Image sensor-based mobile phone glass cover quality inspection method

Abstract

This invention relates to the field of image recognition quality inspection technology, specifically disclosing a method for inspecting the quality of mobile phone glass covers based on image sensors. An industrial camera is used to perform omnidirectional scanning of a mobile phone glass cover fixed on an inspection platform. Based on the acquired image data, crack concentration areas are determined according to crack distribution. Outlier values ​​of crack distribution characteristics are calculated through cluster analysis to identify areas with abnormal crack distributions. For each crack concentration area, the curvature fluctuation characteristics and crack propagation rate characteristics of the cracks are extracted. The curvature is approximated using the finite difference method and its Haar wavelet transform to obtain curvature fluctuation characteristic values. The local outlier factor algorithm is used to calculate crack propagation anomaly characteristic values. A comprehensive feature vector is constructed as input to a support vector machine model to predict the quality score of each mobile phone glass cover, thereby determining its quality level and marking low-quality products.

Description

Technical Field

[0001] This invention relates to the field of image recognition quality inspection technology, and more specifically to a method for inspecting the quality of mobile phone glass covers based on image sensors. Background Technology

[0002] With the rapid development of the smartphone market, consumers have increasingly higher requirements for the quality of mobile phone glass covers, paying attention not only to their aesthetic appearance but also to their durability and resistance to breakage. Traditional quality inspection methods mainly rely on manual visual inspection and simple automated optical inspection equipment. While these methods can identify obvious defects to some extent, their ability to identify micro-cracks, minor scratches, and defects with complex shapes is limited. Furthermore, traditional methods often lack the ability to predict crack propagation trends, failing to provide a comprehensive quality assessment. Therefore, developing an efficient, accurate, and real-time monitoring automated inspection system has become a key requirement for improving the quality of mobile phone glass covers.

[0003] The existing technology has the following shortcomings:

[0004] Current technologies for inspecting the quality of mobile phone glass covers have several shortcomings. First, traditional manual visual inspection is inefficient and easily affected by subjective factors, making it difficult to guarantee consistency and accuracy. Second, most existing automated optical inspection equipment uses fixed threshold segmentation algorithms, which have poor adaptability to images under different lighting conditions, leading to frequent missed or false detections. Furthermore, current technologies typically only identify visible surface defects, lacking in-depth analysis of crack propagation rates and curvature fluctuation characteristics, thus failing to effectively predict potential quality problems. More importantly, existing systems lack an integrated feedback mechanism to guide continuous optimization of the production process, making it difficult to achieve closed-loop management from inspection to improvement. These problems limit production efficiency. Summary of the Invention

[0005] The purpose of this invention is to provide a method for detecting the quality of mobile phone glass covers based on image sensors, so as to solve the problems mentioned above.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A method for quality inspection of mobile phone glass covers based on image sensors includes the following steps:

[0008] S1: Use an image sensor to scan the phone's glass cover from all angles to acquire raw image data and preprocess the raw image data;

[0009] The image sensor is an industrial camera;

[0010] S2: Based on the preprocessed original image data, identify the cracks in the mobile phone glass cover in the original image, and identify the concentrated crack areas according to the crack distribution of the mobile phone glass cover.

[0011] S3: Based on the crack concentration area, for each crack concentration area, extract the curvature fluctuation characteristics and crack propagation rate characteristics of the crack, analyze the curvature fluctuation and crack propagation rate characteristics of the crack, construct a comprehensive feature vector based on the analysis results, and use it as the input of the machine learning model. Based on the model output, determine the quality of each mobile phone glass cover.

[0012] S4: Based on the judgment results, classify the quality of each mobile phone glass cover into low quality and high quality, and mark the low quality mobile phone glass cover.

[0013] As a further aspect of the present invention: the identification of cracks in the mobile phone glass cover in the original image specifically includes:

[0014] The original grayscale image of the mobile phone glass cover is acquired, and an initial global threshold is calculated. Then, the image is divided into multiple non-overlapping small regions. A local threshold is calculated for each small region. The calculated local threshold is used to perform preliminary binarization processing on each small region to generate a series of partially binarized images. The location and bounding box of each crack are marked in the binarized images.

[0015] As a further aspect of the present invention: the identification of concentrated crack regions specifically includes:

[0016] The image is segmented into multiple non-overlapping small regions. Based on the crack distribution in each region, the crack distribution feature anomaly value in the corresponding region is calculated. It is then determined whether the crack distribution feature anomaly value in each region is greater than or equal to a preset threshold. If so, it is recorded as a crack concentration region.

[0017] As a further aspect of the present invention: the process for obtaining the outlier values ​​of the crack distribution characteristics is as follows:

[0018] The mobile phone glass cover is divided into several non-overlapping regions of the same size, and crack distribution feature values ​​are extracted for each region.

[0019] K-means clustering analysis was performed on the crack distribution feature values, and K cluster centers were initialized. For each region, the nearest cluster center was found according to the Euclidean distance formula and assigned to the corresponding cluster. The new center point position of each cluster was calculated and updated. The assignment and update process was repeated until the cluster center reached the maximum number of iterations. After obtaining the final clustering results, the average crack distribution feature value was calculated for each cluster. For each region, the sum of squares of the differences between the crack distribution feature value and the average feature value of its cluster was calculated, and the square root was taken to obtain the crack distribution feature outlier value.

[0020] As a further aspect of the present invention: based on crack concentration regions, for each crack concentration region, the curvature fluctuation and crack propagation rate characteristics of the crack are extracted, and the curvature fluctuation and crack propagation rate characteristics of the crack are analyzed, specifically including:

[0021] For each crack concentration area, the curvature fluctuation characteristics and crack propagation rate characteristics of the cracks are extracted. Based on the degree of curvature fluctuation, the curvature fluctuation characteristic value is calculated, and based on the crack propagation rate, the crack propagation anomaly characteristic value is calculated. The curvature fluctuation characteristic value and the crack propagation anomaly characteristic value are constructed into a comprehensive feature vector. The comprehensive feature vector of all regions is used as the input of the machine learning model. The output of the model is the quality score of each mobile phone glass cover. Based on the quality score, the quality of each mobile phone glass cover is determined.

[0022] As a further aspect of the present invention: the process for obtaining the curvature fluctuation characteristic value is as follows:

[0023] For each point on the center line of all cracks, the first and second derivatives of the arc length are approximately calculated using the finite difference method, and the curvature of each crack is calculated using the curvature calculation formula.

[0024] Apply Haar wavelet transform to the calculated curvature sequence to decompose the curvature sequence into different frequency components, and obtain a series of wavelet coefficients.

[0025] Based on the wavelet coefficients after wavelet transform, the sum of the standard deviations of all wavelet coefficients is calculated to obtain the curvature fluctuation characteristic value.

[0026] As a further aspect of the present invention: the process for obtaining the abnormal feature value of crack propagation is as follows:

[0027] Acquire and standardize the crack propagation rate dataset, determine the B-distance and B-neighborhood of each standardized data point, and calculate the reachability density of each data point relative to other points; calculate the local outlier factor of each data point based on the reachability density, and calculate the mean of all local outliers to obtain the crack propagation anomaly feature value.

[0028] As a further aspect of the present invention: the determination of the quality of each mobile phone glass cover specifically includes:

[0029] Determine whether the quality score of each mobile phone glass cover is greater than or equal to a preset threshold. If yes, mark it as a high-quality mobile phone glass cover; otherwise, mark it as a low-quality mobile phone glass cover.

[0030] As a further aspect of the present invention: the construction process of the machine learning model is as follows:

[0031] The curvature fluctuation characteristic value and crack propagation anomaly characteristic value of each crack concentration region are constructed into a comprehensive feature vector;

[0032] The comprehensive feature vectors of all regions are used as input to the machine learning model, which is a support vector machine model; the support vector machine model is used to output the quality score of each mobile phone glass cover, and the quality of each mobile phone glass cover is judged based on the quality score.

[0033] A series of mobile phone glass cover samples with known quality scores were collected and labeled to form a training set. The support vector machine model was trained using the training set. During the training process, the model parameters were adjusted using cross-validation techniques. After multiple iterations, the final support vector machine model was obtained. The comprehensive feature vector of each newly input mobile phone glass cover was input into the trained support vector machine model, and the quality score of the mobile phone glass cover was output.

[0034] The beneficial effects of this invention are:

[0035] (1) By employing high-resolution industrial cameras and advanced image processing technologies, such as local threshold segmentation, curvature fluctuation feature extraction, and crack propagation rate analysis, this invention enables accurate identification and quantitative assessment of cracks and other minute defects in mobile phone glass covers. Specifically, local threshold segmentation technology allows for accurate capture of crack edge information under different lighting conditions, while curvature fluctuation feature extraction accurately calculates the morphological characteristics of each crack using the finite difference method and Haar wavelet transform. Furthermore, it combines this with a local outlier factor algorithm based on reachability density to analyze the crack propagation rate, thereby comprehensively describing the development trend and potential risks of the crack. In particular, these feature vectors are input into an optimized support vector machine model for quality score prediction, which not only significantly improves the accuracy of individual defect judgment but also effectively identifies abnormal areas with complex patterns, greatly improving the overall accuracy and efficiency of the inspection. In addition, this automated inspection scheme significantly reduces the need for manual inspection and shortens the inspection cycle, making it particularly suitable for real-time quality monitoring on large-scale production lines. This ensures that the quality of each mobile phone glass cover meets high standards, thus providing strong technical support for improving production efficiency and reducing the defect rate. Through this systematic approach, the present invention not only achieves efficient and reliable defect detection, but also provides a data-driven foundation for continuous optimization of production processes, helping enterprises maintain a leading position in fierce market competition.

[0036] (2) By precisely marking low-quality mobile phone glass covers and arranging further manual or advanced automated equipment verification, this invention can record detailed defect information for each defective product, including multi-dimensional data such as the location, morphological characteristics, and propagation rate of cracks. These detailed defect analysis results not only provide valuable data support for subsequent process improvements but also enable the production team to quickly locate weak links in the production process, such as improper parameter settings or equipment aging in specific procedures. Based on this data, production process engineers can specifically adjust key process parameters and optimize production line configuration, thereby effectively reducing the frequency of similar defects and significantly improving overall product quality. Furthermore, the real-time feedback mechanism and continuous monitoring function provided by this invention enable manufacturing enterprises to continuously optimize their production processes in daily operations, identify potential quality risks through data analysis, and take preventative measures. In the long term, this systematic quality management strategy helps to significantly reduce the defect rate, improve resource utilization, reduce waste, and enhance the market competitiveness of products, ensuring that enterprises occupy a favorable position in fierce market competition. Through this comprehensive quality inspection and management solution, the present invention not only enables detailed evaluation of individual products, but also provides enterprises with a path for sustainable improvement, driving the entire production system toward higher quality standards. Attached Figure Description

[0037] The invention will now be further described with reference to the accompanying drawings.

[0038] Figure 1 This is a flowchart illustrating the specific steps of the mobile phone glass cover quality inspection method based on an image sensor according to the present invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] Please see Figure 1 As shown, this invention is a method for quality inspection of mobile phone glass covers based on image sensors, including the following steps:

[0041] S1: Use an image sensor to scan the phone's glass cover from all angles to acquire raw image data and preprocess the raw image data;

[0042] The image sensor is an industrial camera;

[0043] S2: Based on the preprocessed original image data, identify the cracks in the mobile phone glass cover in the original image, and identify the concentrated crack areas according to the crack distribution of the mobile phone glass cover.

[0044] S3: Based on the crack concentration area, for each crack concentration area, extract the curvature fluctuation characteristics and crack propagation rate characteristics of the crack, analyze the curvature fluctuation and crack propagation rate characteristics of the crack, construct a comprehensive feature vector based on the analysis results, and use it as the input of the machine learning model. Based on the model output, determine the quality of each mobile phone glass cover.

[0045] S4: Based on the judgment results, classify the quality of each mobile phone glass cover into low quality and high quality, and mark the low quality mobile phone glass cover.

[0046] In S1, an image sensor is used to perform a full-range scan of the phone's glass cover to acquire raw image data, and the raw image data is preprocessed, specifically including:

[0047] A high-resolution industrial camera is used to perform a full-range scan of the mobile phone's glass cover. First, the glass cover is fixed on a testing platform, and its position and angle are adjusted by an automated mechanical device to ensure the sensor covers the entire surface area. An appropriate light source system (such as a ring LED) is configured to reduce reflection interference and enhance surface texture features. The image sensor acquires data line by line or area by area at a preset resolution and frame rate, generating raw image data containing information about the glass cover's surface.

[0048] After acquiring the raw image data, preliminary preprocessing is performed to improve the accuracy of subsequent analysis. Preprocessing steps include image denoising (e.g., using Gaussian filtering), brightness and contrast adjustment to eliminate the effects of uneven illumination, and image registration to correct distortions caused by changes in scanning angle or position. The result is clear, standardized image data, laying the foundation for subsequent crack detection and feature extraction.

[0049] In S2, based on the preprocessed original image data, cracks in the phone's glass cover in the original image are identified, and the crack concentration areas are identified according to the crack distribution of the phone's glass cover. Specifically, this includes:

[0050] The original grayscale image of the phone's glass cover is acquired, and an initial global threshold is calculated. Then, the image is segmented into multiple non-overlapping small regions, and a local threshold is calculated for each region. The calculation expression is: T i =m(x,y)+A*(x,y)*(1+α*ΔC);

[0051] In the formula, A is the preset sensitivity value, α is the preset weighting factor, ΔC represents the contrast difference factor between adjacent regions, m(x,y) represents the mean value of pixels in the region, (x,y) represents the region coordinates, Ti represents the local threshold in the i-th region, and i represents the number of non-overlapping small regions into which the image is segmented; the calculated local threshold is used to perform preliminary binarization processing on each small region to generate a series of partially binarized images; and the location and bounding box of each crack are marked in the binarized images.

[0052] The image is divided into multiple non-overlapping small regions. Based on the crack distribution in each region, the crack distribution feature anomaly value of the corresponding region is calculated. It is determined whether the crack distribution feature anomaly value of each region is greater than or equal to a preset threshold. If so, it is recorded as a crack concentration region.

[0053] The process for obtaining the outlier values ​​of the crack distribution characteristics is as follows:

[0054] The mobile phone glass cover is divided into several non-overlapping regions of the same size, and the crack distribution feature value F is extracted for each region. i =[L i C i D i ];

[0055] In the formula, L i C represents the total length of all cracks in the i-th region. i D represents the total number of cracks in the i-th region. i F represents the crack density in the i-th region. i This represents the characteristic value of crack distribution in the i-th region;

[0056] K-means clustering analysis was performed on the crack distribution feature values, and K cluster centers were initialized. For each region, the nearest cluster center was found according to the Euclidean distance formula and assigned to the corresponding cluster. The new center point position of each cluster was calculated and updated. The assignment and update process was repeated until the cluster center reached the maximum number of iterations. After obtaining the final clustering results, the average crack distribution feature value was calculated for each cluster. For each region, the sum of squares of the differences between the crack distribution feature value and the average feature value of its cluster was calculated, and the square root was taken to obtain the crack distribution feature outlier value.

[0057] It should be noted that by comprehensively analyzing the characteristic values ​​of crack distribution from multiple dimensions, regions with abnormal crack distributions can be identified more accurately. Combined with the K-means clustering algorithm, crack distribution pattern classification and outlier calculation can be completed automatically without manual intervention. This is suitable for real-time detection on large-scale production lines. The identified abnormal regions can serve as an important basis for improving production processes, helping to reduce the defect rate and improve product quality.

[0058] In S3, based on crack concentration regions, for each crack concentration region, the curvature fluctuation characteristics and crack propagation rate characteristics of the cracks are extracted. These characteristics are analyzed, and based on the analysis results, a comprehensive feature vector is constructed as input to the machine learning model. Based on the model output, the quality of each mobile phone glass cover is determined, specifically including:

[0059] For each crack concentration area, the curvature fluctuation characteristics and crack propagation rate characteristics of the cracks are extracted. Based on the degree of curvature fluctuation, the curvature fluctuation characteristic value is calculated, and based on the crack propagation rate, the crack propagation anomaly characteristic value is calculated. The curvature fluctuation characteristic value and the crack propagation anomaly characteristic value are constructed into a comprehensive feature vector. The comprehensive feature vector of all regions is used as the input of the machine learning model. The output of the model is the quality score of each mobile phone glass cover. Based on the quality score, the quality of each mobile phone glass cover is determined.

[0060] The process for obtaining the curvature fluctuation characteristic value is as follows:

[0061] For every point on the centerline of all cracks, the first and second derivatives of the arc length are approximated using the finite difference method. The curvature of each crack is then calculated using the curvature calculation formula, expressed as follows: Where a represents the number of cracks, x′ a and y′ a The first derivative of the image coordinates, x″ a and y″ a The second derivative of the image coordinates, k a Let represent the curvature of the a-th crack;

[0062] Applying Haar wavelet transform to the calculated curvature sequence decomposes the curvature sequence into different frequency components, resulting in a series of wavelet coefficients. The Haar wavelet transform is represented as a matrix operation: W = H·K, where H represents the Haar wavelet transform matrix, K represents the vector composed of the curvature sequence, and W represents the transformed coefficient vector.

[0063] Based on the wavelet coefficients after wavelet transform, the sum of the standard deviations of all wavelet coefficients is calculated to obtain the curvature fluctuation characteristic value.

[0064] The process for obtaining the abnormal feature value of crack propagation is as follows:

[0065] Collect crack propagation rate measurements at multiple time points or spatial locations, standardize the collected crack propagation rates, and for each standardized data point, find its distance to the B-th nearest neighbor as the distance to the current point; define the set of data points containing the current point and its B-th nearest neighbors as its B-neighborhood.

[0066] For each standardized data point and another data point, calculate the average distance from all points in the B-neighborhood of the standardized data point to the other data point. Divide the number of points in the B-neighborhood of the other data point by this average distance to obtain the reachability density of the other data point relative to the standardized data point. For each standardized data point, calculate the average of the ratios of the reachability density of all points in its B-neighborhood relative to the other data point to the reachability density of the other data point relative to these points to obtain the local outlier factor. Calculate the mean of all local outlier factors to obtain the crack propagation anomaly characteristic value.

[0067] It should be noted that this method achieves intelligent quality assessment of mobile phone glass cover plates by accurately extracting and analyzing the curvature fluctuation characteristics and crack propagation rate characteristics of cracks. Its core technological advantage lies in accurately calculating the curvature fluctuation characteristic values ​​of cracks using the finite difference method and Haar wavelet transform, and evaluating the abnormal crack propagation characteristic values ​​through a local outlier factor algorithm. This constructs a comprehensive feature vector as input to a machine learning model to predict the quality score of each glass cover plate. This method not only improves detection accuracy but also effectively identifies minute defects and crack propagation anomalies that are difficult to detect using traditional methods. The innovation lies in providing a novel multi-dimensional feature extraction and analysis framework by combining mathematical analysis (finite difference method, wavelet transform) with statistical methods (local outlier factor) for intelligent assessment of surface crack quality. This method offers higher accuracy and reliability than traditional single-parameter assessments, and is particularly suitable for the refined analysis of complex crack morphologies.

[0068] The process of constructing the machine learning model is as follows:

[0069] The curvature fluctuation characteristic value and crack propagation anomaly characteristic value of each crack concentration region are constructed into a comprehensive feature vector;

[0070] The comprehensive feature vectors of all regions are used as input to the machine learning model, which is a support vector machine model; the support vector machine model is used to output the quality score of each mobile phone glass cover, and the quality of each mobile phone glass cover is judged based on the quality score.

[0071] The construction process of the support vector machine model is as follows: A linear kernel is selected based on the characteristics of the data to separate samples of different categories in a high-dimensional space; key parameters of the support vector machine are set, including regularization parameters and the selection of the RBF kernel; these key parameters are optimized using cross-validation; a series of mobile phone glass cover samples with known quality scores are collected and labeled to form a training set; the support vector machine model is trained using the training set, and the optimal hyperplane is found by minimizing structural risk, which maximizes the separation of samples of different quality levels; during training, cross-validation is used to adjust the model parameters to ensure good generalization ability; after multiple iterations, the final support vector machine model is obtained; for each newly input mobile phone glass cover, the comprehensive feature vector is input into the trained support vector machine model, which outputs the quality score of the mobile phone glass cover.

[0072] It should be noted that by constructing a comprehensive feature vector and using a support vector machine model to score the quality of each mobile phone glass cover, the entire process from feature vector construction to model training and output quality scores is covered, ensuring the systematic nature and operability of the technical solution. This method not only improves the accuracy of quality inspection but also provides strong support for automated production.

[0073] In S4, based on the judgment results, the quality of each mobile phone glass cover is divided into low quality and high quality, and the low quality mobile phone glass cover is marked, specifically including:

[0074] Determine whether the quality score of each mobile phone glass cover is greater than or equal to a preset threshold. If it is, mark it as a high-quality mobile phone glass cover; otherwise, mark it as a low-quality mobile phone glass cover.

[0075] It should be noted that low-quality glass covers undergo further manual inspection or verification using advanced automated testing equipment to confirm the specific defects. Based on the verification results, these covers are either repaired or scrapped, and relevant data is recorded to improve production processes and quality control procedures. This process not only ensures high-quality standards for outgoing products but also provides data support for continuous production optimization.

[0076] The working principle of this invention is as follows: A high-resolution industrial camera is used to perform an all-around scan of a mobile phone glass cover plate fixed on a testing platform. An appropriate light source system (such as a ring LED) is configured to reduce reflection interference and enhance surface texture features. Raw image data is acquired and preprocessed to ensure the accuracy of subsequent analysis. Based on the preprocessed image data, local threshold segmentation technology is used to identify cracks, and crack concentration areas are determined according to the crack distribution. Local thresholds within each small region are calculated for preliminary binarization, marking the crack location and bounding box. Further, cluster analysis is used to calculate outlier values ​​of crack distribution characteristics, identifying regions with abnormal crack distributions. For each crack concentration area, the curvature fluctuation characteristics and crack propagation rate characteristics of the cracks are extracted. The curvature is approximated using the finite difference method and its Haar wavelet transform to obtain curvature fluctuation characteristic values, and the local outlier factor algorithm is used to calculate crack propagation anomaly characteristic values. A comprehensive feature vector is constructed as the input to a support vector machine model. This model predicts the quality score of each mobile phone glass cover plate based on the knowledge learned from the training set, thereby determining its quality level. Based on quality scores, mobile phone glass covers are categorized into low-quality and high-quality products. Low-quality products are marked and further reviewed manually or using advanced automated equipment. After confirming the specific defects, a decision is made on whether to repair or scrap them. Relevant data is recorded for process improvement. This entire process not only improves inspection efficiency and accuracy but also provides a real-time feedback mechanism for the production line, helping to continuously optimize production processes and improve product quality.

[0077] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for quality inspection of mobile phone glass covers based on image sensors, characterized in that, Includes the following steps: S1: Use an image sensor to scan the phone's glass cover from all angles to acquire raw image data and preprocess the raw image data; The image sensor is an industrial camera; S2: Based on the preprocessed original image data, identify the cracks in the mobile phone glass cover in the original image, and identify the concentrated crack areas according to the crack distribution of the mobile phone glass cover. S3: Based on the crack concentration area, for each crack concentration area, extract the curvature fluctuation characteristics and crack propagation rate characteristics of the crack, analyze the curvature fluctuation and crack propagation rate characteristics of the crack, construct a comprehensive feature vector based on the analysis results, and use it as the input of the machine learning model. Based on the model output, determine the quality of each mobile phone glass cover. S4: Based on the judgment results, classify the quality of each mobile phone glass cover into low quality and high quality, and mark the low quality mobile phone glass cover.

2. The method for quality inspection of mobile phone glass cover plates based on image sensors according to claim 1, characterized in that, The identification of cracks in the mobile phone glass cover in the original image specifically includes: The original grayscale image of the mobile phone glass cover is acquired, and an initial global threshold is calculated. Then, the image is divided into multiple non-overlapping small regions. A local threshold is calculated for each small region. The calculated local threshold is used to perform preliminary binarization processing on each small region to generate a series of partially binarized images. The location and bounding box of each crack are marked in the binarized images.

3. The method for quality inspection of mobile phone glass cover plates based on image sensors according to claim 1, characterized in that, The identification of concentrated crack areas specifically includes: The image is segmented into multiple non-overlapping small regions. Based on the crack distribution in each region, the crack distribution feature anomaly value in the corresponding region is calculated. It is then determined whether the crack distribution feature anomaly value in each region is greater than or equal to a preset threshold. If so, it is recorded as a crack concentration region.

4. The method for quality inspection of mobile phone glass cover plates based on image sensors according to claim 3, characterized in that, The process for obtaining the outlier values ​​of the crack distribution characteristics is as follows: The mobile phone glass cover is divided into several non-overlapping regions of the same size, and crack distribution feature values ​​are extracted for each region. K-means clustering analysis was performed on the crack distribution characteristic values, and K cluster centers were initialized. For each region, the nearest cluster center was found according to the Euclidean distance formula, and it was assigned to the corresponding cluster. The new center point position of each cluster is calculated and updated, and the allocation and update process is repeated until the cluster center reaches the maximum number of iterations; After obtaining the final clustering results, for each cluster, calculate its average crack distribution characteristic value. For each region, calculate the sum of squares of the differences between the crack distribution characteristic value and the average characteristic value of its cluster, and take the square root to obtain the crack distribution characteristic outlier value.

5. The method for quality inspection of mobile phone glass cover plates based on image sensors according to claim 1, characterized in that, Based on crack concentration regions, for each crack concentration region, the curvature fluctuation and crack propagation rate characteristics of the crack are extracted, and the crack curvature fluctuation and crack propagation rate characteristics are analyzed, specifically including: For each crack concentration area, the curvature fluctuation characteristics and crack propagation rate characteristics of the cracks are extracted. Based on the degree of curvature fluctuation, the curvature fluctuation characteristic value is calculated, and based on the crack propagation rate, the crack propagation anomaly characteristic value is calculated. The curvature fluctuation characteristic value and the crack propagation anomaly characteristic value are constructed into a comprehensive feature vector. The comprehensive feature vector of all regions is used as the input of the machine learning model. The output of the model is the quality score of each mobile phone glass cover. Based on the quality score, the quality of each mobile phone glass cover is determined.

6. The method for quality inspection of mobile phone glass cover plates based on image sensors according to claim 5, characterized in that, The process for obtaining the curvature fluctuation characteristic value is as follows: For each point on the center line of all cracks, the first and second derivatives of the arc length are approximately calculated using the finite difference method, and the curvature of each crack is calculated using the curvature calculation formula. Apply Haar wavelet transform to the calculated curvature sequence to decompose the curvature sequence into different frequency components, and obtain a series of wavelet coefficients. Based on the wavelet coefficients after wavelet transform, the sum of the standard deviations of all wavelet coefficients is calculated to obtain the curvature fluctuation characteristic value.

7. The method for quality inspection of mobile phone glass cover plates based on image sensors according to claim 5, characterized in that, The process for obtaining the abnormal feature value of crack propagation is as follows: Acquire and standardize the crack propagation rate dataset, determine the B distance and B neighborhood of each standardized data point, and calculate the reachability density of each data point relative to other points. The local outlier factor for each data point is calculated based on the reachability density. The mean of all local outliers is then calculated to obtain the crack propagation anomaly characteristic value.

8. The method for quality inspection of mobile phone glass cover plates based on image sensors according to claim 5, characterized in that, The determination of the quality of each mobile phone glass cover specifically includes: Determine whether the quality score of each mobile phone glass cover is greater than or equal to a preset threshold. If yes, mark it as a high-quality mobile phone glass cover; otherwise, mark it as a low-quality mobile phone glass cover.

9. The method for quality inspection of mobile phone glass cover plates based on image sensors according to claim 5, characterized in that, The process of constructing the machine learning model is as follows: The curvature fluctuation characteristic value and crack propagation anomaly characteristic value of each crack concentration region are constructed into a comprehensive feature vector; The comprehensive feature vectors of all regions are used as input to the machine learning model, which is a support vector machine model; the support vector machine model is used to output the quality score of each mobile phone glass cover, and the quality of each mobile phone glass cover is judged based on the quality score. A series of mobile phone glass cover samples with known quality scores were collected and labeled to form a training set. The support vector machine model was trained using the training set. During the training process, the model parameters were adjusted using cross-validation techniques. After multiple iterations, the final support vector machine model was obtained. The comprehensive feature vector of each newly input mobile phone glass cover was input into the trained support vector machine model, and the quality score of the mobile phone glass cover was output.

Images