresponding relationship database is queried to obtain characters matching the independent burr areas, and the characters are filled into preset grid positions to obtain a preliminary character matrix; the binary grammar model is used to calculate the probability of adjacent character combinations in the preliminary character matrix, and it is determined whether the probability is lower than a preset threshold; if the probability is lower than the preset threshold, the matching degree of the burr feature at the position and other candidate characters is recalculated, and the suboptimal matching result is selected to replace the corresponding characters in the preliminary character matrix to obtain an optimized character matrix.

[0028] Exemplarily, a database of correspondence between representative features of burr defects and features of verification code characters is constructed, and the similarity between burr contours and standard character contours is calculated by contour matching method based on Fourier descriptor, and the area ratio of burrs to characters is compared at the same time. The threshold range is set to 0.8-1.2, and a preliminary mapping relationship between burrs and characters is established. The original burr defect image is segmented by adaptive threshold using Otsu algorithm, and each independent burr area is extracted. The centroid coordinates and main direction of each burr area are calculated. According to these features, the image is divided into M×N grids, and the size of each grid is equivalent to the average burr size. The grid boundary is determined according to the distribution of the centroid coordinates of the burr. Each grid is traversed. If there is a burr defect in the grid, the mapping database is queried according to the representative features of the burr, the corresponding character is obtained, and the character identifier is filled into the corresponding grid position. For large burrs spanning multiple grids, the grid with the largest coverage area is selected to fill in the character. If there are multiple burrs in a grid, the burr with the largest area is selected as the representative of the grid to form a preliminary character matrix. The character matrix is ​​checked for local consistency. The probability of adjacent character combinations is calculated using a binary grammar model. The threshold is set to 0.01. If the probability is lower than the threshold, it is considered an unreasonable combination. The matching degree between the burr feature at that position and other candidate characters is recalculated, and the suboptimal matching result is selected for replacement to optimize the overall consistency of the character matrix. When constructing the mapping relationship between the burr defect and the verification code character, the 64-point Fourier descriptor of the burr contour is first extracted, and the Euclidean distance is calculated with the pre-stored standard character contour descriptor. A distance less than 0.2 is considered a match. At the same time, the ratio of the burr area to the standard character area is calculated. If it is within the range of 0.8-1.2, the match is further confirmed. The Otsu algorithm is applied to the original image, and the optimal threshold is selected for binarization to obtain the burr area. The centroid coordinates and main direction of each burr are calculated, and the 800×600 pixel image is divided into 10×8 grids based on this information. Traverse the grid. If there is a burr in a grid, its area, perimeter ratio, and direction features are extracted, and the corresponding character is queried in the mapping database. For glitches that span multiple grids, that is, those with an area exceeding 150% of the average, characters are filled in the grid with the largest coverage area. If there are multiple glitches in a grid, the one with the largest area is selected as the representative. Finally, a 10×8 character matrix is ​​obtained. A binary grammar model based on 100,000 common words is used to calculate the probability of adjacent character pairs. If it is less than 0.01, it is considered an unreasonable combination. At this time, the similarity between the glitches at that position and the suboptimal matching characters is recalculated. If the similarity difference is less than 20%, it is replaced with the suboptimal character to improve the overall consistency.

[0029] According to the representative characteristics of the burr defects in the grid, the corresponding characters are found in the verification code character set, the characters corresponding to the grid are arranged according to the grid positions, a character matrix is ​​generated, the edge contour of the character matrix is ​​analyzed, the missing parts are identified and completed, and the completed character matrix is ​​obtained.

[0030] A feature vector is constructed according to the area, perimeter ratio and direction angle of the burr defects in the grid, and the character with the highest matching degree is searched in the pre-established verification code character feature library through the support vector machine classifier to obtain the initial character matrix; the Canny edge detection algorithm is applied to the initial character matrix to calculate the image gradient amplitude and direction, and the edge contour of the character matrix is ​​obtained through the double threshold method and edge connection, and the area with discontinuous or missing edges is identified; for the identified missing area, the morphological reconstruction algorithm is used to preliminarily fill the missing part in combination with the shape features and semantic information of adjacent characters, and a structure close to the character stroke width is selected. elements, reconstructing the character contour by iterative expansion and intersection operations to obtain a repaired character contour; using a convolutional neural network with a U-Net structure to perform fine processing on the repaired character contour, the network input is the repaired character image, and the output is the complete character image, by learning the local structural features of the characters, generating character details that conform to the context, and obtaining a completed character matrix; applying an optical character recognition algorithm to the completed character matrix for verification, calculating the recognition confidence, and if the recognition confidence is lower than a preset threshold, fine-tuning the convolutional neural network until a high-confidence completed character matrix is ​​obtained.

[0031] Exemplarily, a feature vector is constructed based on the area, perimeter ratio, and direction angle of the burr defects in the grid, and the support vector machine classifier is used to find the character with the highest matching degree in the pre-established verification code character feature library. The feature library expands the samples by multi-angle rotation, scaling, and noise addition, and arranges the matching results according to the grid position to generate an initial character matrix. The Canny edge detection algorithm is applied to the initial character matrix to calculate the image gradient amplitude and direction. The edge contour of the character matrix is ​​obtained by the double threshold method and edge connection, and the area with discontinuous or missing edges is identified. For the identified missing area, the morphological reconstruction algorithm is used to preliminarily fill the missing part in combination with the shape features and semantic information of adjacent characters, and the structural elements close to the character stroke width are selected. The character contour is reconstructed by iterative expansion and intersection operations to obtain the repaired character contour. Based on the repaired character contour, a convolutional neural network with a U-Net structure is used to refine the completion area. The network input is the repaired character image and the output is the complete character image. The pixel-level cross entropy loss function and the Adam optimizer are used for training. By learning the local structural features of the characters, the character details that conform to the context are generated, and finally the completed character matrix is ​​obtained. The optical character recognition algorithm is applied to the completion result for verification, and the recognition confidence is calculated. If it is lower than the preset threshold, the convolutional neural network step is returned for fine-tuning until a high-confidence completed character matrix is ​​obtained. When processing burr defects in the grid, the feature vector is first extracted, including area, perimeter ratio, and direction angle, and a 10-15 dimensional vector is usually obtained. The verification code character feature library contains 3000 samples of common fonts, and each sample is expanded to 9000 by ±15° rotation, 0.8-1.2 times scaling, and adding 5% Gaussian noise. The support vector machine with RBF kernel, C=1.0, gamma='scale', is used to match characters in this library. Canny edge detection uses a Gaussian kernel size of 5x5, a low threshold of 50, and a high threshold of 150. Morphological reconstruction uses a 3x3 rectangular structural element, and the number of iterations is usually 5-10. The U-Net network contains 4 downsampling and upsampling, with a starting number of 64 convolution kernels, which doubles at each layer, with a maximum of 512. The training set contains 100,000 pairs of defect-complete character image pairs, a batch size of 32, a learning rate of 0.001, and 100 training rounds. Pixel-level cross entropy loss is used with the Adam optimizer. After completion, the Tesseract OCR engine is used for verification, and the confidence threshold is set to 0.8. If it is lower than the threshold, U-Net is fine-tuned for 5 rounds at a learning rate of 0.0001. The entire process from feature extraction to final verification usually takes 0.5-1 seconds per character and is processed on a quad-core CPU.

[0032] The spacing and size relationship between characters in the completed character matrix are analyzed, and the overlapping characters that exceed the overlap requirement are segmented to obtain optimized characters; the feature image of each character is spliced ​​according to the matrix position to obtain a preliminary verification code character matrix.

[0033] The connected domain analysis algorithm is applied to the completed character matrix, and the independent character area is identified by judging the 8-neighborhood connectivity to obtain the position and size information of the character. The ratio of the overlapping area of ​​the adjacent character boundary boxes to the character area is calculated. If the ratio exceeds the preset threshold, the overlapping characters are separated by a segmentation method based on contour analysis. The separated character image is morphologically processed, and the rectangular structure element is used for opening operation to remove noise points and the closed operation to fill the internal holes to obtain a character image with a complete shape. The spacing and size relationship between characters are analyzed, and the Euclidean distance of the center points of adjacent characters and the character aspect ratio are calculated. If the spacing is less than the preset ratio of the average character width or the aspect ratio difference exceeds the preset threshold, the character is scaled and the position is adjusted. According to the adjusted character matrix position information, the optimized character feature image is spliced ​​in the original arrangement order, and the Gaussian kernel is applied to the transition area between the characters for fuzzy processing to obtain a preliminary verification code character matrix.

[0034] Exemplarily, the connected domain analysis algorithm is applied to the completed character matrix, and each independent character area is identified through 8-neighborhood connectivity judgment, and the position and size information of each character is obtained, and the bounding box coordinates of the character are calculated at the same time. The ratio of the overlapping area of ​​the bounding boxes of adjacent characters to the character area is calculated. If the ratio exceeds the preset threshold value of 0.2, the overlapping characters are separated by a segmentation method based on contour analysis. The character contours are extracted, the concave points are found, and the concave points are segmented along the concave point connection line to obtain the separated character image. The segmented character image is morphologically processed, and a 3x3 rectangular structural element is used to perform an open operation once to remove small noise points, and two closed operations are performed to fill the internal holes to ensure the integrity and consistency of the character shape, while preserving the bounding box information of each character. The spacing and size relationship between characters are analyzed, and the Euclidean distance of the center points of adjacent characters and the character aspect ratio are calculated. If the spacing is less than 1 / 4 of the average character width or the aspect ratio difference exceeds 20%, the character is scaled and the position is adjusted. According to the adjusted character matrix position information, the optimized character feature images are spliced ​​in the original arrangement order, and the 3x3 Gaussian kernel is used to perform blur processing in the transition area between characters to smooth the character edges and obtain a preliminary verification code character matrix. When processing the completed character matrix, the 8-neighborhood connected domain analysis algorithm is first applied, and the pixel brightness threshold is set to 128 to identify each independent character area. For a 100x30 pixel verification code image, 4-6 connected areas are usually obtained. The bounding box of each area is calculated, and the upper left corner coordinates (x, y) and width and height (w, h) are recorded. Then the overlap of the bounding boxes of adjacent characters is calculated. If the ratio of the overlapping area to the character area exceeds 0.2, segmentation is triggered. During the segmentation process, the character contour is extracted, the Douglas-Peucker algorithm is used to simplify the contour, and the concave points are detected. The characters are segmented along the concave point connection line to obtain the separated sub-images. Each character image is morphologically processed, and a 3x3 rectangular structure element is used to first remove noise points with an area less than 9 pixels, and then fill the holes with an area less than 9 pixels. Analyze the character spacing and calculate the distance between the center points of adjacent characters. If it is less than 1 / 4 of the average character width, adjust the position. Calculate the character aspect ratio. If the difference in the aspect ratio of adjacent characters exceeds 20%, scale the abnormal characters. Finally, splice the character images in the original order and apply a 3x3 Gaussian kernel to blur the character intervals to obtain the final verification code character matrix, which is usually 100x30 pixels in size and contains 4-6 characters.

[0035] Step S105 , performing image enhancement processing on the preliminary verification code character matrix, adjusting the color contrast of the characters to highlight the thickness of the character strokes, and generating a verification code image.

[0036] A contrast-limited adaptive histogram equalization process is applied to the verification code character matrix, wherein the process divides the image into local areas, performs histogram equalization on each area, and combines the processing results through an interpolation method; based on the equalization processing results, an Otsu algorithm is used to perform adaptive threshold segmentation, an optimal threshold is determined, the character area is separated from the background, and a binary character image is obtained; a morphological operation is performed on the binary character image, a skeleton extraction algorithm based on distance transformation is used to retain the main structure of the character, and a structural element is used to perform an expansion operation; the processed character image is fused with a background layer generated by a Perlin noise algorithm, and the degree of fusion between the character and the background is controlled by an alpha blending algorithm; Gaussian noise is added to the fused image, the mean of the Gaussian noise is a preset value, the standard deviation is a preset value, and a bilinear interpolation method is used to randomly distort the local area, and the distortion amplitude is controlled within a preset range to generate a final verification code image.

[0037] Exemplarily, the contrast-limited adaptive histogram equalization processing is applied to the preliminary verification code character matrix, the image is divided into 8x8 local areas, each area is subjected to histogram equalization, the contrast limit threshold is set to 0.01, and the processing results are merged by the interpolation method to enhance the local contrast of the image and highlight the contour features of the characters. The Otsu algorithm is used for adaptive threshold segmentation, and the optimal threshold is determined according to the principle of maximizing the inter-class variance of the image grayscale histogram, the character area is separated from the background, and a binary character image is obtained, while retaining the edge information map. The binary character image is subjected to morphological operations, and the skeleton extraction algorithm based on distance transformation is used to retain the main structure of the character, and then a 3x3 cross-shaped structural element is used for expansion operation. The number of iterations is adjusted according to the required stroke thickness to achieve thickening or thinning of the character strokes. The processed character image is fused with the background layer generated by the Perlin noise algorithm, and the degree of fusion between the character and the background is controlled by the alpha blending algorithm. The alpha value is set between 0.6 and 0.8 to adjust the color and transparency of the character. Finally, Gaussian noise is added with a mean of 0 and a standard deviation of 10, and the local area is randomly distorted using the bilinear interpolation method. The distortion amplitude is controlled within the range of 1-3 pixels to generate the final verification code image. When processing the verification code character matrix of 100x30 pixels, contrast-limited adaptive histogram equalization is first applied to divide the image into 5x2 local areas, each of which is 20x15 pixels in size. The histogram is calculated for each area, the contrast threshold is limited to 0.01, and the pixel values ​​exceeding the threshold are redistributed to other bins. The processing results are merged using bilinear interpolation to obtain the enhanced image. Subsequently, the Otsu algorithm is applied for adaptive threshold segmentation, the image grayscale histogram is calculated, the threshold is traversed from 0 to 255, and the value that maximizes the inter-class variance is selected as the optimal threshold, which is usually between 100 and 150. Pixels below the threshold are set to 0, and those above the threshold are set to 255 to obtain a binary image. Distance transform is applied to the binary image, the distance from each foreground pixel to the nearest background pixel is calculated, and the local maximum point is taken as the skeleton. Use a 3x3 cross-shaped structural element [[(0, 1, 0), (1, 1, 1), (0, 1, 0)]] to perform 2-3 dilation operations to thicken the character strokes. When generating the background, use the Perlin noise algorithm, set an 8x8 grid, and generate smooth noise through cosine interpolation. When the character image is merged with the background, set the alpha value to 0.7, that is, I_result = 0.7I_char + 0.3I_background. Add Gaussian noise and use Box-Muller transform to generate random numbers with a mean of 0 and a standard deviation of 10. Finally, apply bilinear interpolation to 20% of the randomly selected areas for local distortion, with a distortion amplitude between 1 and 3 pixels to obtain the final verification code image.

[0038] Step S106, compressing and formatting the generated verification code image to obtain a final industrial part burr defect verification code for use in defect detection of workpiece parts.

[0039] The verification code image is subjected to grayscale conversion, wherein the grayscale conversion converts the RGB value into a single-channel grayscale value by a weighted average method; the grayscale image is subjected to frequency domain analysis according to the single-channel grayscale value, wherein the frequency domain analysis includes dividing the grayscale image into 8x8 small blocks, performing a two-dimensional discrete cosine transform on the small blocks, and obtaining a frequency domain coefficient matrix; a zigzag scan is performed on the frequency domain coefficient matrix, wherein the zigzag scan converts the frequency domain coefficient matrix into a one-dimensional sequence; the one-dimensional sequence is compressed by arithmetic coding, wherein the compression encodes symbols by a probability model to obtain a compressed bit stream; the compressed bit stream is encapsulated into a JPEG format, wherein the encapsulation includes adding file header information, wherein the file header information includes an SOI tag, an APP0 segment, a DQT segment, a SOF0 segment, a DHT segment, and an SOS segment; wherein the DQT segment stores a quantization table, the SOF0 segment records an image size and a sampling factor, and the DHT segment contains a Huffman table.

[0040] Exemplarily, the generated verification code image is gray-scale converted, and the RGB value is converted into a single-channel gray-scale value using a weighted average method to reduce the amount of image data. The gray-scale image is analyzed in the frequency domain by applying discrete cosine transform, the image is divided into 8x8 small blocks, and each small block is subjected to a two-dimensional DCT transform to obtain a frequency domain coefficient matrix, and the high-frequency coefficients are quantized by the JPEG standard quantization table, and the quantization table is scaled and adjusted according to the target compression ratio, and the high-frequency information that the human eye is not sensitive to is discarded. The quantized coefficient matrix is ​​scanned in a zigzag shape, the two-dimensional matrix is ​​converted into a one-dimensional sequence, and then the sequence is compressed using arithmetic coding, and the symbols are encoded by a probability model to generate a compressed bit stream. The compressed bit stream is encapsulated in JPEG format, and file header information is added, including SOI marker, APP0 segment, DQT segment, SOF0 segment, DHT segment and SOS segment, wherein the DQT segment stores the quantization table, the SOF0 segment records the image size and sampling factor, and the DHT segment contains the Huffman table, and finally the industrial parts burr defect verification code JPEG file is generated. The generated verification code image is integrated with the defect detection system of the workpiece parts. The verification code decoding step is added in the preprocessing stage of the image recognition algorithm. The verification code information is extracted by reverse JPEG decompression and character recognition algorithm as a reference for subsequent defect detection. When processing the verification code image of 100x30 pixels, the grayscale conversion is first performed using the weighted average method, and the weight coefficients are R: 0.299, G: 0.587, and B: 0.114. The converted grayscale image is divided into 4x4 blocks of 8x8 pixels. A two-dimensional DCT transform is applied to each block to obtain 64 frequency coefficients. Quantization is performed using the JPEG standard brightness quantization table, and the quantization table is scaled according to the target compression ratio. The quantized coefficient matrix is ​​converted into a one-dimensional sequence by zigzag scanning. The sequence is compressed using adaptive arithmetic coding. The initial probability model assumes that the probability of all symbols is equal, and the probability is dynamically updated during the encoding process. The compressed bitstream is encapsulated in JPEG format. The file header contains SOI marker, APP0 segment, DQT segment stores quantization table, SOF0 segment records image size and sampling factor, DHT segment stores Huffman table, and SOS segment marks the start of image data. The size of the final generated JPEG file is about 30-40% of the original image. In the defect detection system, the verification code image is first decoded by JPEG and restored to a grayscale image. Then the adaptive threshold segmentation algorithm is applied to separate the characters from the background to obtain a binary image. Connected region analysis is used to extract each character, and the area, perimeter ratio, and direction of the character are calculated. Finally, the characters are recognized by the pre-trained support vector machine classifier, and the recognition results are used as auxiliary information for defect detection to verify the reliability of the detection results.

[0041] The above embodiment is only one of the preferred implementation modes of the present invention and should not be used to limit the protection scope of the present invention. Any changes or modifications that are made to the main design concept and spirit of the present invention and have no substantive significance, and the technical problems they solve are still consistent with the present invention, should be included in the protection scope of the present invention.

Claims

1. A method for detecting defects in industrial parts, characterized in that: The method comprises: Obtain an input burr defect image of an industrial part, extract the shape, size, and distribution position features of the burrs in the image, obtain a burr feature map, segment the burr defect area according to the edge roughness and color difference in the burr feature map, and generate a binary burr defect mask; Perform morphological operations on the binary burr defect mask, eliminate isolated noise points through dilation and corrosion, connect adjacent burr areas, segment the processed burr mask, and extract independent burr defect areas based on the 3D contour and surface texture features of the burr; For each independent burr defect area, its geometric features are calculated, a burr defect feature vector is constructed, and the semantic features of the burr defect feature vector are extracted to obtain the feature representation of the burr defect. Different types of burr defects are grouped based on the feature similarity of the burr defects, and the shape and size features of the most representative samples in each group of burr defects are selected as the representative features of the burr defects in this group. Establishing a mapping relationship between each group of burr defect representatives and verification code characters, including mapping the burr shape and size attributes to the shape and size characteristics of the character, dividing the original burr defect image into a number of grids, and generating a preliminary verification code character matrix according to the grid position and the verification code character mapping results; Perform image enhancement processing on the preliminary verification code character matrix, adjust the color contrast of the characters to highlight the thickness of the character strokes, and generate a verification code image; The generated verification code image is compressed and format converted to obtain the final industrial parts burr defect verification code for defect detection of workpiece parts.

2. The method according to claim 1, wherein: The method obtains an input burr defect image of an industrial part, extracts the shape, size, and distribution position features of the burrs in the image, obtains a burr feature map, segments the burr defect area according to the edge roughness and color difference in the burr feature map, and generates a binary burr defect mask, including: The acquired industrial parts burr defect image is preprocessed by linear grayscale transformation and adaptive histogram equalization to obtain a preprocessed image with enhanced contrast and highlighted burr features; Extracting the shape, size and distribution position of the burr according to the preprocessed image, and generating a burr feature map including the burr contour, area and coordinates; Using the Canny operator to perform edge detection on the burr feature map, obtain edge information in the burr feature map, and calculate the gradient amplitude and direction of edge pixel points; If the edge information satisfies the high and low thresholds automatically determined by the OTSU algorithm, the edge is binarized to obtain a preliminary burr contour; Performing regional segmentation on the preliminary burr contour using a regional growing algorithm based on edge and grayscale gradient, starting with a seed point obtained by edge detection, and gradually expanding the segmented region; If the gray value difference and gradient direction difference of adjacent pixels are both less than the set threshold, they will be included in the same area until they cannot be expanded further; Evaluating and optimizing the segmented regions, and using shape features and texture features to evaluate the rationality of the segmented regions; If the segmented regions are unreasonable, a merge or split operation is performed on them.

3. The method according to claim 1, wherein: The binary burr defect mask is subjected to morphological operation, isolated noise points are eliminated by dilation and corrosion, adjacent burr regions are connected, the processed burr mask is segmented, and independent burr defect regions are extracted according to the three-dimensional contour and surface texture features of the burr, including: Applying a dilation operation to the binary burr defect mask, iterating with a structural element to expand the burr region boundary, and then performing an erosion operation to remove isolated noise points to obtain an optimized burr mask image; Extracting contour information of the optimized burr mask using a Sobel operator, calculating curvature change and direction consistency based on the contour information, and determining connectivity of adjacent burr regions; A rigid body transformation algorithm is used to align the two-dimensional burr mask image with the three-dimensional point cloud data collected by the depth camera, and the two-dimensional image coordinate system is converted to the three-dimensional point cloud coordinate system; Performing bilateral filtering and noise reduction on the point cloud data, using the moving least squares method to perform surface reconstruction and smoothing, extracting the depth value and normal vector information of each pixel, and combining the grayscale value of the two-dimensional image to construct a feature vector including the depth value, the component of the normal vector and the grayscale value; The feature vector is segmented using a region growing algorithm, the edge points of the burr area are selected as seed points, and the area is gradually expanded. When the Euclidean distance of the feature vectors of adjacent points is less than a preset threshold, they are included in the same area until they can no longer be expanded, and an independent burr defect area is obtained.

4. The method according to claim 1, wherein: The method of calculating the geometric features of each independent burr defect area, constructing a burr defect feature vector, extracting the semantic features of the burr defect feature vector, obtaining a feature representation of the burr defect, grouping different types of burr defects based on the feature similarity of the burr defects, and selecting the shape and size features of the most representative samples in each group of burr defects as representative features of the group of burr defects includes: Obtain the geometric features and texture features of the burr defect area and construct a multi-dimensional burr defect feature vector; Wherein, the geometric features include area, perimeter, circularity, rectangularity, aspect ratio and orientation angle; Using an autoencoder with three layers of encoders and three layers of decoders to perform dimensionality reduction processing on the multi-dimensional burr defect feature vector; If the number of nodes in the middle layer is 1 / 4 of the dimension of the original feature vector, a low-dimensional burr defect feature representation is obtained through the compression and reconstruction process; Performing Z-score normalization on the feature vector output by the autoencoder, and calculating the cosine similarity between different burr defect samples; Constructing a similarity matrix according to the cosine similarity, and analyzing the similarity matrix using a density-based clustering algorithm; If the neighborhood radius is 0.5 and the minimum number of samples is 4, burr defects with similar characteristics are automatically grouped; For each group of burr defects, the average feature vector and variance of the samples in the group are calculated; If the Euclidean distance weight between the sample and the average feature vector in the group is 0.7 and the variance weight of each dimension of the sample feature is 0.3, the sample with the smallest weighted sum is selected as the representative sample of the group.

5. The method according to claim 1, wherein: The mapping relationship between each group of burr defect representatives and verification code characters is established, including mapping the burr shape and size attributes to the shape and size features of the characters, dividing the original burr defect image into a plurality of grids, and generating a preliminary verification code character matrix according to the grid positions and the verification code character mapping results, including: Acquire representative features of burr defects and features of verification code characters, and establish a database of corresponding relationships between the representative features of burr defects and features of verification code characters; Applying the Otsu algorithm to the original burr defect image according to the corresponding relationship database, and performing adaptive threshold segmentation to extract independent burr areas in the burr defect image; For the independent burr area, query the corresponding relationship database to obtain characters matching the independent burr area, fill the characters into preset grid positions, and obtain a preliminary character matrix; Using a bigram model to calculate the probability of adjacent character combinations in the preliminary character matrix, and determining whether the probability is lower than a preset threshold; If the probability is lower than a preset threshold, recalculate the matching degree between the burr feature at the position and other candidate characters, select the suboptimal matching result to replace the corresponding character in the preliminary character matrix, and obtain an optimized character matrix; The method also includes: finding corresponding characters in the verification code character set according to representative features of the burr defects in the grid, arranging the characters corresponding to the grid according to the grid positions, generating a character matrix, analyzing the edge contours of the character matrix, identifying and completing the missing parts, and obtaining the completed character matrix; analyzing the spacing and size relationship between the characters in the completed character matrix, segmenting the overlapping characters that exceed the overlap requirements, and obtaining optimized characters; splicing the feature images of each character according to the matrix position to obtain a preliminary verification code character matrix.

6. The method according to claim 5, wherein: The method includes: finding corresponding characters in the verification code character set according to representative features of the burr defects in the grid, arranging the characters corresponding to the grid according to the grid positions, generating a character matrix, analyzing the edge contours of the character matrix, identifying and completing the missing parts, and obtaining the completed character matrix, including: A feature vector is constructed according to the area, perimeter ratio and direction angle of the burr defect in the grid, and the character with the highest matching degree is searched in the pre-established verification code character feature library through the support vector machine classifier to obtain the initial character matrix; Applying the Canny edge detection algorithm to the initial character matrix, calculating the image gradient amplitude and direction, obtaining the edge contour of the character matrix through a double threshold method and edge connection, and identifying areas where the edges are discontinuous or missing; For the identified missing area, the missing part is preliminarily filled using a morphological reconstruction algorithm in combination with shape features and semantic information of adjacent characters, a structural element close to the character stroke width is selected, and the character contour is reconstructed through iterative expansion and intersection operations to obtain a repaired character contour; A convolutional neural network with a U-Net structure is used to refine the restored character contours. The network input is the restored character image, and the output is a complete character image. By learning the local structural features of the characters, character details that conform to the context are generated to obtain a completed character matrix. The completed character matrix is ​​verified by applying an optical character recognition algorithm, and the recognition confidence is calculated. If the recognition confidence is lower than a preset threshold, the convolutional neural network is fine-tuned until a completed character matrix with high confidence is obtained.

7. The method according to claim 5, wherein: The spacing and size relationship between characters in the completed character matrix are analyzed, and overlapping characters exceeding the overlap requirement are segmented to obtain optimized characters; The feature images of each character are spliced ​​according to the matrix position to obtain a preliminary verification code character matrix, including: The connected domain analysis algorithm is applied to the completed character matrix, and the independent character regions are identified through 8-neighborhood connectivity judgment to obtain the position and size information of the characters; Calculating the ratio of the overlapping area of ​​the adjacent character bounding boxes to the character area, and if the ratio exceeds a preset threshold, separating the overlapping characters using a segmentation method based on contour analysis; Morphological processing is performed on the separated character images. The rectangular structure element is used to perform opening operations to remove noise points and closing operations to fill the internal holes to obtain a character image with a complete shape. Analyze the spacing and size relationship between characters, calculate the Euclidean distance between the center points of adjacent characters and the character aspect ratio, and if the spacing is less than a preset ratio of the average character width or the aspect ratio difference exceeds a preset threshold, scale and adjust the position of the characters; According to the adjusted character matrix position information, the optimized character feature images are spliced ​​in the original arrangement order, and the Gaussian kernel is applied to the transition area between characters for fuzzy processing to obtain a preliminary verification code character matrix.

8. The method according to claim 1, wherein: The image enhancement process is performed on the preliminary verification code character matrix, the color contrast of the characters is adjusted to highlight the thickness of the character strokes, and the verification code image is generated, including: Applying contrast-limited adaptive histogram equalization processing to the verification code character matrix, the processing divides the image into local areas, performs histogram equalization on each area, and combines the processing results by interpolation method; According to the equalization processing result, the Otsu algorithm is used to perform adaptive threshold segmentation, determine the optimal threshold, separate the character area from the background, and obtain a binary character image; Performing morphological operations on the binary character image, using a skeleton extraction algorithm based on distance transformation to retain the main structure of the character, and using structural elements to perform dilation operations; The processed character image is merged with the background layer generated by using the Perlin noise algorithm, and the degree of fusion between the character and the background is controlled by using an alpha blending algorithm; Gaussian noise is added to the fused image, the mean value of the Gaussian noise is a preset value, the standard deviation is a preset value, and a bilinear interpolation method is used to randomly distort the local area, and the distortion amplitude is controlled within a preset range to generate a final verification code image.

9. The method according to claim 1, wherein: The generated verification code image is compressed and format converted to obtain a final industrial part burr defect verification code for defect detection of workpiece parts, including: Performing grayscale conversion on the verification code image, wherein the grayscale conversion converts the RGB value into a single-channel grayscale value by a weighted average method; Performing frequency domain analysis on the grayscale image according to the single-channel grayscale value, wherein the frequency domain analysis includes dividing the grayscale image into 8x8 small blocks, performing two-dimensional discrete cosine transform on the small blocks, and obtaining a frequency domain coefficient matrix; Performing a zigzag scan on the frequency domain coefficient matrix, wherein the zigzag scan converts the frequency domain coefficient matrix into a one-dimensional sequence; Compressing the one-dimensional sequence by using arithmetic coding, wherein the compression encodes symbols through a probability model to obtain a compressed bit stream; Encapsulating the compressed bit stream into a JPEG format, wherein the encapsulation includes adding file header information, wherein the file header information includes an SOI marker, an APP0 segment, a DQT segment, a SOF0 segment, a DHT segment, and an SOS segment; The DQT segment stores a quantization table, the SOF0 segment records image size and sampling factor, and the DHT segment contains a Huffman table.

Images