Material surface embossing defect identification method and system based on visual inspection

Abstract

The invention provides a material surface embossing defect identification method and system based on visual inspection, and relates to the technical field of visual inspection.The method comprises the steps that visual inspection process data, used for defect identification, of a current material surface embossing production line are collected, and the visual inspection process data are classified and aggregated according to defect type identifiers; performing identification performance consistency analysis on the multiple types of visual inspection process data, extracting an enhanced feature vector corresponding to a defect type identifier to be optimized, optimizing current visual inspection control data corresponding to the defect type identifier according to the enhanced feature vector, and outputting the optimized visual inspection control data, and overlying and storing visual detection process data. The technical problem that in the prior art, when different defect types are embossed on the surface of a treated material through visual detection, performance expressions are inconsistent, and consequently overall defect recognition is unstable is solved. The technical effect of improving the overall performance and reliability of identifying the embossing defects on the surface of the material through visual inspection is achieved.

Description

Technical Field

[0001] This invention relates to the field of visual inspection technology, and specifically to a method and system for identifying embossed defects on material surfaces based on visual inspection. Background Technology

[0002] In the embossing process of material surfaces, visual inspection technology is widely used for defect identification and quality control. However, existing visual inspection technologies exhibit significant performance differences when handling different types of defects in material surface embossing. Some defect types, due to their complexity or subtle features, may result in low detection efficiency and long processing times, while others, due to their indistinct features or high similarity to normal patterns, suffer from insufficient detection accuracy. This inconsistency in visual inspection performance not only affects the efficiency and accuracy of defect identification but also makes it difficult to guarantee the stability of the overall inspection process. In actual production, the instability of defect detection leads to missed or false detections, thereby affecting the final quality of the product.

[0003] Existing technologies suffer from inconsistent performance of visual inspection when processing different types of defects in embossed patterns on material surfaces, leading to unstable overall defect identification. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for identifying embossed defects on material surfaces based on visual inspection, in order to solve the technical problem that existing technologies have inconsistent performance of visual inspection when processing different types of embossed defects on material surfaces, resulting in unstable overall defect identification.

[0005] In view of the above problems, this application provides a method and system for identifying embossed defects on material surfaces based on visual inspection.

[0006] The first aspect of this application provides a method for identifying defects in embossing on material surfaces based on visual inspection. The method includes: collecting visual inspection process data from a current material surface embossing production line for defect identification; the visual inspection process data includes visual inspection control data, defect type identifiers, and identification performance data; classifying and aggregating the visual inspection process data according to the defect type identifiers to output multiple types of visual inspection process data; performing a consistency analysis of the identification performance of the multiple types of visual inspection process data to determine defect type identifiers to be optimized, and extracting enhanced feature vectors corresponding to the defect type identifiers to be optimized; optimizing the current visual inspection control data corresponding to the defect type identifiers according to the enhanced feature vectors, outputting optimized visual inspection control data, and storing the optimized visual inspection control data overlaying the visual inspection process data.

[0007] Optionally, the optimized visual inspection control data is applied to the visual inspection system of the current material surface embossing production line; when the current material surface embossing production line is detected to switch to the next material surface embossing production line, the visual inspection process data for defect identification of the next material surface embossing production line is re-collected, and the optimized visual inspection control data corresponding to the next visual inspection process data is obtained.

[0008] Optionally, the consistency of recognition performance includes efficiency consistency, quality consistency, and stability consistency. The efficiency consistency includes average processing time and time consumption distribution quantiles. The quality consistency includes the median recognition confidence and the standard deviation of recognition confidence. The stability consistency includes the standard deviation of processing time and the process complexity index. The mean recognition performance of the multi-class visual inspection process data is calculated. Defect types with performance values ​​lower than the mean recognition performance are selected based on the recognition performance of the multi-class visual inspection process data, and the defect type identifiers to be optimized are output.

[0009] Optionally, a consistency tolerance threshold is set, which includes an efficiency consistency threshold, a quality consistency threshold, and a stability consistency threshold; the average recognition performance is defined based on the consistency tolerance threshold to obtain an average recognition performance threshold; defect types that deviate from the average recognition performance threshold are filtered according to the recognition performance of the multi-class visual inspection process data, and defect type identifiers to be optimized are output.

[0010] Optionally, the output defect type identifier to be optimized includes an optimization label, which includes at least one of efficiency inconsistency, quality inconsistency, and stability inconsistency; a first set of defect performance feature vectors of the defect type identifier and a second set of defect performance feature vectors of the optimization label are extracted; enhanced correlation analysis is performed according to the first set of defect performance feature vectors and the second set of defect performance feature vectors, and the enhanced feature vector corresponding to the defect type identifier to be optimized is output.

[0011] Optionally, feature standardization is performed on the first group of defect manifestation feature vectors and the second group of defect manifestation feature vectors to output the standardized first group of defect manifestation feature vectors and the second group of defect manifestation feature vectors; mutual information is sampled to calculate the feature correlation matrix between the standardized first group of defect manifestation feature vectors and the second group of defect manifestation feature vectors; a semantic distance matrix is ​​introduced to fuse the feature correlation matrix to obtain a semantic-feature fusion matrix; the semantic-feature fusion matrix is ​​analyzed to obtain feature enhancement weights, and the first group of defect manifestation feature vectors is enhanced according to the feature enhancement weights to obtain enhanced feature vectors.

[0012] Optionally, PCA principal component analysis is performed on the semantic-feature fusion matrix to obtain the first k groups of left and right feature vectors that are greater than a preset principal component threshold according to the PCA principal component analysis results; feature enhancement weights are calculated based on the first k groups of left and right feature vectors.

[0013] Optionally, a subset of control data related to the defect type identifier to be optimized is extracted from the current visual inspection control data; the enhanced feature vector is mapped to the subset of control data to obtain a control parameter adjustment vector; the corresponding visual inspection control data is optimized according to the control parameter adjustment vector, and the optimized visual inspection control data is output.

[0014] Optionally, the visual inspection control data includes at least one of the following: image preprocessing parameters, selection identifiers for the defect feature extraction algorithm, processing parameters for the defect feature extraction algorithm, and confidence threshold of the defect classification model.

[0015] A second aspect of this application provides a visual inspection-based material surface embossing defect recognition system. The system includes: a data acquisition module for acquiring visual inspection process data used for defect recognition on a current material surface embossing production line, the visual inspection process data including visual inspection control data, defect type identifiers, and recognition performance; a data classification module for classifying and aggregating the visual inspection process data according to the defect type identifiers, outputting multiple types of visual inspection process data; a defect type determination module for performing consistency analysis of the recognition performance of the multiple types of visual inspection process data, determining the defect type identifiers to be optimized, and extracting the enhanced feature vectors corresponding to the defect type identifiers to be optimized; and a data overlay module for optimizing the current visual inspection control data corresponding to the defect type identifiers according to the enhanced feature vectors, outputting optimized visual inspection control data, and overlaying the stored visual inspection process data with the optimized visual inspection control data.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages: The method provided in this application collects visual inspection process data for defect identification on a current material surface embossing production line. This visual inspection process data includes visual inspection control data, defect type identifiers, and identification performance data. The method categorizes and aggregates the visual inspection process data according to the defect type identifiers, outputting multiple types of visual inspection process data. A consistency analysis of the identification performance of these multiple types of visual inspection process data is performed to determine the defect type identifiers to be optimized. Enhanced feature vectors corresponding to the defect type identifiers to be optimized are extracted. The current visual inspection control data for the corresponding defect type identifier is optimized according to the enhanced feature vectors, and optimized visual inspection control data is output. This optimized visual inspection control data overwrites the stored visual inspection process data. This achieves the technical effect of improving the efficiency, accuracy, and stability of visual inspection for each type of defect identification, thereby improving the overall performance and reliability of visual inspection for identifying embossed defects on material surfaces.

[0017] The above description is merely an overview of the technical solution of this application. To enable a clearer understanding of the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 A flowchart illustrating the visual inspection-based method for identifying embossed defects on material surfaces provided in this application.

[0020] Figure 2 This is a schematic diagram of the structure of the visual inspection-based material surface embossing defect recognition system provided in this application.

[0021] Figure labeling: Data acquisition module 11, data classification module 12, defect type determination module 13, data coverage module 14. Detailed Implementation

[0022] This application provides a method and system for identifying embossed defects on material surfaces based on visual inspection. It addresses the technical problem of inconsistent performance of visual inspection in handling different types of embossed defects on material surfaces, leading to overall instability in defect identification. This method improves the overall performance and reliability of visual inspection in identifying embossed defects on material surfaces.

[0023] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all of them.

[0024] Example 1, as Figure 1 As shown, this application provides a method for identifying embossed defects on material surfaces based on visual inspection. The method includes: The visual inspection process data for defect identification is collected from the current material surface embossing production line. The visual inspection process data includes visual inspection control data, defect type identification, and identification performance.

[0025] Specifically, a material surface embossing production line is a production line used for embossing processes on material surfaces. Visual inspection equipment, including high-resolution industrial cameras and lighting equipment, is installed on this production line. The high-resolution industrial cameras are used to capture images of the embossed surface, and the lighting equipment provides suitable illumination to ensure clear and stable image quality. Visual inspection process data for defect identification is collected from the current material surface embossing production line. This data refers to a series of relevant data collected, processed, and analyzed during the identification of embossed surface defects. This includes visual inspection control data, defect type identification, and identification performance. The visual inspection control data is the detection process data for all defects on the current material surface embossing production line. Defect type identification refers to the identification of material surface embossing defects detected and identified during the production process, such as scratches, indentations, and bubbles, each type corresponding to different abnormal characteristics. Identification performance refers to the specific performance of defect identification during the visual inspection process, including but not limited to a comprehensive evaluation of the efficiency, quality, and stability of each type of defect identification.

[0026] By comprehensively and accurately collecting visual inspection process data from the material surface embossing production line, comprehensive and reliable data support is provided for defect identification optimization, thereby improving the overall performance of material surface embossing defect identification.

[0027] Furthermore, the visual inspection control data includes at least one of the following: image preprocessing parameters, selection identifiers for the defect feature extraction algorithm, processing parameters for the defect feature extraction algorithm, and confidence threshold of the defect classification model.

[0028] Specifically, visual detection and identification of embossed defects on material surfaces includes, but is not limited to, preprocessing the images acquired by the visual inspection device before they enter the defect detection algorithm; processing the preprocessed images using defect feature extraction algorithms, such as edge detection algorithms and gray-level co-occurrence matrix algorithms, based on the defect characteristics and detection requirements of the embossed material surfaces; and then using a defect classification model to identify embossed defects on the material surfaces.

[0029] Based on this, the visual inspection control data should include at least one of the following: image preprocessing parameters, selection identifiers for defect feature extraction algorithms, processing parameters for defect feature extraction algorithms, and confidence thresholds for defect classification models. Image preprocessing parameters refer to a series of operational parameter settings performed on the image before the defect detection algorithm processes it, including but not limited to image grayscale conversion, filtering, and contrast enhancement. The selection identifier for defect feature extraction algorithms refers to which algorithm is chosen to extract defect features from the image during the visual inspection process. For example, for embossing defects with obvious edge features, the Canny edge detection algorithm based on edge detection is selected; for defects with specific texture features, an algorithm based on texture analysis, such as the gray-level co-occurrence matrix algorithm, is selected. After algorithm selection, the identifier information of the selected algorithm is recorded. Different defect feature extraction algorithms have different processing parameters. For example, for edge detection-based algorithms, parameters such as the edge detection threshold, the type and size of the gradient operator, etc., need to be set. The processing parameters of the defect feature extraction algorithm are obtained synchronously according to the defect feature extraction algorithm. The confidence threshold of a defect classification model refers to the threshold used to judge the reliability of the classification result during the defect classification process. Defect classification results with a confidence level higher than the confidence threshold are considered reliable defect identification results.

[0030] The visual inspection process data is classified and aggregated according to the defect type identifier, and multiple types of visual inspection process data are output.

[0031] Specifically, each visual inspection process data point includes a defect type identifier, indicating the type of defect identified in the image, such as scratches, indentations, or bubbles. Based on the defect type identifier, the visual inspection process data is categorized into multiple classes. After categorization, the corresponding visual inspection process data for each class is aggregated to achieve data merging. Through classification and aggregation, multi-class visual inspection process data is output. This multi-class data is divided into multiple subsets based on defect type, with each subset containing all relevant data for that defect type during the visual inspection process.

[0032] By classifying visual inspection process data according to defect type, it is possible to analyze different defect types and then make targeted optimizations and adjustments to improve the accuracy and efficiency of material surface embossing defect detection.

[0033] A consistency analysis of the recognition performance of the various types of visual inspection process data is performed to determine the defect type identifiers to be optimized, and the enhanced feature vectors corresponding to the defect type identifiers to be optimized are extracted.

[0034] Furthermore, a consistency analysis of the recognition performance of the multiple types of visual inspection process data is performed to determine the defect type identifiers to be optimized. The method includes: wherein the consistency of recognition performance includes efficiency consistency, quality consistency, and stability consistency; the efficiency consistency includes average processing time and time consumption distribution quantiles; the quality consistency includes the median recognition confidence and the standard deviation of recognition confidence; and the stability consistency includes the standard deviation of processing time and the process complexity index; the mean recognition performance of the multiple types of visual inspection process data is calculated; and defect types with recognition performance values ​​lower than the mean recognition performance are selected according to the recognition performance of the multiple types of visual inspection process data, and defect type identifiers to be optimized are output.

[0035] Specifically, after obtaining data from multiple visual inspection processes, a consistency analysis of recognition performance is performed to identify which defect types have unstable detection performance, low efficiency, or poor recognition quality, thus providing a clear direction for visual inspection optimization. This consistency includes efficiency consistency, quality consistency, and stability consistency. Efficiency consistency includes average processing time and time distribution quantiles. For each type of visual inspection process data, the defect detection processing time of all samples is statistically analyzed, and the average value is calculated. If the average processing time for a certain defect type is significantly higher than that of other defect types, it indicates that the detection efficiency of that defect type is low and needs optimization. Calculating time distribution quantiles, such as the median and upper quartile, can further detect time fluctuations. Quality consistency includes the median recognition confidence score and the standard deviation of the recognition confidence score. Recognition confidence score refers to the degree of confidence the defect classification model has in the detection results, representing the probability that a certain defect type is correctly identified. For each type of visual inspection process data, the median confidence score of all recognition results is calculated. If the median confidence score for a certain defect type is low, it indicates that the recognition results for that defect type are unreliable and the recognition quality is poor. The standard deviation of the recognition confidence score is calculated for each type of visual inspection process data. The standard deviation measures the stability of the recognition results. If the standard deviation of the recognition confidence score for a certain defect type is large, it indicates that the recognition results for that defect category fluctuate significantly across different inspection processes, and the recognition quality is unstable.

[0036] Stability consistency includes processing time standard deviation and process complexity indicators. By analyzing the processing time fluctuations for each defect type across different inspection batches, processing time standardization is obtained. Significant time fluctuations for a particular defect type across different batches indicate insufficient stability in defect detection for that type. Each defect type's detection process may involve different image processing algorithms, feature extraction methods, etc. Due to differences in algorithm complexity and feature dimensions, the complexity of the detection process for each defect type is calculated using weighted averages to obtain a process complexity indicator. Excessive complexity in the detection process for a particular defect may lead to unstable results. By assigning different weights to efficiency, quality, and stability, the results of efficiency consistency, quality consistency, and stability consistency are weighted and calculated to obtain the recognition performance of each type of visual inspection process data. Weight allocation can be dynamically set according to actual needs; if recognition quality is more important for production line defect identification, higher weight can be assigned to quality consistency.

[0037] By performing efficiency, quality, and stability consistency analyses on each type of visual inspection process data across multiple categories, the recognition performance of each category is obtained. The average recognition performance of each category is then calculated. Each category of visual inspection process data is compared to this average performance, and defect types with performance below the average are identified as defect types to be optimized. Corresponding identifiers for these defect types are then output. Defect types to be optimized are those with excessively long processing times, poor recognition quality, and poor stability compared to other defect types.

[0038] By analyzing the consistency of recognition performance in visual inspection process data for different defect types, we can identify defect types that perform poorly and optimize them according to the difficulty of the defect process for that type. This will achieve a unified recognition standard, thereby improving the overall recognition performance of embossed defects on material surfaces and ensuring efficient, accurate, and stable defect recognition in production.

[0039] Furthermore, the method for filtering defect types that deviate from the average recognition performance of the multiple types of visual inspection process data according to the average recognition performance is further divided into: setting a consistency tolerance threshold, which includes an efficiency consistency threshold, a quality consistency threshold, and a stability consistency threshold; defining the average recognition performance based on the consistency tolerance threshold to obtain an average recognition performance threshold; filtering defect types that deviate from the average recognition performance threshold according to the recognition performance of the multiple types of visual inspection process data, and outputting defect type identifiers to be optimized.

[0040] Specifically, consistency tolerance thresholds are set, including efficiency consistency thresholds, quality consistency thresholds, and stability consistency thresholds. These can be set based on the mean ± 2 standard deviations or through analysis of historical data and dynamic equipment. The efficiency consistency threshold refers to whether the time required for the visual inspection process to process defect identification is within an acceptable range. When the inspection time exceeds the efficiency consistency threshold, it indicates poor efficiency. The quality consistency threshold refers to the acceptable accuracy range of the visual inspection process when identifying defect types. The stability consistency threshold reflects whether the visual inspection process exhibits consistent stability during processing. A mean recognition performance threshold is defined based on the consistency tolerance thresholds. For example, the efficiency consistency threshold, quality consistency threshold, and stability consistency threshold are weighted and summed to obtain the mean recognition performance threshold. Then, the recognition performance of multiple types of visual inspection process data is compared with the mean recognition performance threshold to identify defect types that deviate from the mean recognition performance threshold, which are then identified as defect types to be optimized.

[0041] By setting consistency tolerance thresholds and defining mean thresholds, defect types with poor recognition performance during visual inspection can be automatically screened out, thereby avoiding blind optimization, ensuring the targeted and efficient nature of optimization work, and improving the overall recognition performance of embossed defects on material surfaces, thus ensuring efficient, accurate and stable defect recognition in production.

[0042] Further, the method for extracting the enhanced feature vector corresponding to the defect type identifier to be optimized includes: wherein the output defect type identifier to be optimized includes an optimization label, the optimization label including at least one of efficiency inconsistency, quality inconsistency, and stability inconsistency; extracting a first set of defect performance feature vectors of the defect type identifier and a second set of defect performance feature vectors of the optimization label; performing enhanced correlation analysis according to the first set of defect performance feature vectors and the second set of defect performance feature vectors, and outputting the enhanced feature vector corresponding to the defect type identifier to be optimized.

[0043] Specifically, through analysis, defect types requiring optimization are identified, and identifiers for these defect types are output. These identifiers include labels for optimization, which may include at least one of efficiency inconsistency, quality inconsistency, and stability inconsistency. The first set of defect performance feature vectors from the defect type identifiers is extracted. This first set refers to the basic static features of the defect type to be optimized, which are static information closely related to the defect type itself, such as image texture features, shape features, and color distribution. Simultaneously, the second set of defect performance feature vectors for the labels to be optimized is extracted. This second set refers to dynamic process features strongly correlated with the labels to be optimized, closely related to dynamic changes during the visual inspection process, including the image preprocessing stage, the feature extraction algorithm selection process, and the confidence threshold setting of the defect classification model. For example, quality inconsistency may be related to the selection of the feature extraction algorithm and the adjustment of the model's confidence threshold.

[0044] Enhanced correlation analysis is performed on the first group of defect performance feature vectors and the second group of defect performance feature vectors. For example, methods such as mutual information and Pearson correlation coefficient are used to analyze the correlation between static features and dynamic process features. The enhanced feature vector corresponding to the defect type identifier to be optimized is output. The enhanced feature vector describes the specific reasons that lead to the bottleneck of defect type identification performance, thereby achieving targeted optimization.

[0045] By performing enhanced correlation analysis on the first set of defect performance feature vectors for defect type identification and the second set of defect performance feature vectors for the label to be optimized, the root cause of the inconsistency in efficiency, quality, or stability of the visual inspection process for the defect to be optimized can be accurately obtained, and then targeted optimization can be carried out to improve the overall performance of visual inspection in identifying embossed defects on material surfaces.

[0046] Furthermore, enhanced correlation analysis is performed based on the first group of defect manifestation feature vectors and the second group of defect manifestation feature vectors. The method includes: standardizing the first group of defect manifestation feature vectors and the second group of defect manifestation feature vectors to output the standardized first group of defect manifestation feature vectors and the second group of defect manifestation feature vectors; sampling mutual information to calculate the feature correlation matrix between the standardized first group of defect manifestation feature vectors and the second group of defect manifestation feature vectors; introducing a semantic distance matrix to fuse the feature correlation matrix to obtain a semantic-feature fusion matrix; analyzing the semantic-feature fusion matrix to obtain feature enhancement weights; and enhancing the first group of defect manifestation feature vectors according to the feature enhancement weights to obtain enhanced feature vectors.

[0047] Specifically, firstly, feature standardization is performed on the first and second sets of defect performance feature vectors to eliminate differences between different feature dimensions, ensuring all features are compared within the same scale and range, thus avoiding interference from differences in feature magnitudes on correlation analysis. Standardization methods include Z-score standardization or Min-Max normalization. Standardization yields the first and second sets of defect performance feature vectors. Then, mutual information is used to calculate the feature correlation between the standardized first and second sets of defect performance feature vectors. Mutual information is a measure of the dependency between two random variables, reflecting the degree of information sharing between the two features. Based on the calculated mutual information, a feature correlation matrix is ​​obtained between the first and second sets of defect performance feature vectors, reflecting the relationship between each pair of feature vectors.

[0048] Based on the semantic meanings of the first and second sets of defect manifestation feature vectors, semantic similarity between each pair of features is calculated using similarity metrics such as cosine similarity or Euclidean distance, yielding semantic distance. A semantic distance matrix is ​​constructed based on the calculated semantic distances, where each element represents the semantic similarity or distance between two features. The semantic distance matrix is ​​then used to fuse the feature correlation matrix. A semantic weight is assigned to the correlation between each pair of features in the feature correlation matrix based on the semantic distance matrix. Matrix operations, such as weighted summation, are then used to fuse the matrix, resulting in a semantic-feature fusion matrix. This semantic-feature fusion matrix contains both feature correlation and semantic information. The semantic-feature fusion matrix is ​​analyzed to obtain feature enhancement weights. These weights are then used to weight the first set of defect manifestation feature vectors, achieving enhancement processing and yielding enhanced feature vectors. These enhanced feature vectors represent the feature enhancement information for the defect type identifier to be optimized.

[0049] By introducing a semantic distance matrix and fusing it with the feature correlation matrix, the numerical and semantic correlations between features are comprehensively considered, improving the comprehensiveness and accuracy of feature analysis. Enhanced feature vectors are obtained through augmentation processing, thereby improving the discriminative power of the feature vectors. This enables more accurate localization and optimization of key factors affecting detection performance, improving the efficiency, accuracy, and stability of defect identification, and further enhancing the performance and adaptability of visual inspection.

[0050] Furthermore, the semantic-feature fusion matrix is ​​analyzed to obtain feature enhancement weights. The method includes: performing PCA principal component analysis on the semantic-feature fusion matrix, obtaining the first k groups of left and right feature vectors that are greater than a preset principal component threshold according to the PCA principal component analysis results; and calculating the feature enhancement weights based on the first k groups of left and right feature vectors.

[0051] Specifically, PCA principal component analysis is performed on the semantic-feature fusion matrix. First, the semantic-feature fusion matrix is ​​standardized, converting the value of each feature into standardized data with zero mean and unit variance. Then, a covariance matrix is ​​calculated from the standardized semantic-feature fusion matrix. Each element in the covariance matrix represents the covariance between different features. Positive covariance values ​​indicate that two features have similar trends, negative values ​​indicate opposite trends, and zero values ​​indicate no linear relationship between features. Singular value decomposition (SVD) is then performed on the covariance matrix to obtain the PCA principal component analysis results, including left and right eigenvectors. The left eigenvector represents the principal direction of the data, i.e., the contribution of each feature to the principal components, while the right eigenvector represents the specific representation of the data in the feature space, reflecting the direction of the principal component space.

[0052] A preset principal component threshold is set. The PCA principal component analysis results are compared with the preset threshold. The top k left eigenvectors and k right eigenvectors (where k is a positive integer) that are greater than the preset threshold are selected. The feature enhancement weights are calculated based on the top k left and right eigenvectors: W = α•U k +(1-α)•V k , among which, U k Let V be the k groups of left eigenvectors. k Let k be the right feature vectors, and α be 0-1, representing the fusion ratio of the first group of defect manifestation feature vectors and the second group of defect manifestation feature vectors.

[0053] By analyzing the relationship between each feature and the principal components and evaluating the importance of the features to the dimensionality reduction results, feature enhancement weights are obtained. This allows us to identify the features that contribute the most to the defect recognition task for the defect type to be optimized, thereby improving the targeting and efficiency of optimization and ultimately enhancing the overall recognition performance of visual inspection.

[0054] The current visual inspection control data is optimized according to the enhanced feature vector to identify the corresponding defect type, and the optimized visual inspection control data is output to overwrite the stored visual inspection process data.

[0055] Furthermore, the method for optimizing the current visual inspection control data corresponding to the defect type identifier according to the enhanced feature vector and outputting the optimized visual inspection control data includes: extracting a subset of control data related to the defect type identifier to be optimized from the current visual inspection control data; mapping the enhanced feature vector to the subset of control data to obtain a control parameter adjustment vector; optimizing the corresponding visual inspection control data according to the control parameter adjustment vector and outputting the optimized visual inspection control data.

[0056] Specifically, the data in the visual inspection process includes control data such as image processing, feature extraction, and classification decisions. A subset of control data related to the defect type identifier to be optimized is extracted from the current visual inspection control data. This subset includes image preprocessing parameters related to the defect type, the selection of feature extraction algorithms, and the confidence threshold of the defect classification model. For example, if the defect type to be optimized is a scratch, image preprocessing parameters and feature extraction algorithms related to scratches are selected. The enhanced feature vector is mapped to the control data subset to obtain a control parameter adjustment vector, which represents the optimization direction and magnitude of the visual inspection control parameters for each optimization objective.

[0057] The current visual inspection control data is adjusted and optimized one by one according to the obtained control parameter adjustment vector, and the optimized visual inspection control data is output. The optimized visual inspection control data includes the updated parameters, which enhances the ability of the visual inspection process to identify the types of defects to be optimized. The optimized visual inspection control data is overwritten with the stored visual inspection process data through a configuration file, so that the updated control data can be directly applied to subsequent defect detection tasks, achieving a unified defect identification standard, improving the performance of the visual inspection process, and enhancing the accuracy and stability of material surface embossing defect identification.

[0058] Furthermore, after outputting the optimized visual inspection and control data, the method also includes: applying the optimized visual inspection and control data to the visual inspection system of the current material surface embossing production line; when it is detected that the current material surface embossing production line is switched to the next material surface embossing production line, the next visual inspection process data for defect identification in the next material surface embossing production line is re-collected, and the optimized visual inspection and control data corresponding to the next visual inspection process data is obtained.

[0059] Specifically, when the optimized visual inspection control data is directly applied to the visual inspection system of the current material surface embossing production line, the optimized control data has already adjusted the control parameters for relevant defect identification through the previous optimization process, which can improve the overall defect identification effect of the current production line. When the current material surface embossing production line completes its work, the production line switches to the next material surface embossing production line. For example, if the material type or embossing process changes, it triggers the re-collection of visual inspection process data for the next material surface embossing production line. This visual inspection process data is then analyzed and optimized to obtain the optimized visual inspection control data corresponding to the next visual inspection process data. The visual inspection control data is then adjusted to adapt the visual inspection to the new material and process characteristics, thereby improving the overall performance of defect identification.

[0060] By continuously collecting and updating visual inspection control data, it is possible to adapt to changes in embossing production lines for different material surfaces, ensuring the effectiveness, accuracy, and stability of defect identification for each type, thereby improving the effectiveness of visual inspection and overall defect detection performance.

[0061] In summary, the visual inspection-based method for identifying embossed defects on material surfaces presented in this application has the following technical advantages: By collecting visual inspection process data for defect identification from the current material surface embossing production line, including visual inspection control data, defect type identifiers, and identification performance data, the visual inspection process data is categorized and aggregated according to the defect type identifiers, outputting multiple types of visual inspection process data. Consistency analysis of the identification performance of these multiple types of visual inspection process data is performed to determine the defect type identifiers to be optimized. Enhanced feature vectors corresponding to the defect type identifiers to be optimized are extracted. The current visual inspection control data for the corresponding defect type identifier is optimized according to the enhanced feature vectors, and the optimized visual inspection control data is output, overwriting the stored visual inspection process data. This achieves the technical effect of improving the efficiency, accuracy, and stability of visual inspection for each type of defect identification by analyzing the detection process data of all defects in the current material surface embossing production line, accurately identifying defect types with poor performance, and specifically extracting enhanced feature vectors to optimize visual inspection control data. This improves the overall performance and reliability of visual inspection for identifying defects in material surface embossing.

[0062] Example 2, based on the same inventive concept as the visual inspection-based material surface embossing defect identification method in the foregoing examples, such as... Figure 2 As shown, this application provides a visual inspection-based material surface embossing defect identification system, wherein the visual inspection-based material surface embossing defect identification system includes: The data acquisition module 11 is used to acquire visual inspection process data for defect identification on the current material surface embossing production line. The visual inspection process data includes visual inspection control data, defect type identifiers, and identification performance. The data classification module 12 is used to classify and aggregate the visual inspection process data according to the defect type identifiers and output multiple types of visual inspection process data. The defect type determination module 13 is used to perform consistency analysis of the identification performance of the multiple types of visual inspection process data, determine the defect type identifiers to be optimized, and extract the enhanced feature vectors corresponding to the defect type identifiers to be optimized. The data overlay module 14 is used to optimize the current visual inspection control data corresponding to the defect type identifiers according to the enhanced feature vectors, output optimized visual inspection control data, and overlay the stored visual inspection process data with the optimized visual inspection control data.

[0063] Furthermore, the data coverage module 14 is also used to: apply the optimized visual inspection control data to the visual inspection system of the current material surface embossing production line; when it is detected that the current material surface embossing production line is switched to the next material surface embossing production line, re-collect the next visual inspection process data used for defect identification in the next material surface embossing production line, and obtain the optimized visual inspection control data corresponding to the next visual inspection process data.

[0064] Furthermore, the defect type determination module 13 is also used to: identify performance consistency including efficiency consistency, quality consistency, and stability consistency, wherein efficiency consistency includes average processing time and time consumption distribution quantiles, quality consistency includes the median of recognition confidence and the standard deviation of recognition confidence, and stability consistency includes the standard deviation of processing time and process complexity index; calculate the mean of recognition performance of the multi-class visual inspection process data; filter defect types that are less than the mean of recognition performance according to the recognition performance of the multi-class visual inspection process data, and output the defect type identifier to be optimized.

[0065] Furthermore, the defect type determination module 13 is also used to: set a consistency tolerance threshold, which includes an efficiency consistency threshold, a quality consistency threshold, and a stability consistency threshold; define the average recognition performance based on the consistency tolerance threshold to obtain an average recognition performance threshold; filter defect types that deviate from the average recognition performance threshold according to the recognition performance of the multi-class visual inspection process data, and output the defect type identifier to be optimized.

[0066] Furthermore, the defect type determination module 13 is also used to: output the defect type identifier to be optimized, which includes an optimization label, wherein the optimization label includes at least one of efficiency inconsistency, quality inconsistency, and stability inconsistency; extract a first set of defect performance feature vectors of the defect type identifier and a second set of defect performance feature vectors of the optimization label; perform enhanced correlation analysis according to the first set of defect performance feature vectors and the second set of defect performance feature vectors, and output the enhanced feature vector corresponding to the defect type identifier to be optimized.

[0067] Furthermore, the defect type determination module 13 is also used to: standardize the first group of defect performance feature vectors and the second group of defect performance feature vectors, and output the standardized first group of defect performance feature vectors and the second group of defect performance feature vectors; sample mutual information to calculate the feature correlation matrix between the standardized first group of defect performance feature vectors and the second group of defect performance feature vectors; introduce a semantic distance matrix to fuse the feature correlation matrix to obtain a semantic-feature fusion matrix; analyze the semantic-feature fusion matrix to obtain feature enhancement weights, and enhance the first group of defect performance feature vectors according to the feature enhancement weights to obtain enhanced feature vectors.

[0068] Furthermore, the defect type determination module 13 is also used to: perform PCA principal component analysis on the semantic-feature fusion matrix, obtain the first k groups of left and right feature vectors that are greater than the preset principal component threshold according to the PCA principal component analysis results; and calculate the feature enhancement weights based on the first k groups of left and right feature vectors.

[0069] Furthermore, the data overlay module 14 is also used to: extract a subset of control data related to the defect type identifier to be optimized from the current visual inspection control data; map the enhanced feature vector to the subset of control data to obtain a control parameter adjustment vector; optimize the corresponding visual inspection control data according to the control parameter adjustment vector, and output the optimized visual inspection control data.

[0070] Furthermore, the data acquisition module 11 also includes: the visual detection control data includes at least one of the following: image preprocessing parameters, selection identifier of defect feature extraction algorithm, processing parameters of defect feature extraction algorithm, and confidence threshold of defect classification model.

[0071] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The material surface embossing defect identification method and specific examples based on visual detection in the aforementioned embodiment 1 are also applicable to the material surface embossing defect identification system based on visual detection in this embodiment. Through the foregoing detailed description of the material surface embossing defect identification method based on visual detection, those skilled in the art can clearly understand the material surface embossing defect identification system based on visual detection in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0072] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0073] Obviously, those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for identifying embossed defects on material surfaces based on visual inspection, characterized in that, The method includes: Collect visual inspection process data for defect identification on the current material surface embossing production line. The visual inspection process data includes visual inspection control data, defect type identification, and identification performance. The visual inspection process data is classified and aggregated according to the defect type identifier, and multiple types of visual inspection process data are output. A consistency analysis of the recognition performance of the various types of visual inspection process data is performed to determine the defect type identifiers to be optimized, and the enhanced feature vectors corresponding to the defect type identifiers to be optimized are extracted. The current visual inspection control data is optimized according to the enhanced feature vector to identify the corresponding defect type, and the optimized visual inspection control data is output to overwrite the stored visual inspection process data.

2. The method for identifying embossed defects on material surfaces based on visual inspection as described in claim 1, characterized in that, After outputting the optimized visual detection and control data, the method also includes: The optimized visual inspection control data is applied to the visual inspection system of the current material surface embossing production line; When the current material surface embossing production line is detected to switch to the next material surface embossing production line, the next visual inspection process data for defect identification in the next material surface embossing production line is re-collected, and the optimized visual inspection control data corresponding to the next visual inspection process data is obtained.

3. The method for identifying embossed defects on material surfaces based on visual inspection as described in claim 1, characterized in that, A consistency analysis of the recognition performance of the various types of visual inspection process data is performed to determine the defect type identifiers to be optimized. include: Among them, the consistency of recognition performance includes efficiency consistency, quality consistency and stability consistency. The efficiency consistency includes average processing time and time consumption distribution quantile. The quality consistency includes the median recognition confidence and the standard deviation of recognition confidence. The stability consistency includes the standard deviation of processing time and process complexity index. Calculate the average recognition performance of the various types of visual detection process data; Based on the recognition performance of the various visual inspection process data, defect types that are less than the average recognition performance are selected, and defect type identifiers to be optimized are output.

4. The method for identifying embossed defects on material surfaces based on visual inspection as described in claim 3, characterized in that, The method further includes filtering defect types that are less than the average recognition performance value based on the average recognition performance value of the multiple types of visual detection process data: Set a consistency tolerance threshold, which includes an efficiency consistency threshold, a quality consistency threshold, and a stability consistency threshold; The average recognition performance is defined based on the consistency tolerance threshold to obtain the average recognition performance threshold. Based on the recognition performance of the various visual inspection process data, defect types that deviate from the average threshold of the recognition performance are selected, and defect type identifiers to be optimized are output.

5. The method for identifying embossed defects on material surfaces based on visual inspection as described in claim 1, characterized in that, Extract the enhanced feature vector corresponding to the defect type identifier to be optimized, method include: The output defect type identifier to be optimized includes an optimization label, which includes at least one of efficiency inconsistency, quality inconsistency, and stability inconsistency. Extract the first set of defect performance feature vectors of the defect type identifier and the second set of defect performance feature vectors of the label to be optimized; Enhanced correlation analysis is performed based on the first set of defect performance feature vectors and the second set of defect performance feature vectors to output the enhanced feature vector corresponding to the defect type identifier to be optimized.

6. The method for identifying embossed defects on material surfaces based on visual inspection as described in claim 5, characterized in that, Enhanced correlation analysis is performed based on the first set of defect manifestation feature vectors and the second set of defect manifestation feature vectors, including the following methods: The first set of defect manifestation feature vectors and the second set of defect manifestation feature vectors are standardized to output the standardized first set of defect manifestation feature vectors and the second set of defect manifestation feature vectors. The feature correlation matrix between the first set of defect manifestation feature vectors and the second set of defect manifestation feature vectors after standardization is calculated using sampled mutual information. A semantic distance matrix is ​​introduced to fuse the feature correlation matrix, resulting in a semantic-feature fusion matrix; The semantic-feature fusion matrix is ​​analyzed to obtain feature enhancement weights. The first group of defect manifestation feature vectors are then enhanced according to the feature enhancement weights to obtain enhanced feature vectors.

7. The method for identifying embossed defects on material surfaces based on visual inspection as described in claim 6, characterized in that, The method for analyzing the semantic-feature fusion matrix to obtain feature enhancement weights includes: PCA principal component analysis is performed on the semantic-feature fusion matrix, and the first k left and right feature vectors with values ​​greater than the preset principal component threshold are obtained according to the PCA principal component analysis results. The feature enhancement weights are calculated based on the first k groups of left and right feature vectors.

8. The method for identifying embossed defects on material surfaces based on visual inspection as described in claim 1, characterized in that, The method includes optimizing the current visual inspection and control data corresponding to the defect type identifier according to the enhanced feature vector, and outputting the optimized visual inspection and control data. Extract a subset of control data related to the defect type identifier to be optimized from the current visual inspection control data; The enhanced feature vector is mapped to the control data subset to obtain the control parameter adjustment vector; The corresponding visual detection control data is optimized by adjusting the vector according to the control parameters, and the optimized visual detection control data is output.

9. The method for identifying embossed defects on material surfaces based on visual inspection as described in claim 1, characterized in that, The visual inspection control data includes at least one of the following: image preprocessing parameters, selection identifiers for the defect feature extraction algorithm, processing parameters for the defect feature extraction algorithm, and confidence threshold of the defect classification model.

10. A material surface embossing defect identification system based on visual inspection, characterized in that, The step of implementing the visual inspection-based material surface embossing defect identification method according to any one of claims 1 to 9, wherein the visual inspection-based material surface embossing defect identification system comprises: The data acquisition module is used to collect visual inspection process data for defect identification on the current material surface embossing production line. The visual inspection process data includes visual inspection control data, defect type identification, and identification performance. The data classification module is used to classify and aggregate the visual inspection process data according to the defect type identifier, and output multiple types of visual inspection process data; The module for determining the type of defect to be optimized is used to perform a consistency analysis of the recognition performance of the multi-type visual inspection process data, determine the defect type identifier to be optimized, and extract the enhanced feature vector corresponding to the defect type identifier to be optimized. The data overlay module is used to optimize the current visual inspection control data corresponding to the defect type identifier according to the enhanced feature vector, output the optimized visual inspection control data, and overlay the stored visual inspection process data with the optimized visual inspection control data.

Images