Screen indentation detection method and detection system based on area-array camera

Abstract

The invention discloses a screen indentation detection method based on an area-array camera, and the method comprises the following steps: employing the area-array camera to shoot a to-be-detected screen, and obtaining the global screen image data; preprocessing the collected original image to obtain a potential indentation feature region; performing statistical analysis on the historical indentation samples, and extracting historical indentation features; carrying out region division on the preprocessed image and generating a detection mask in a potential indentation feature region; image information in the mask area is processed, and an indentation area is extracted according to historical indentation features; and establishing a mapping relationship between the OTSU value and the manually judged defect level, generating a visual report, and marking the indentation defect position on the screen.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of defect detection, in particular to a screen indentation defect detection method and system based on machine vision. BACKGROUND

[0002] With the popularity of electronic devices such as smartphones, tablets and laptops, the screen as the main interface for user interaction directly affects the user experience. During the assembly process of electronic device screens, indentation defects are often caused by the extrusion of the mainboard or other components on the back of the screen. Although this indentation defect is weak, it has an important impact on product quality.

[0003] Currently, line-scan cameras are mainly used in industry to detect screens. Line-scan cameras obtain high-resolution images by scanning line by line, which can theoretically capture small indentation defects. However, this detection method has the following problems: Long detection time: Line-scan cameras need to scan the entire screen line by line, resulting in low detection efficiency, which is not suitable for high-capacity production lines; Bar noise interference: Line scanning process is prone to introduce bar noise, affecting detection accuracy; Defect detection limitations: Line-scan cameras have poor detection results for common screen appearance defects such as surface contaminants, scratches, and pits, and cannot achieve comprehensive screen quality detection; High device complexity: Line-scan camera systems require precise scanning mechanisms and synchronization control, resulting in complex device structures and high maintenance costs.

[0004] In addition, existing screen indentation detection techniques generally have the following technical difficulties: Weak indentation difficult to identify: Screen indentation is usually very weak and difficult to observe directly on the surface through conventional methods; Serious background interference: Screen surfaces often have various background interference such as reflections and textures, increasing the difficulty of indentation recognition; Lack of feature modeling: indentation shapes are diverse, lacking systematic feature modeling and quantitative standards; Inconsistent manual detection: Existing detection methods often rely on human experience and lack objective and uniform evaluation criteria.

[0005] Therefore, there is an urgent need to develop an efficient and accurate screen indentation detection method to improve the quality control level of electronic device screens. SUMMARY

[0006] In view of the deficiencies of the prior art, the present application provides a kind of electronic equipment screen defect detection method and system based on machine vision.The present application is particularly suitable for the indentation defect automatic detection in the screen assembly process of smart phone, tablet computer, notebook computer and other consumer electronic products, and can also be applied to the quality control link of display equipment such as liquid crystal display and television panel.The present application combines image processing, computer vision, artificial intelligence and industrial automation technology, realizes the accurate identification and objective evaluation of weak indentation defect through face array camera image acquisition and efficient algorithm processing, and provides technical support for quality control in electronic product manufacturing process.

[0007] The present application provides a kind of screen indentation detection method based on face array camera, comprising the following steps: Step S1: using face array camera to shoot the screen to be detected, and obtaining global screen image data;Face array camera has the characteristics of high resolution and large field of view, and can shoot the image of the entire screen, so as to obtain comprehensive screen information.

[0008] Step S2: preprocessing the collected original image to obtain potential indentation feature area;The purpose of preprocessing is to improve the quality of image and enhance the indentation feature in image, so as to facilitate subsequent analysis and processing.

[0009] Step S3: statistical analysis of historical indentation sample, and extraction of historical indentation feature;Collect a large number of historical indentation samples, which should be representative and cover different types and degrees of indentation. Statistical analysis is carried out on these samples, and parameters such as indentation shape, size, gray value and texture are extracted, which can describe the characteristics of indentation. These historical indentation features will be used as the basis for subsequent detection, and by comparing with the features in the image to be detected, it can be judged whether there is indentation and the type and degree of indentation.

[0010] Step S4: region division is carried out on the preprocessed image, and detection mask is generated in potential indentation feature area;The preprocessed image is divided into different regions, so that each region can be analyzed and processed separately. Detection mask is generated in potential indentation feature area, which is a binary image, in which indentation area is marked as 1 and non-indentation area is marked as 0. By generating detection mask, attention can be focused on potential indentation area, calculation amount can be reduced, and detection efficiency can be improved.

[0011] Step S5: processing the image information in mask area and extracting indentation area according to historical indentation feature;Further processing is carried out on the image information in mask area, such as feature extraction, matching, etc. According to historical indentation feature, corresponding algorithm and model are used to analyze the image in mask area, and indentation area is extracted.

[0012] Step S6: Establish the mapping relationship between OTSU value and artificial defect grade determination, and generate a visual report to mark the indentation defect position on the screen. Establishing the mapping relationship between OTSU value and artificial defect grade determination can automatically determine the indentation defect grade according to the OTSU value. At the same time, a visual report is generated to mark the detected indentation defect position on the screen image, so that the user can intuitively understand the indentation condition of the screen.

[0013] As a further scheme of the present application, in the step S2, the collected original image is preprocessed, including: The color image is converted into a gray-scale image; Gaussian filtering and median filtering are applied to reduce image background noise; and the potential indentation feature area in the image is enhanced through contrast stretching and histogram equalization. The color information of each pixel point in the color image is converted into a single gray value. The conversion method is based on the sensitivity of the human eye to different colors, reduces the data volume of the image, facilitates subsequent processing, and also highlights the brightness information of the image, so that the potential indentation features are more easily detected in the gray-scale image. A weighted matrix is generated using a Gaussian distribution function, and then the matrix is convolved with the image. In the image, the closer the point to the center pixel, the greater the weight, and the farther the point, the smaller the weight, thereby achieving smoothing processing of the image; effectively reducing the Gaussian noise in the image, while reducing noise, the edge information of the image can be well preserved, providing a clearer image basis for subsequent feature extraction and analysis; As a further scheme of the present application, the step S3 specifically includes: statistical analysis of historical indentation samples, extracting indentation area size, shape, and distribution position; establishing an indentation feature library based on the statistical results, including the morphological features and position distribution rules of typical indentations; and generating an indentation priori knowledge model.

[0014] Specifically, it includes: data collection: collect a large number of representative historical indentation sample images. These samples should cover different types, degrees and locations of indentation to ensure the comprehensiveness and accuracy of subsequent analysis. Area size extraction: determine the boundary of each indentation in the image through image analysis techniques such as threshold segmentation, edge detection, etc., and then calculate the area, perimeter and other quantitative indicators of the indentation area to represent the size of the indentation. Shape feature extraction: various methods can be used to describe the shape of the indentation, such as Fourier descriptor, which extracts the shape features of the indentation boundary through Fourier transform; or calculate the moments of the shape, such as central moment, Hu moment, etc. These moments can reflect the geometric properties of the shape, such as circularity, rectangularity, eccentricity, etc., so as to quantitatively describe the shape of the indentation. Distribution position extraction: determine the coordinate position of the indentation in the screen image, or divide the screen into several sub-regions, and count the frequency or probability of the occurrence of the indentation in each sub-region to represent the distribution position features of the indentation. The relative position relationship between the indentation and the screen edge, corner or other specific areas can also be analyzed.

[0015] Establishing an indentation feature library; feature arrangement and classification: arrange the extracted indentation area size, shape and distribution position features, and classify the indentation according to different feature combinations. For example, according to the shape of the indentation, it can be divided into circular indentation, rectangular indentation, irregular indentation, etc. According to the size of the indentation, each category can be further subdivided.

[0016] Typical indentation selection: select representative typical indentation samples from each category. These typical indentations should be able to well represent the main features of this type of indentation, for example, select indentations with regular shape and obvious features as templates for this category.

[0017] Feature library establishment: store the morphological features (such as shape, size, etc.) of typical indentations and the position distribution rules (such as the common appearing area on the screen, the relative position with other components, etc.) in the indentation feature library. The feature library can be managed in the form of a database, which is convenient for subsequent query, update and maintenance.

[0018] Model construction: Based on the data in the indentation feature library, use machine learning, statistics or other modeling methods to construct the indentation priori knowledge model. For example, clustering analysis method can be used to cluster the indentation features, and the clustering center and distribution range of different types of indentation are found out as the basis of the priori knowledge model; or use Bayesian model, according to the feature distribution of historical indentation samples, calculate the probability of indentation under given feature conditions. Model training and optimization: Use a large number of historical indentation samples to train the constructed model, adjust the parameters of the model, so that it can accurately describe the characteristics and distribution rules of indentation. In the training process, cross-validation and other methods can be used to evaluate the performance of the model, and the structure or parameters of the model can be adjusted to optimize the model and improve its generalization ability and accuracy. Model application: The generated indentation priori knowledge model can be used for subsequent indentation detection and analysis. When processing new screen images to be detected, according to the priori knowledge provided by the model, more targeted feature extraction and matching can be carried out, which can improve the efficiency and accuracy of indentation detection, for example, according to the model to predict the possible position and shape of indentation, and focus on detection in these areas.

[0019] As a further scheme of the present application, the specific process of step S4 is: based on priori knowledge, the preprocessed image is divided into regions; a detection mask is generated for high-risk regions where indentation may exist; a multi-scale detection framework is established to adapt to indentation defects of different sizes.

[0020] Specifically: analyze priori knowledge: according to the indentation priori knowledge model established in step S3, understand the common distribution position of indentation on the screen, the relative position relationship with each part of the screen and other information. Select partition method: a variety of ways can be used for regional division. For example, based on spatial clustering algorithm, the image is divided into multiple sub-regions, so that the image features in each sub-region are relatively consistent, while the different sub-regions have obvious differences. It can also be divided according to the structural characteristics of the screen, such as uniformly dividing the screen into several rectangular regions, or dividing according to the functional areas of the screen (such as display area, frame area, etc.). Execute regional division: according to the selected partition method, divide the preprocessed image. The image is divided into multiple smaller sub-regions, so that each region can be analyzed in detail whether there is indentation in the subsequent process.

[0021] Generate a detection mask for high-risk areas where indentations may exist: Determine high-risk areas: Based on prior knowledge and the results of region division, combined with information such as the probability and frequency of indentation occurrence, determine which sub-regions are high-risk areas where indentations may occur. For example, in historical indentation samples, some regions have a higher number of indentation occurrences, or according to the production process and usage of the screen, some regions are more likely to be pressed by external forces and produce indentations. These regions can be determined as high-risk areas.

[0022] Generate a detection mask: For each determined high-risk area, create a corresponding detection mask. The mask is usually a binary image of the same size as the region, where the target area (i.e., the area where indentations may exist) is marked as 1 and the background area is marked as 0. The mask can be generated by various methods, such as manual drawing, automatic generation based on threshold segmentation algorithm, etc. For example, according to the gray scale features or shape features of the high-risk area, set appropriate thresholds to mark the pixels in the region that meet the conditions as target pixels, and generate the mask.

[0023] Establish a multi-scale detection framework to adapt to indent defects of different sizes: Scale selection: According to the distribution of indentation sizes in the indentation prior knowledge model, determine the different scales to be considered. For example, indentations are divided into small, medium, and large scales, corresponding to different pixel areas or length ranges. For small-scale indentations, a smaller detection window or convolution kernel can be selected; for large-scale indentations, a larger detection window or convolution kernel is needed to ensure that indentations of different sizes can be completely detected.

[0024] Construct a multi-scale detection model: Various methods can be used to construct a multi-scale detection framework.

[0025] Fuse multi-scale detection results: Fuse the detection results under different scales to obtain the final indentation detection results.

[0026] As a further scheme of the present application, the specific process of step S5 is: processing the image information in the mask region to remove high-light noise; designing a targeted binarization kernel according to indentation characteristics; Apply various binarization methods to extract the indentation region; calculate the OTSU value of the binarization result as a quantitative indicator of indentation severity.

[0027] As a further scheme of the present application, the specific process of step S6 is: establishing a mapping relationship between the OTSU value and the artificial judgment of the defect level; training an evaluation model according to the judgment results of a large number of sample indentation levels; defining the OTSU threshold standard of indentation defect severity; and realizing an automatic grading function consistent with human eye perception.

[0028] As a further scheme of the present application, the step S2 specifically comprises: converting the RGB image into an 8-bit gray image; applying a Gaussian filter to eliminate random noise; obtaining a background image by using large-scale Gaussian blur, and eliminating the effect of uneven illumination by subtracting the background image from the original image; and improving the image contrast by using the CLAHE algorithm to enhance the visibility of weak indentations.

[0029] As a further scheme of the present application, for the image in the mask region, the specific process of removing highlight noise points by using an adaptive threshold algorithm is as follows: the brightness mean value and standard deviation of the pixels in the mask region are counted, and a highlight threshold is dynamically set to adapt to the brightness distribution of different images; the pixel brightness is compared with the threshold, and the pixels exceeding the threshold are marked as highlight noise points to identify possible interference regions; the detected noise points are smoothed by using a median filter or an interpolation method, and the noise points are replaced by surrounding normal pixels, so that the interference is reduced and the image details are maintained.

[0030] As a further scheme of the present application, the specific process of extracting the indentation region by using multiple binarization methods is as follows: the best segmentation threshold is automatically found by calculating the inter-class variance under different thresholds according to the image gray histogram, so that the discrimination between the foreground and the background is maximized; the threshold is calculated in different local regions, which can adapt to the uneven illumination, so that the indentation region can be effectively segmented even in the case of large brightness variation in different regions; the indentations may have different degrees of gray distribution, and a single threshold may not be able to completely distinguish all indentation regions, so the multi-level threshold method divides the image into multiple gray levels by setting multiple thresholds, thereby more finely distinguishing indentation regions of different severity levels; the OTSU value of the binarization result is calculated as a quantitative indicator of indentation severity.

[0031] The present application also provides a screen indentation detection system based on the screen indentation detection method based on a plane array camera, an image acquisition module, which uses a plane array camera to shoot a screen to be detected to obtain global screen image data; an image preprocessing module, which pre-processes the collected original image; an indentation feature modeling module, which is used for modeling indentation features; an indentation region positioning module, which is used for positioning indentation regions; an indentation feature extraction module, which is used for extracting indentation features; a human eye consistency evaluation module, which is used for realizing automatic grading function consistent with human eye perception; and a result output module, which is used for providing defect level judgment results to guide production process decision-making.

[0032] The present application has the following advantages: the present application solves the problems of low efficiency, large noise interference, limited detection range and other problems in the existing screen indentation detection technology, provides a screen indentation detection method and system based on an area array camera, and achieves the following objectives: improve detection efficiency: by using an area array camera instead of a line scan camera, the image acquisition time is significantly shortened; improve detection accuracy: by using specific image preprocessing and indentation feature extraction algorithms, weak indentations can be effectively identified; establish a unified standard: build an indentation feature library and a human eye consistency mapping to achieve objective quantitative evaluation of indentation defects; enhance detection adaptability: capable of simultaneously handling indentation and other surface defect detection requirements.

[0033] Specifically, improve detection efficiency: an area array camera can take a picture of the entire screen area at once, while a line scan camera needs to scan line by line to obtain the image. In comparison, the area array camera greatly shortens the image acquisition time, thereby improving the detection efficiency. For example, in mobile phone screen detection, using an area array camera can complete image acquisition of the entire screen in a short time, while a line scan camera needs a longer time to scan the entire screen. Especially for large size screens, the efficiency advantage of the area array camera is more obvious.

[0034] Improve detection accuracy: image preprocessing: by converting the RGB image to an 8-bit grayscale image, the data volume is reduced while the basic outline and texture information of the image is retained. Random noise is eliminated by applying Gaussian filtering, a large-scale Gaussian blur is used to obtain a background image, and the effect of uneven illumination is eliminated by subtracting the background image from the original image. The CLAHE algorithm is used to improve the image contrast and enhance the visibility of weak indentations. These preprocessing steps can improve the image quality and provide a better foundation for subsequent indentation feature extraction.

[0035] Indentation feature extraction algorithm: according to the image grayscale histogram, the inter-class variance under different thresholds is calculated to automatically find the best segmentation threshold, so that the discrimination between foreground and background is maximized. The threshold is calculated in different local regions to adapt to uneven lighting and effectively segment the indentation region. For different gray scale distributions that the indentation may have, a multi-level threshold method is used to set multiple thresholds to segment the image into multiple gray levels, and to more carefully distinguish indentation regions of different severity.

[0036] Establish a unified standard: build an indentation feature library to collect indentation images of different types and different severity and their feature data, and establish a mapping relationship between these features and the defect levels determined by the human eye. Through training and analysis of a large amount of sample data, the OTSU threshold standard for indentation defect severity is determined, thereby achieving objective quantitative evaluation of indentation defects. In this way, regardless of the screen or detection environment, the severity of the indentation can be evaluated according to a unified standard.

[0037] Enhanced detection adaptability: The detection method and system can not only detect indentations, but also cope with the detection needs of other surface defects through comprehensive analysis of images. For example, for scratches, stains, bubbles and other defects on the screen surface, these defects can be distinguished and identified from indentations by adjusting and expanding the feature extraction algorithm and classification model, realizing one-stop detection of various surface defects, and improving the versatility and adaptability of the detection system.

[0038] Compared with the prior art, the present application has the following remarkable effects: Improved detection efficiency: The full-screen image is collected by the area array camera at one time, which shortens the detection time by more than 80% compared with the line scanning camera, significantly improving the production line efficiency; Good noise suppression effect: The specific preprocessing algorithm effectively reduces the background noise interference, and the image signal-to-noise ratio is improved by about 40%, avoiding the common strip noise problem of the line scanning camera; High detection accuracy: Based on the indentation feature library and the design of a specific binary kernel, the detection rate of weak indentations reaches more than 95%, and the false positive rate is controlled to be less than 5%; Objective evaluation standard: The human eye consistency mapping model is established, so that the consistency of the automatic detection result and the judgment of the professional engineer reaches more than 90%, realizing the objective quantization of indentation defect evaluation; Strong adaptability: The method can not only detect indentation defects, but also be extended to detect various defects such as stains, scratches and pits on the screen surface, realizing comprehensive detection of screen quality; Convenient system integration: The area array camera system has simple structure and is easy to deploy and maintain on the existing production line, and the comprehensive use cost is reduced by about 30%.

[0039] In order to more clearly illustrate the structural features and effects of the present application, the present application will be described in detail below in combination with the drawings and specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 It is a whole architecture diagram of the screen indentation detection system of the present application; Figure 2 It is a flow chart of the screen indentation detection method; Figure 3 It is a processing flow chart of the preprocessing module; Figure 4 It is a comparison chart of the effects before and after preprocessing; Figure 5 It is a work flow chart of the feature modeling module; Figure 6 It is a work flow chart of the indentation area positioning module; Figure 7 It is a work flow chart of the indentation feature extraction module; Figure 8 It is an indentation area extraction effect diagram; Figure 9 It is a work flow chart of the human eye consistency evaluation module; Figure 10 For detecting effect comparison chart; Figure 11 For performance comparison chart with traditional line scan camera detection method. DETAILED DESCRIPTION

[0041] The application will be further described in conjunction with the drawings and related knowledge, and a clear and complete description will be made. Obviously, the described application is only a part of the embodiments of the application, not all the embodiments.

[0042] The application provides a screen indentation detection method based on a surface array camera, comprising the following steps: collecting an image of a screen to be detected by using a surface array camera; pre-processing the collected image, including gray scale conversion, filter processing and contrast enhancement; establishing an indentation feature library and generating an indentation priori knowledge model; generating an indentation defect detection mask based on the priori knowledge model; designing a specific binary kernel for indentation feature extraction for the image in the mask area; calculating the OTSU value of the binary result as a quantitative index of indentation severity; establishing a mapping relationship between the OTSU value and the artificial judgment defect grade; and outputting the indentation position, range and severity and other detection results.

[0043] The image preprocessing step comprises: converting the RGB image into a gray scale image; applying Gaussian filtering to eliminate random noise; eliminating the influence of uneven illumination by background equalization; and applying the CLAHE algorithm to enhance the image contrast.

[0044] The indentation feature library contains shape features, size features and position distribution features, which are established by statistical analysis of historical indentation samples.

[0045] The design of the binary kernel is based on the morphological features in the indentation feature library, including an elliptical kernel and an irregular closed curve kernel.

[0046] The mapping relationship is established by collecting the grade evaluation results of indentation samples by professional engineers and analyzing the corresponding relationship between the OTSU value and the artificial judgment grade.

[0047] In the application, another screen indentation detection system based on a surface array camera comprises a surface array camera, a light source system, an image acquisition controller, an image processing unit and a display output unit. The surface array camera is a high-resolution industrial camera equipped with a low-distortion lens. The light source system is a ring-shaped LED light source with adjustable angle. The image processing unit is based on GPU acceleration, realizing parallel computation of indentation detection algorithms.

[0048] Example 1: Refer to Figures 1-11As shown, the present application provides a screen indentation detection method based on an area array camera, comprising the following steps: Step S1: using an area array camera to shoot the screen to be detected to obtain global screen image data; using an area array camera to shoot the screen to be detected can collect full-screen images at one time, greatly shortening the image acquisition time compared with a line scanning camera, improving the detection efficiency, and providing global image data basis for subsequent detection work.

[0049] Step S2: preprocessing the collected original image to obtain potential indentation feature area; Specifically, it includes: gray scale transformation: converting a color image into a gray scale image; image filtering: applying Gaussian filtering, median filtering and other algorithms to reduce image background noise; image enhancement: enhancing potential indentation features in the image through contrast stretching, histogram equalization and other methods; converting a color image into a gray scale image reduces the dimension and information quantity of the image, facilitates subsequent processing, and also highlights the brightness information of the image, which is conducive to the display of potential indentation features; applying Gaussian filtering, median filtering and other algorithms. Gaussian filtering can effectively eliminate random noise in the image, making the image smoother; median filtering has a good inhibitory effect on salt and pepper noise and can protect the edge information of the image, further reducing image background noise and improving image quality; through contrast stretching, histogram equalization and other methods, the dynamic range of image gray scale values is expanded, the contrast between potential indentations and the background in the image is enhanced, and the potential indentation features are more obvious, which facilitates subsequent feature extraction and detection.

[0050] Step S3: statistically analyzing historical indentation samples to extract historical indentation features; specifically, statistically analyzing historical indentation samples to extract indentation area size, shape, distribution position and other features; establishing an indentation feature library based on the statistical results, containing morphological features and position distribution rules of typical indentations; generating an indentation priori knowledge model to provide a reference for subsequent detection. Statistically analyzing historical indentation samples to extract the size, shape, distribution position and other features of the indentation area, these features can reflect the typical morphology and distribution rules of the indentation; establishing an indentation feature library based on the statistical results, containing morphological features and position distribution rules of typical indentations, providing a reference model for subsequent detection, so that the detection process can identify indentations according to these priori knowledge; through analysis and feature extraction of historical indentation samples, an indentation priori knowledge model is generated, which can help the algorithm better understand the features of the indentation, so as to more accurately identify the indentation in actual detection.

[0051] Step S4: region division is performed on the preprocessed image, and a detection mask is generated in the potential indentation feature region; specifically, region division is performed on the preprocessed image based on prior knowledge; a detection mask is generated for a high-risk region where indentation may exist; a multi-scale detection framework is established to adapt to indentation defects of different sizes; region division is performed on the preprocessed image based on prior knowledge, and the image is divided into different sub-regions to facilitate more targeted detection; a detection mask is generated for a high-risk region where indentation may exist; the mask can be understood as a binary image, where a region marked as 1 indicates a region that needs to be detected, and a region marked as 0 can be ignored, thus reducing the detection range and computational load and improving detection efficiency; considering that indentations may have different sizes, a multi-scale detection framework is established to better adapt to indentation defects of different sizes and improve the accuracy and comprehensiveness of detection.

[0052] Step S5: image information in the mask region is processed, and an indentation region is extracted according to historical indentation features; specifically, image information in the mask region is processed to remove highlight noise; a targeted binarization kernel is designed according to indentation features; a variety of binarization methods (including but not limited to OTSU and adaptive threshold) are applied to extract the indentation region; the OTSU value of the binarization result is calculated as a quantitative indicator of indentation severity; image information in the mask region is processed by calculating the mean and standard deviation of pixel brightness and setting a dynamic highlight threshold, comparing the pixel brightness with the threshold, marking pixels exceeding the threshold as highlight noise points, and then applying median filtering or interpolation methods for smoothing to replace noise points with normal pixels, thereby reducing interference and preserving image details; a targeted binarization kernel is designed according to indentation features; the kernel is a convolution kernel used for image processing, which can highlight the features of the indentation region and suppress the background region to prepare for subsequent binarization processing; a variety of binarization methods such as OTSU and adaptive threshold are applied to convert the image into a binary image, so that the indentation region and the background region can be clearly separated, thereby extracting the indentation region; the OTSU value of the binarization result is calculated, which can be used as a quantitative indicator of indentation severity; the larger the OTSU value, the greater the difference between the indentation region and the background region, and the more severe the indentation.

[0053] Step S6: Establishing the mapping relationship between OTSU value and artificial defect grade determination, and generating a visual report marking the indentation defect position on the screen. Specifically, it includes: establishing the mapping relationship between OTSU value and artificial defect grade determination; training the evaluation model according to the judgment results of engineers on a large number of sample indentation grades; defining the OTSU threshold standard of indentation defect severity; realizing the automatic grading function consistent with human eye perception; further, outputting the indentation position, range, severity and other detection results; generating a visual report marking the indentation defect position on the screen; providing defect grade determination results to guide production process decision. According to the judgment results of engineers on a large number of sample indentation grades, the mapping relationship between OTSU value and artificial defect grade determination is established, the evaluation model is trained, the OTSU threshold standard of indentation defect severity is defined, and the automatic grading function consistent with human eye perception is realized. The output of indentation position, range, severity and other detection results provides detailed information for production process, helping staff to understand the indentation situation of the screen; a visual report is generated to mark the indentation defect position on the screen, which shows the detection results in an intuitive way, facilitating staff to check and analyze, and providing defect grade determination results as a basis for production process decision, such as deciding whether to repair or scrap the screen.

[0054] Embodiment 2: Refer to Figure 1 As shown in the figure, the present application also provides a screen indentation detection system based on area array camera, specifically including that the image acquisition module obtains the original image and transmits it to the preprocessing module, the preprocessing module preliminarily processes the image and outputs it to the indentation area positioning module, the module generates a detection mask combining the prior knowledge in the indentation feature library and transmits it to the feature extraction module, the feature extraction module inputs the OTSU value to the human eye consistency evaluation module for defect grade determination, and finally the result output module generates a report; Specifically, the area array camera 1: a high-resolution industrial area array camera (at least 15 million pixels) is adopted, equipped with a low distortion lens to ensure sufficient resolution for weak indentation; the light source system 2: an adjustable angle ring LED light source is adopted to enhance the visibility of indentation through specific illumination angle (such as 30° low angle illumination) and reduce screen reflection interference; the image acquisition controller 3: controls the camera exposure parameter, trigger timing and light source brightness to ensure stable image acquisition quality; the image processing unit 4: runs the indentation detection algorithm of the present application, including preprocessing, feature extraction, defect evaluation and other functional modules, specifically including image preprocessing module, indentation feature modeling module, indentation area positioning module and indentation feature extraction module; the display output unit 5: displays the detection results, including indentation position, severity and other information, and generates a quality report.

[0055] In this embodiment, the image acquisition module is composed of a high-resolution industrial area array camera (at least 15 million pixels) with a low-distortion lens, an adjustable angle ring LED light source, and an image acquisition controller. The area array camera is used to acquire full-screen images at one time, and the high-pixel and low-distortion lens ensures the resolution of the weak indentation. The ring LED light source enhances the visibility of the indentation and reduces the interference of the reflection by 30° low-angle illumination. The image acquisition controller controls the camera exposure parameters, trigger timing, and light source brightness to stabilize the image acquisition quality.

[0056] The preprocessing module receives the original image from the image acquisition module, converts the color image to a grayscale image through grayscale transformation, reduces the background noise by using Gaussian filtering, median filtering, and other algorithms, and enhances the potential indentation features by contrast stretching, histogram equalization, and other methods to provide clearer images for subsequent processing.

[0057] The indentation area positioning module combines the prior knowledge in the indentation feature library to divide the image after preprocessing into regions, generates a detection mask for high-risk areas where indentation may exist, establishes a multi-scale detection framework to adapt to indentation defects of different sizes, and accurately locates the areas where indentation may appear.

[0058] The feature extraction module processes the image information in the mask region, removes high-brightness noise, designs a specific binary kernel according to the indentation features, applies OTSU, adaptive thresholding, and other binary methods to extract the indentation area, and calculates the OTSU value of the binary result to provide a quantitative index for defect level determination.

[0059] The human eye consistency evaluation module trains the evaluation model according to the judgment results of engineers on a large number of sample indentation levels, establishes the mapping relationship between the OTSU value and the artificial judgment defect level, defines the OTSU threshold standard of indentation defect severity, realizes the automatic grading function consistent with human eye perception, and makes the detection results more consistent with the actual situation.

[0060] The result output module displays the detection results, including indentation position, range, severity, and other information, generates a visual report and a quality report, and provides a basis for production process decision-making to guide subsequent production operations.

[0061] Embodiment 3: A screen indentation detection method based on an area array camera, comprising the following steps: referring to Figure 2 , image acquisition: the area array camera shoots the screen to be detected with fixed parameters to obtain an original RGB image with a resolution of 4896x3264. The system ensures consistent shooting timing through a trigger signal to ensure the repeatability of the detection; referring to Figure 3 , Figure 4As shown, the preprocessing module processing flow, gray scale conversion: convert RGB image to 8-bit gray scale image; filter processing: apply Gaussian filter (kernel size = 5x5, sigma = 1.5) to eliminate random noise; background equalization: use large-scale Gaussian blur (kernel size = 51x51) to obtain the background image, and eliminate the uneven illumination effect by subtracting the background image from the original image; contrast enhancement: improve image contrast through CLAHE (Contrast Limited Adaptive Histogram Equalization) algorithm, and enhance the visibility of weak indentation. Referring to Figure 5 As shown, indentation feature modeling, statistical analysis of more than 100 groups of historical indentation samples, and extraction of typical indentation features: shape feature: indentation is mostly elliptical or irregular closed curve; size feature: common indentation area is between 3mm² 20mm²; position distribution: indentation mostly occurs in the screen edge area and the corresponding position of internal components; build indentation feature library, including morphological descriptor of indentation and position probability distribution map; generate indentation prior knowledge model to guide the subsequent detection process.

[0062] Referring to Figure 6 As shown, indentation area positioning, based on the prior knowledge model, the image after preprocessing is divided into regions, and the high-risk areas are focused on; generate indentation defect detection mask, this mask mainly includes: screen edge area (area within 10mm from the screen boundary); internal component corresponding position (determined according to product structure diagram); high incidence area of indentation in historical data; establish multi-scale detection framework, use different size of sliding window to scan and detect in the mask area.

[0063] Referring to Figure 7 , Figure 8For the image within the mask region, an adaptive threshold algorithm is applied to remove highlight noise points. The adaptive threshold is calculated: the mean brightness and standard deviation of the pixels within the mask region are calculated, and a dynamic highlight threshold is set to adapt to the brightness distribution of different images. The highlight noise points are detected: the pixel brightness is compared with the threshold, and the pixels exceeding the threshold are marked as highlight noise points to identify possible interference areas. The highlight noise is removed: the detected noise points are smoothed using median filtering or interpolation methods, and the noise points are replaced with normal pixels to reduce interference and preserve image details. According to the indentation feature library, specific morphological kernels (mainly elliptical and irregular closed curve shapes) are designed. Various binarization methods are applied to extract potential indentation regions: OTSU automatic threshold method: the inter-class variance under different thresholds is calculated based on the image gray histogram, and the optimal segmentation threshold is automatically found to maximize the discrimination between the foreground (indentation region) and background (normal region). Adaptive threshold method: the threshold is calculated for different local regions, which can adapt to uneven lighting conditions, so that even in cases of large brightness changes in different regions, the indentation region can be effectively segmented. Multi-level threshold method: indentations may have different degrees of gray distribution, and a single threshold may not be able to completely distinguish all indentation regions. The multi-level threshold method divides the image into multiple gray levels by setting multiple thresholds, thereby more finely distinguishing indentation regions of different severity levels. The OTSU value (threshold size and inter-class variance) of the binarization result is calculated as a quantitative indicator of indentation severity.

[0064] Referring to Figure 9 As shown, human eye consistency evaluation collects the grade evaluation results of more than 200 indentation samples from professional engineers (usually divided into four grades: no defect, slight, moderate, and severe); analyzes the correspondence between OTSU value and artificial judgment grade, and according to the analysis results, a mapping model can be established using linear fitting, segmented threshold, or machine learning method to make the classification results of OTSU value more consistent with the artificial judgment standard, thereby improving the accuracy and reliability of automatic detection. Linear fitting: by performing linear regression on the data of OTSU value and artificial judgment grade, a linear relationship between OTSU value and grade is established, thereby converting OTSU value to the corresponding indentation grade; segmented threshold: according to the distribution of OTSU value, multiple threshold intervals are set, and a segmented function is used to map OTSU value to the corresponding indentation grade to more accurately reflect the indentation characteristics of different grades; machine learning method: by introducing machine learning algorithms such as support vector machine (SVM) or decision tree, OTSU value and other features are used as input to train the model to automatically predict the grade of indentation, thereby improving the accuracy and adaptability of automatic detection; OTSU threshold criteria for determining the severity of indentation defects: OTSU < T1: No defect (level 0); T1 ≤ OTSU < T 2. : Minor defect (level 1); T2 ≤ OTSU < T3: Moderate defect (level 2); OTSU ≥ T3: Severe defect (level 3); where T1, T2, and T3 are thresholds determined through a large number of experiments. Implement an automatic grading function consistent with human eye perception to ensure the consistency of the detection results with the judgments of professionals.

[0065] Result output, mark the detected indentation positions and ranges on the original image; output the OTSU value of each indentation and the corresponding defect level; generate a detection report, including defect statistics and severity distribution; give the product quality determination result (qualified / unqualified) according to the preset criteria to guide the production process decision-making.

[0066] The technical principle of the present invention is described above in combination with specific embodiments, which are only the preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. Any technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. Those skilled in the art can readily conceive of other specific embodiments of the present invention without creative efforts, and these embodiments will fall within the protection scope of the present invention.

Claims

1. A method for screen indentation detection based on an area array camera, characterized in that, The following steps are involved: Step S1: Use an area array camera to shoot the screen to be inspected to obtain a global screen image; Step S2: pre-processing the collected image to obtain potential indentation feature areas; Step S3: Statistically analyzing historical indentation samples and extracting historical indentation features; Step S4: Divide the preprocessed image into regions and generate a detection mask in the potential indentation feature region; Step S5: Processing the image information in the mask area and extracting the indentation area based on the historical indentation features; Step S6: Establish a mapping relationship between the OTSU value and the manually determined defect level, and generate a visual report to mark the indentation defect location on the screen.

2. The screen press mark detection method based on an area array camera of claim 1, wherein, In step S2, the collected original image is pre-processed, including: Convert the color image into a grayscale image; apply Gaussian filtering and median filtering to reduce the image background noise; enhance the potential indentation feature area in the image through contrast stretching and histogram equalization.

3. A screen press mark detection method based on an area array camera as claimed in claim 2, characterized in that, The step S3 specifically includes: statistically analyzing historical indentation samples to extract the size, shape, and distribution position of the indentation area; establishing an indentation feature library based on the statistical results, which includes the morphological characteristics and position distribution patterns of typical indentations; and generating an indentation prior knowledge model.

4. The screen press mark detection method based on an area array camera of claim 3, wherein, The specific process of step S4 is as follows: based on the prior knowledge model, the preprocessed image is divided into regions; a detection mask is generated for high-risk areas where indentations may exist; and a multi-scale detection framework is established to accommodate indentation defects of different sizes.

5. A screen press mark detection method based on an area array camera as claimed in claim 4, characterized in that, The specific process of step S5 is as follows: processing the image information in the mask area to remove highlight noise; designing a targeted binarization kernel according to the indentation characteristics; Apply various binarization methods to extract the indentation area; The OTSU value of the binarized result is calculated as a quantitative indicator of the indentation severity.

6. A screen press mark detection method based on an area array camera as claimed in claim 5, characterized in that, The specific process of step S6 is as follows: establishing a mapping relationship between the OTSU value and the manually determined defect level; training an evaluation model based on the determination results of the indentation levels of a large number of samples; defining the OTSU threshold standard for the severity of the indentation defect; and realizing an automatic grading function consistent with human eye perception.

7. A screen press mark detection method based on an area array camera as claimed in claim 6, characterized in that, The step S2 specifically includes: converting the RGB image into an 8-bit grayscale image; applying Gaussian filtering to eliminate random noise; using large-scale Gaussian blur to obtain a background image, and eliminating the influence of uneven lighting by subtracting the background image from the original image; and improving the image contrast through the CLAHE algorithm to enhance the visibility of weak indentations.

8. A screen press mark detection method based on an area array camera as claimed in claim 7, characterized in that, For the image within the mask area, the specific process of applying the adaptive threshold algorithm to remove highlight noise points is as follows: the brightness mean and standard deviation of the pixels in the mask area are counted, and the highlight threshold is dynamically set to adapt to the brightness distribution of different images; the pixel brightness is compared with the threshold, and pixels exceeding the threshold are marked as highlight noise points to identify possible interference areas; the detected noise points are smoothed by applying the median filter or interpolation method, and the noise points are replaced with surrounding normal pixels.

9. A screen press mark detection method based on an area array camera as claimed in claim 8, characterized in that, The specific process of extracting the indentation area by using multiple binarization methods is as follows: according to the image gray histogram, the between-class variance under different threshold values is calculated to automatically find the best segmentation threshold, so that the discrimination between the foreground and the background is maximized; the threshold value is calculated in different local regions, the image is segmented into multiple gray levels, and thus the indentation areas with different degrees are more finely distinguished; The OTSU value of the binarization result is calculated as a quantitative index of the indentation severity.

10. A screen press mark detection system employing the screen press mark detection method based on an area array camera according to claim 1, characterized by, An image acquisition module is configured to capture the screen to be detected by using a face array camera to obtain global screen image data; an image preprocessing module is configured to preprocess the acquired original image; an indentation feature modeling module is configured to model the indentation feature; An indentation area positioning module is configured to position the indentation area; An indentation feature extraction module is configured to extract the indentation feature; An eye consistency evaluation module is configured to realize the automatic grading function consistent with the human eye perception; and a result output module is configured to provide the defect grade determination result to guide the production process decision.

Images