Panel dust and defect detection method, device, equipment and storage medium

Abstract

The present application discloses a panel dust and defect detection method, device, equipment and storage medium, which relates to the field of image recognition. The AOI machine is used to shoot the illumination image of the panel to be tested under the test light source and the light-free image of the panel to be tested under no light source, and all suspected defect points contained in the illumination image of the panel to be tested are identified; then the illumination sub-image to be tested and the light-free sub-image to be tested are cropped from the two illumination images, and threshold binarization processing is performed to extract the defect contour from the binarized image; the suspected dust points are screened according to the pixel area of ​​the defect contour in the light-free sub-image to be tested, and the defect contours corresponding to the suspected dust points in the two sub-images to be tested are similarly matched, and the dust points and the real defect points are determined according to the matching results. Through the image feature comparison and geometric morphology matching mechanism under multi-light source conditions, combined with the pixel area screening of the defect contour and the cross-light source similarity analysis, the false defects caused by environmental interference and the defects of the panel itself can be effectively distinguished.

Description

Technical Field

[0001] The present application relates to the field of image recognition, and in particular to a panel dust and defect detection method, device, equipment and storage medium. Background Art

[0002] Liquid crystal display (LCD) panels, as a core component of modern information display technology, are widely used in various electronic products, including smartphones, tablets, televisions, and in-car displays. The LCD panel manufacturing process involves multiple complex steps, including coating, etching, developing, panel assembly, die casting and sealing, and driver chip installation. Even the slightest deviation in any of these steps can lead to various defects on the panel, such as point defects, line defects, mura (cloudiness), scratches, edge chipping, burrs, and water droplets. Traditional methods for detecting defects in LCD panels rely primarily on manual visual inspection and automated optical inspection (AOI) technology.

[0003] AOI systems integrate optical, mechanical, electronic control, and software technologies, using machine vision to replace the human eye for automated panel scanning and defect identification. These systems can detect a variety of defects, including bright spots, dark spots, bright lines, dark lines, and mura. In precision manufacturing, such as semiconductor and LCD panel production, production and testing are typically performed in cleanroom environments to minimize the impact of dust and other contaminants on product quality. Cleanrooms utilize high-efficiency particulate matter (HEPA / ULPA) filters and airflow control systems to strictly control airborne particulate matter, temperature, humidity, and pressure, providing a controlled, dust-free environment for the manufacture of precision products.

[0004] However, in real-world applications, even in cleanroom environments, tiny dust particles or foreign matter may still adhere to panel surfaces. During optical inspection, especially when illuminated by specific test light sources, these particles scatter, reflect, or absorb light, resulting in features similar to real defects in the inspection image, such as bright spots, dark spots, and specks. In these situations, AOI systems often struggle to accurately distinguish between these "false defects" caused by dust and "real defects" in the panel itself (such as pixel defects, mura, and scratches). Existing AOI systems typically use a single or fixed light source (such as backlight, coaxial light, or high-intensity line light) for image acquisition. However, the optical characteristics of dust and real defects vary significantly under different lighting conditions. A single or fixed light source cannot fully capture these differences, making it difficult for the system to effectively separate the image features of dust and real defects. For example, in "spotlight inspection" mode, where the panel is illuminated, dust may appear as a shadow or bright spot. However, in the absence of light or under low light conditions at a specific angle, dust may simply appear as a non-luminous foreign object. This lack of information is a major cause of misjudgment. Misjudgments caused by dust interference will cause the AOI system to report a large number of "false defects", thereby reducing detection accuracy and yield, increasing the burden of manual re-inspection, and leading to unnecessary rework or scrap. Summary of the Invention

[0005] The present application provides a panel dust and defect detection method, device, equipment and storage medium, which have the technical effect of improving the accuracy of distinguishing dust from real defects and reducing misjudgment.

[0006] In one aspect, the present application provides a panel dust and defect detection method, the method comprising:

[0007] Using an automated optical inspection (AOI) machine to capture a point-reporting illumination image of the panel under a test light source, and when identifying suspected defect points in the point-reporting illumination image, obtaining a point-reporting dark image of the panel under no light source;

[0008] Based on the suspected defect point, a corresponding illuminated sub-image to be inspected and a corresponding dark sub-image to be inspected are cut out from the reported point illumination image and the reported point dark sub-image, respectively performing threshold binarization processing on the illuminated sub-image to be inspected and the dark sub-image to be inspected, and extracting the defect outline from the binarized image;

[0009] Suspected dust spots are screened based on the pixel area of ​​the defect contour in the light-free sub-image to be inspected, and similarity matching is performed on the defect contours corresponding to the suspected dust spots in the illuminated sub-image to be inspected and the light-free sub-image to be inspected, including calculating the similarity value between the contour shape parameters and the area parameters, and determining the dust spots and the real defect points based on the similarity value matching results; the contour shape parameters include the Hu moment value.

[0010] Specifically, the step of cutting out corresponding illumination sub-images to be inspected and dark sub-images to be inspected from the reported point illumination image and the reported point dark image based on the suspected defect point includes:

[0011] Obtaining a point illumination diagram of the panel under test under different test light sources;

[0012] Polling to identify all suspected defect points in the reporting point illumination map, and establishing a mapping relationship between the suspected defect points and the corresponding reporting point illumination map;

[0013] During the recognition phase, polling the reporting point illumination map, marking all the determined dust points and the real defect points in all the unidentified reporting point illumination maps;

[0014] The suspected defect points at the same position in the unidentified reported point illumination image are filtered according to the determined dust and defect coordinates, and the corresponding illuminated sub-image to be inspected and the dark sub-image to be inspected are cut out based on the remaining suspected defect points.

[0015] Specifically, the threshold binarization process includes:

[0016] Performing grayscale processing on the illuminated sub-image to be inspected and the dark sub-image to be inspected respectively to obtain a grayscale illuminated sub-image and a grayscale dark sub-image;

[0017] All pixels in the grayscale non-photon sub-image and the grayscale illuminated sub-image are selected by polling respectively; for the selected target pixel, a local pixel image is intercepted with the target pixel as the center;

[0018] Calculating a Gaussian weighted average of all pixels in the local pixel map;

[0019] Determine a pixel threshold of the target pixel based on the Gaussian weighted average;

[0020] Comparing all original pixel values ​​of the grayscale photoless sub-image and the grayscale illuminated sub-image with corresponding pixel thresholds to determine the image foreground and image background;

[0021] An image binarization operation is performed on the grayscale non-photon sub-image and the grayscale illuminated sub-image based on the image foreground and the image background to form a binary image.

[0022] Specifically, the process of extracting defect contours includes:

[0023] Scanning the binary image line by line from the edge of the image inward to obtain foreground pixels;

[0024] The first scanned foreground pixel is used as the boundary starting point. The boundary is tracked according to the set tracking direction, and the boundary pixels are scanned one by one. Each pixel has up, down, left, right, and four diagonal lines. When tracking the boundary, starting from the current pixel, the eight directions are checked in sequence to find the next boundary pixel belonging to the contour.

[0025] The coordinates of the boundary pixels are recorded to form a contour point sequence of the defect contour.

[0026] Specifically, screening suspected dust spots according to the pixel area of ​​the defect outline in the dark sub-image to be inspected includes:

[0027] Calculating the area of ​​each defect contour of the dark sub-image to be inspected, or counting the number of foreground pixels in the defect contour;

[0028] When the defect contour area exceeds a set threshold, or the number of foreground pixels exceeds a set number, it is determined to be the suspected dust point;

[0029] When the defect contour area does not exceed a set threshold, or the number of foreground pixels does not exceed a set number, it is determined to be the real defect point.

[0030] Specifically, the contour shape parameters include at least one of the Hu moment value, the perimeter ratio, and the roundness difference, and the area parameters include at least one of the area ratio and the bounding box ratio; at least one of the Hu moment value, the area ratio, the perimeter ratio, the bounding box ratio, and the roundness difference; the smaller the Hu moment value, the more similar the contour shapes are, the area ratio is used to represent the relative difference between the areas of the two contours, the perimeter ratio is used to represent the relative difference between the perimeters of the two contours, and the bounding box ratio is used to represent the relative difference between the width and height of the two minimum circumscribed rectangles for calculating the two contours. The higher the roundness value, the closer the contour shape is to a circle.

[0031] Performing similarity matching on the defect outlines corresponding to the suspected dust spots in the illuminated sub-image to be inspected and the dark sub-image to be inspected, including:

[0032] At least one set of contour shape parameters and area parameters of the defect contours in the two images are calculated respectively, and the corresponding weighted similarity value is calculated according to the weight ratio of the different contour shape parameters and area parameters of the panel to be tested; wherein the contour shape parameters include Hu moment values.

[0033] Specifically, determining the dust point and the real defect point according to the similarity value matching result includes:

[0034] When the weighted similarity value of the illuminated sub-image to be inspected and the dark sub-image to be inspected is lower than a set threshold, the suspected defect point in the reported illumination image is determined to be the real defect point; when the weighted similarity value is not lower than the set threshold, the suspected defect point is determined to be the dust point.

[0035] On the other hand, the present application provides a panel dust and defect detection device, the device comprising:

[0036] The suspected point recognition module is used to capture the reported point illumination image of the panel under the test light source and the reported point dark image without the test light source through the automated optical inspection (AOI) machine, and identify all suspected defect points contained in the reported point illumination image;

[0037] A contour extraction module is used to cut out corresponding illuminated sub-images to be inspected and dark sub-images to be inspected from the reported point illumination image and the reported point dark sub-image based on the suspected defect point, perform threshold binarization processing on the illuminated sub-images to be inspected and the dark sub-images to be inspected, and extract the defect contour from the binarized images;

[0038] The defect confirmation module is used to screen suspected dust points based on the pixel area of ​​the defect contour in the light-free sub-image to be inspected, and perform similarity matching on the defect contours corresponding to the suspected dust points in the illuminated sub-image to be inspected and the light-free sub-image to be inspected, including calculating the similarity value between the contour shape parameters and the area parameters, and determining the dust points and the real defect points based on the similarity value matching results; the contour shape parameters include the Hu moment value.

[0039] On the other hand, the present application provides a computer device, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the panel dust and defect detection method described in the above aspect.

[0040] On the other hand, the present application provides a computer-readable storage medium, which stores at least one instruction, at least one program, code set or instruction set. The at least one instruction, the at least one program, the code set or instruction set is loaded and executed by a processor to implement the panel dust and defect detection method described in the above aspects.

[0041] The beneficial effects brought about by the technical solution provided by the embodiment of the present application include at least: establishing a discrimination model between dust and real defects through the image feature comparison and geometric morphology matching mechanism under multiple light source conditions. By synchronously acquiring image data under the test light source and no light source conditions, combined with the pixel area screening of the defect contour and the cross-light source similarity analysis, it is possible to effectively distinguish between false defects caused by environmental interference and defects in the panel itself, and solve the problem of misjudgment caused by a single light source mode in traditional AOI inspection. At the same time, by accurately identifying real defects, the reliability of production yield statistics is improved, which helps to improve the manufacturing process. This method is applicable to defect detection of various types of liquid crystal display panels and can significantly improve the level of product quality control. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a flow chart of a panel dust and defect detection method provided by an embodiment of the present application;

[0043] Figure 2 It shows a flow chart of polling different reported point illumination maps to screen suspected defect points and crop images;

[0044] Figure 3 A schematic diagram showing the generation of a binarized image using an adaptive threshold binarization algorithm and an Otsu algorithm for a poweroff image is shown;

[0045] Figure 4 A boundary tracking adjacent pixel and direction coding comparison table is shown;

[0046] Figure 5 shows a partial enlarged view of a poweroff diagram;

[0047] Figure 6 A comparative schematic diagram of defect contour extraction based on power-on graph and power-off graph is shown;

[0048] Figure 7 shows a dust and defect detection algorithm flow chart;

[0049] Figure 8 The following is a structural block diagram of a panel dust and defect detection device provided in an embodiment of the present application;

[0050] Figure 9 A structural block diagram of a computer device provided by an exemplary embodiment of the present application is shown. DETAILED DESCRIPTION

[0051] In order to make the objectives, technical solutions and advantages of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0052] In this document, "plurality" refers to two or more. "And / or" describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can mean: A exists alone, A and B exist simultaneously, or B exists alone. The character " / " generally indicates an "or" relationship between the associated objects.

[0053] Figure 1 This is a flow chart of a panel dust and defect detection method provided in an embodiment of the present application, comprising the following steps:

[0054] S1. Using an automated optical inspection (AOI) machine, a point-indicating illumination image of the panel under test is captured. When identifying suspected defect points in the point-indicating illumination image, a point-indicating dark image of the panel under test without light is obtained.

[0055] Automated Optical Inspection (AOI) machines utilize an optical imaging system combined with mechanical motion control. Specifically, they employ high-resolution industrial cameras coupled with controllable light source modules. Through programmable control, they capture images of panels under various lighting conditions (such as backlighting, coaxial lighting, and high-brightness linear lighting). The equipment simultaneously acquires image data under both test lighting and no lighting, providing multimodal image input for subsequent differentiation between dust and real defects.

[0056] The "Power On" and "Power Off" images of the panel surface are captured with the test light source on and off, respectively. These images can be acquired using either line scan or area scan imaging, with the light source switching module enabling rapid switching between these two modes. These images capture the differences in reflection, scattering, and transmission characteristics between dust and actual defects with and without external light, providing fundamental data for feature comparison.

[0057] The AOI machine scans and images the LCD panel under different test light sources. Traditional image inspection algorithms or manual screening can be used to detect all suspected defects on the panel. If no suspected defects are detected in the power-on image, there is no need to obtain a power-off image. However, if suspected defects are detected in any power-on image, all possible suspected defects are extracted, all light sources are turned off, and a power-off image is captured.

[0058] S2. Based on the suspected defect point, the corresponding illuminated sub-image to be inspected and the dark sub-image to be inspected are cut out from the reported point illumination image and the reported point dark sub-image, and the illuminated sub-image to be inspected and the dark sub-image to be inspected are respectively subjected to threshold binarization processing, and the defect contour is extracted from the binarized image;

[0059] For the power-on image where suspected defects are detected, the AOI machine provides bounding box (bbox) information in the image, maps it to the power-off image, and cuts out the illuminated sub-image and the dark sub-image for inspection in both light and dark conditions.

[0060] Threshold binarization converts the cropped illuminated and dark sub-images into black and white binary images. This can be achieved using either an adaptive thresholding algorithm or a global thresholding segmentation method. The segmentation threshold is determined by calculating the grayscale statistics of the local pixel region. This process eliminates interference caused by uneven illumination and highlights the contours of the defect area, facilitating subsequent shape analysis.

[0061] The converted binary image appears in black and white in terms of pixel distribution. Based on the pixel color difference, the boundary information of the target defect area can be identified and tracked from the binary image. The coordinates of the contour points are determined by pixel-by-pixel scanning, and the defect contours in the illuminated sub-image and the dark sub-image can be extracted.

[0062] S3. Suspected dust points are screened based on the pixel areas of the defect outlines in the dark sub-image to be inspected, and similarity matching is performed on the defect outlines of the corresponding suspected dust points in the illuminated sub-image to be inspected and the dark sub-image to be inspected, and the dust points and the real defect points are determined based on the similarity matching results.

[0063] False defects like "dust spots" only require subsequent cleaning; however, true defects, such as scratches, burrs, and clouding, require repair or scrapping. False defects like "dust spots" are typically small, measured in pixels, so their area can be used to initially screen suspected dust spots from larger, true defects. The challenge lies in focusing on identifying those small, suspected dust spots. If they are true defects, they must be accurately identified to avoid impacting production line yield.

[0064] In this embodiment, the similarity matching of small-area suspected dust spots refers to calculation to identify and recognize the feature correlation of the corresponding defect contours in the two images. This feature correlation can be started from the two aspects of contour shape parameters and area parameters. Specifically, it can be implemented by a weighted comparison method of geometric feature combinations such as Hu moment value, area ratio, perimeter ratio, bounding box ratio, and roundness difference. Specifically, one or more parameter control variables can be set according to the accuracy requirements, and a comprehensive evaluation model can be established by setting feature weights. Of course, in order to ensure the importance of calculation, this application sets an important Hu moment value in the contour shape parameters for calculating contour similarity. The role of this matching is to utilize the difference in morphological stability between dust and real defects under different lighting conditions to improve classification accuracy through multi-dimensional feature fusion.

[0065] In a possible implementation, the working process and principle of the solution can be expressed as follows:

[0066] The AOI machine captures images of the panel under test, both under test light and without light. Leveraging the complementary nature of the two lighting modes, the image captures the morphological differences between dust and real defects under different lighting conditions. All suspected defects are identified from the image, ensuring initial coverage of potential defect locations.

[0067] For each suspected defect point, the corresponding illuminated sub-image and dark sub-image are cropped from the reported point's illuminated image and dark sub-image. Local image analysis is used to mitigate global noise interference. The two sub-images are then thresholded and binarized, respectively, to isolate the defect area more clearly based on the grayscale distribution differences under different lighting conditions. Defect outlines are extracted from the binarized images, providing the shape and area features necessary for subsequent differentiation between dust and real defects.

[0068] Suspected dust spots are screened based on the pixel area of ​​the defect outlines in the dark sub-image. This method exploits the differences in area characteristics of dust in the absence of light due to surface scattering or occlusion. A similarity match is performed between the corresponding defect outlines in illuminated and dark conditions. By comparing the correlation between features such as shape and area, dust particles are further distinguished from real defects. Finally, the matching of similarity values ​​is used to accurately classify dust particles from real defects.

[0069] In summary, this application establishes a discrimination model between dust and real defects through image feature comparison and geometric morphology matching mechanism under multiple light source conditions. By synchronously acquiring image data under test light source and no light source conditions, combined with pixel area screening of defect contours and cross-light source similarity analysis, it effectively distinguishes between false defects caused by environmental interference and defects in the panel itself, and solves the problem of misjudgment caused by a single light source mode in traditional AOI inspection. At the same time, by accurately identifying real defects, the reliability of production yield statistics is improved, which helps to improve the manufacturing process. This method is suitable for defect detection of various types of liquid crystal display panels and can significantly improve the level of product quality control.

[0070] In some embodiments, light dust, water droplets, and special defect drops will produce special texture effects under different test light sources. In order to ensure that no suspected defects are missed, this application can also fully cover all reported point illumination maps under different test light sources, test all reported point illumination maps one by one, and screen out all possible suspected defect points. Figure 2 As shown in the flowchart, the solution may include the following steps:

[0071] A. Obtain the illumination diagram of the panel under test under different test light sources;

[0072] B. Poll and identify all suspected defect points in the reporting point illumination map, and establish a mapping relationship between the suspected defect points and the reporting point illumination map to which they belong;

[0073] The polling identification process can adopt a per-image scanning approach or a parallel processing method, such as using multi-threading technology to simultaneously process multiple illumination maps. The mapping relationship is established through a coordinate index table. The location information of each suspected defect point is recorded as three-dimensional coordinate data, including the horizontal and vertical coordinates and the identification code of the illumination map to which it belongs.

[0074] C. In the recognition phase, poll the reported point illumination map and mark all identified dust points and real defect points in all unidentified reported point illumination maps;

[0075] The dynamic marking operation is implemented through a shared storage area. The coordinate information of confirmed dust spots and real defect points is written into the global database. The illumination map of unidentified reported points is automatically associated with the database for coordinate comparison when loading.

[0076] D. Filter the suspected defect points at the same position in the illumination image of the unidentified reported points based on the determined dust and defect coordinates, and cut out the corresponding illumination sub-image and dark sub-image based on the remaining suspected defect points.

[0077] The coordinate filtering process sets a position matching threshold. For example, if the coordinates of a suspected defect point in the unidentified illumination map deviate by less than ±2 pixels from the marked coordinates, it is considered a duplicate detection point and removed. The cropping operation expands the pixel range based on the boundary coordinates of the remaining suspected defect points. For example, a 32×32 pixel area is expanded outward from the coordinates of the defect center point to ensure that the sub-image contains the complete defect features.

[0078] Specifically, when processing illumination maps containing reported points from different test light sources, an automated scanning mechanism first traverses all illumination maps. For example, images generated using three test light sources, 500 lux, 800 lux, and 1000 lux, vary in color temperature, wavelength, and intensity. After suspected defect points are extracted from each image, their coordinate information and the corresponding illumination map number are stored in a relational database, forming a defect location index across all illumination conditions. During subsequent processing, when a defect point in a particular illumination map is confirmed to be dust or a true defect, its coordinate information is immediately synchronized with the inspection process for other unprocessed illumination maps. Before defect recognition is initiated in these unprocessed illumination maps, the coordinate index data is first used for position comparison. If the coordinate overlap between the currently scanned pixel area and a confirmed defect exceeds 95%, defect recognition and sub-image cropping for that area are skipped. This mechanism allows repeated suspected defect points generated at the same physical location under different illumination conditions to undergo full contour extraction and matching analysis only during the initial inspection, effectively eliminating redundant calculations for the same location in subsequent inspection processes. As a result, in multi-light source detection scenarios, computing resource consumption can be reduced by 30%-50%, while the misjudgment rate can be reduced by more than 60%.

[0079] The scheme can also be implemented as follows:

[0080] Obtain the point-reporting illumination maps for the panel under test under different test light sources. First, perform a round-robin analysis of all the point-reporting illumination maps, detecting suspected defect points in each map. For each suspected defect point detected, establish a mapping relationship between it and its corresponding point-reporting illumination map. For example, use a hash table to store the coordinates of the suspected defect point and its corresponding point-reporting illumination map identifier.

[0081] During the recognition phase, each reported illumination map is polled sequentially. For the currently processed illumination map, the coordinates of previously identified dust spots and true defect points are added to the map. This can be achieved by adding a specific color or marker to the image.

[0082] Next, based on the determined dust and defect coordinates, suspected defect points at the same location in the unidentified reported point illumination map are filtered. Specifically, the coordinates of each suspected defect point in the unidentified reported point illumination map are compared with the determined dust and defect coordinate list. If the coordinates of a suspected defect point match the determined coordinates, it is removed from the pending list.

[0083] Finally, based on the remaining suspected defect points, illuminated and dark sub-images are cropped from the corresponding reported point illumination and dark sub-images. During cropping, a fixed-size rectangular area, such as a 30x30 pixel image block, is captured centered on the suspected defect point. These cropped sub-images are used for subsequent defect analysis and confirmation.

[0084] Due to the inherent characteristics of the dark spot image, it is very difficult to visually distinguish light dust against a black background. However, when faced with illuminated and dark sub-images, the traditional global threshold binarization method is prone to inaccurate threshold selection due to brightness differences and noise interference in different areas, which in turn affects the accuracy of defect contour extraction and causes misjudgment of dust and real defects. It is obviously not suitable for this scenario. To this end, this application provides a binarization processing method specifically for light dust images against a black background, including the following steps:

[0085] 1. Perform grayscale processing on the illuminated sub-image to be inspected and the dark sub-image to be inspected respectively to obtain a grayscale illuminated sub-image and a grayscale dark sub-image;

[0086] Convert the cropped color image area to a grayscale image. Perform the operation of converting the RGB three-channel image to a grayscale image, converting the three-channel (red, green, blue) color image to a single-channel grayscale image. When converting the three-channel RGB image to a grayscale image, the following weighted summation formula is used:

[0087] Gray(R,G,B)=0.299×R+0.587×G+0.114×B

[0088] Among them, R, G and B are the pixel values ​​of the red, green and blue channels.

[0089] This weighting is based on the human eye's sensitivity to different colors. The human eye is most sensitive to green, followed by red, and finally blue. Therefore, when converting a color image to a grayscale image, green has the highest weight, followed by red, and blue has the lowest. Grayscaling removes color information while retaining the image's brightness information, which is crucial for subsequent binarization and contour extraction. This is because for defect and dust detection, brightness or contrast differences, not color, are the primary basis for judgment. At the same time, it makes subsequent image processing simpler and reduces complexity. Furthermore, since grayscale images only have one channel, the amount of computation and storage space required for processing are smaller, which can accelerate subsequent image processing and analysis tasks.

[0090] 2. Poll all pixels in the grayscale non-photon sub-image and the grayscale illuminated sub-image respectively; for the selected target pixel, intercept the local pixel map with the target pixel as the center;

[0091] This step mainly uses the poweroff diagram as an example. The poweron diagram is the same.

[0092] For each pixel (x, y) in the poweroff image, the algorithm takes a local area (local in local) of size blockSize×blockSize centered on the pixel.

[0093] 3. Calculate the Gaussian weighted average of all pixels in the local pixel map;

[0094] Calculate the Gaussian weighted mean (mean(x, y)) of the pixels in the local area. The Gaussian weighted mean assigns higher weights to the central pixel and nearby pixels, and decreases the weights from the center point toward the surrounding pixels. This better reflects the brightness characteristics of the local area around the central pixel and smoothes out noise.

[0095] 4. Determine the pixel threshold of the target pixel based on the Gaussian weighted average;

[0096] Calculate the local threshold of the target pixel T(x, y) = mean(x, y) - C, where C is a constant used to fine-tune the threshold to adapt to different contrast requirements, so that it can more sensitively capture weak dust signals.

[0097] 5. Compare the original pixel values ​​in the grayscale photoless sub-image and the grayscale illuminated sub-image with the corresponding pixel thresholds to determine the image foreground and image background;

[0098] 6. Based on the image foreground and image background, perform image binarization operation on the grayscale non-photon sub-image and the grayscale illuminated sub-image to form a binary image.

[0099] Compare the original pixel value with the calculated pixel threshold T(x, y). If the original pixel value is greater than T(x, y), the pixel is determined to be foreground (for example, set to white 255); otherwise, it is determined to be background (set to black 0). Then, a complete binary image is obtained through the binarization operation.

[0100] This local adaptive thresholding process dynamically adjusts the threshold based on the brightness distribution of the local area of ​​the image, rather than a single global adaptive threshold. This effectively separates even very faint, difficult-to-visually-distinguish light dust from the black background, forming a clear binary image. This ensures the integrity of the dust and provides a reliable foundation for subsequent contour extraction and shape matching.

[0101] Figure 3 A schematic diagram showing the generation of a binarized image using an adaptive threshold binarization algorithm and an Otsu algorithm for a poweroff image is shown; Figure 3 We specifically selected an image with noticeable dust; otherwise, the Poweroff Original Image would appear to be a pure black image. The image on the left shows the Poweroff Original Image. The image in the middle shows a dust image detected using the well-known Otsu algorithm. The image on the right shows a dust image detected using the local adaptive threshold binarization algorithm of this embodiment (dust is highlighted with white pixels in all images). Comparing the two binarized image generation schemes, we can see that while the Otsu algorithm in the middle can detect dust spots, its detection accuracy falls far short of the required level and it also misidentifies surrounding noise spots as defects. The scheme provided in this application, on the other hand, can relatively accurately identify dust bands, significantly reducing misidentification rates.

[0102] In general, this method effectively addresses the problem of traditional global thresholding's inability to adapt to local image brightness variations through a dynamic local threshold calculation strategy. For each pixel, the threshold is determined by calculating the Gaussian weighted average of its local region, allowing the binarization process to adapt to the brightness distribution of different regions. This locally adaptive thresholding process accurately separates faint dust from the black background, forming a clear binary image and providing a reliable foundation for subsequent defect contour extraction and shape matching. Furthermore, this method maintains good performance under both illuminated and dark conditions, effectively improving the accuracy of distinguishing dust from real defects and reducing the rate of false positives.

[0103] In one possible implementation, the present application also provides a new contour extraction method, which performs contour extraction on the binarized poweroff image to obtain the defect contour contour1; and performs contour extraction on the binarized poweron image to obtain the defect contour contour2. The following uses the extraction of contour1 as an example. The steps may include the following:

[0104] Scan the binary image line by line from the edge of the image to the inside to obtain the foreground pixels;

[0105] The first scanned foreground pixel is used as the boundary starting point, and the boundary is tracked according to the set tracking direction, scanning the boundary pixels one by one;

[0106] The coordinates of the boundary pixels are recorded to form a sequence of contour points of the defect contour.

[0107] In one possible implementation, you can start scanning the image line by line from the upper left corner, looking for foreground pixels (usually white pixels with a value of 255); find the first foreground pixel as the starting point of the boundary and determine the tracking direction (usually clockwise); in a two-dimensional image, each pixel has 8 adjacent pixels (upper, lower, left, right, and 4 diagonal pixels). When tracing the boundary, you need to start from the current pixel and check the 8 directions in order to find the next pixel belonging to the contour. Figure 4 The boundary tracking direction encoding table shown is sequentially encoded.

[0108] Figure 5 A partial enlarged image of a poweroff image is shown, in which the dust that is difficult to distinguish with the naked eye is accurately outlined after the adaptive binarization and contour extraction of this scheme.

[0109] Similarly, we perform adaptive threshold binarization on the grayscale region of the reported point illumination image (the defects in the reported point illumination image are not as difficult to discern with the naked eye, but the above method is still effective, demonstrating the universality of this adaptive threshold binarization method). The outline of the suspected defect is extracted, denoted as contour2. This outline represents the shape of the suspected defect detected by the AOI machine under specific lighting conditions.

[0110] Figure 6This diagram shows a comparison of defect contours extracted based on power-on and power-off images. The first column visualizes the original illumination image of the reported point detected by the AOI and the defect contour extracted by the algorithm. The second column corresponds to the original power-off image and the extracted power-off dust contour for easy comparison. This example shows that defect images reported by AOI inspection are likely to be over-inspected due to dust rather than actual defects. This can be roughly judged based on pixel area, and then the subsequent contour shape matching module can automatically filter out over-inspected defects.

[0111] The initial judgment process of the defect contour is as follows:

[0112] 1. Calculate the area of ​​each defect contour in the dark sub-image to be inspected, or count the number of foreground pixels in the defect contour;

[0113] 2. When the defect contour area exceeds the set threshold, or the number of foreground pixels exceeds the set number, it is determined to be a suspected dust point;

[0114] 3. When the defect contour area does not exceed the set threshold, or the number of foreground pixels does not exceed the set number, it is determined to be a real defect point.

[0115] This application uses the foreground pixel counting of the Poweroff image to illustrate. This step is to count the number n of white pixels within the outline to reduce the cost of subsequent complex calculations.

[0116] Judgment logic: Count the number n of white pixels in the area enclosed by contour1 extracted from the poweroff map.

[0117] See also Figure 7 The dust and defect detection algorithm flow chart shown in the figure shows that after grayscale processing and adaptive threshold binarization, contour 1 and contour 2 are extracted respectively. For contour 1, the number of white pixels is first counted for preliminary judgment, and then further judgment is made.

[0118] If n is less than the preset threshold (for example, n < 2), this indicates that in the absence of light, there is little to no visible physical foreign matter (dust) detected in the area. This means that the "defect" detected by the AOI machine on the point illumination map is likely a true panel defect (such as a bright spot, dark spot, mura, etc.) rather than caused by dust. In this case, the system directly marks the defect as "Retain this defect," indicating that it is a true defect and is retained, eliminating the need for subsequent complex shape matching.

[0119] If n is greater than or equal to a preset threshold (for example, n>= 2), it indicates that a certain number of white pixels exist in the area in the absence of light, strongly suggesting that the suspected defect may be caused by dust. In this case, the suspected defect is marked as "suspected dust" and enters the next step of the precise identification process.

[0120] This step provides a fast and effective initial filtering mechanism. Suspected defects that don't clearly correspond to physical foreign matter on the Poweroff image can be directly confirmed as real defects, avoiding unnecessary subsequent complex calculations and improving overall processing efficiency. This is based on a core assumption: real defects (such as dead pixels and mura) typically don't have a clear physical presence on the Poweroff image, while dust does.

[0121] For some real defects such as tiny scratches or point defects, additional precise identification is required. Specifically, the defect outlines of the suspected dust spots in the illuminated sub-image and the dark sub-image are matched for similarity. The matching results are determined as follows:

[0122] 1. Calculate at least one set of contour shape parameters and area parameters of the defect contours in the two images respectively, where the contour shape parameters include at least one of Hu moment value, perimeter ratio, and roundness difference, and the area parameters include at least one of area ratio and bounding box ratio;

[0123] 2. Calculate the corresponding weighted similarity value based on the weight ratio of different contour shape parameters and area parameters of the panel to be tested.

[0124] 3. When the weighted similarity value of the illuminated sub-image to be inspected and the dark sub-image to be inspected is lower than the set threshold, the suspected defect point in the reported illumination image is determined to be a real defect point; when the weighted similarity value is not lower than the set threshold, the suspected defect point is determined to be a dust point.

[0125] Among them, the calculation of the Hu moment value is achieved by extracting the Hu invariant moment feature of the contour, which is invariant to translation, rotation and scaling, and can quantify shape similarity. The area ratio and perimeter ratio reflect the relative changes in geometric dimensions by calculating the difference in the ratio of the area and perimeter of the two contours. The bounding box ratio analyzes the extension direction of the contour by comparing the difference in the aspect ratio of the minimum circumscribed rectangle. The roundness difference measures the degree to which the shape is close to a circle by calculating the difference in the ratio of the contour area to the equivalent circle area. The calculation of the weighted similarity value needs to dynamically adjust the weights of each feature according to the actual detection scene. For example, in a scene with a lot of dust particles, the weight of the roundness difference can be set to 0.3 to 0.5, the weight of the bounding box ratio is set to 0.2 to 0.4, and the weight of the Hu moment value is set to 0.1 to 0.3. The above area parameters and contour shape parameters can be set in number and weight ratio according to the accuracy requirements, and this application does not limit this.

[0126] The introduction of Hu moments makes shape similarity judgment more robust, unaffected by translation, rotation, and scaling. The calculation of area ratios and perimeter ratios complements the basis for determining geometric dimensions, helping to capture the differences in optical properties between dust and real defects. Bounding box ratio analysis further distinguishes the extension direction and compactness of contours, facilitating the identification of defects with unique shapes such as scratches. The introduction of roundness differences specifically improves the ability to identify dust particles. Finally, through weighted similarity calculation, dynamic optimization of different features is achieved, allowing the matching process to adapt to the needs of different detection scenarios, thereby comprehensively improving the accuracy and reliability of dust and defect detection.

[0127] By comprehensively evaluating these parameters, the system is able to calculate an overall similarity score between contour1 and contour2, and can determine whether the two contours are similar enough to be considered the same physical entity.

[0128] The comprehensive evaluation of multi-dimensional similarity makes discrimination more robust and accurate. These parameters capture the geometric features of the contour from different angles, making the similarity assessment more comprehensive and accurate, and avoiding the misjudgment that can be caused by a single parameter. For example, a scratch may appear very thin on the poweroff image, but under certain lighting conditions, it may appear as a wide, bright line. Its shape characteristics are significantly different from those of dust particles, and multi-parameter matching can effectively distinguish them. On the other hand, if a dust particle produces a scratch-like defect under certain lighting conditions, but its physical shape (contour1) on the poweroff image and its optical shape (contour2) on the point map are highly similar in terms of Hu moment value, area, perimeter, etc., it can be inferred that they are the same object.

[0129] Final defect identification and filtering:

[0130] "Yes" shape match (defect filtered): If the shape similarity between contour1 and contour2 reaches the preset similarity threshold, the system will determine that the suspected defect in the reported point illumination image is caused by dust detected in the poweroff image. In this case, the defect is identified as "dust" and "filtered out," no longer included in the true defect list. This means it is a false alarm / pass on the AOI machine, eliminating the need for manual re-inspection or rework.

[0131] "No" shape mismatch (retain the defect): If the shape similarity between contour1 and contour2 does not meet the preset similarity threshold, even if a foreign object is present in the poweroff image, its shape does not match the defect shape in the reported point illumination image. This indicates that the defect in the reported point illumination image is not simply the optical manifestation of the foreign object in the poweroff image under different lighting conditions. The system considers the suspected defect to be a true panel defect (for example, pixel defect, mura, scratch, etc.) and will "retain the defect." These retained defects will be the actual defects reported by the AOI machine and require subsequent processing (such as manual review or repair).

[0132] In summary, the beneficial effects brought about by this application include the following:

[0133] a) Multi-light source information fusion: By synchronously acquiring power-on and power-off images, defects at the same physical location can be analyzed in multiple dimensions.

[0134] b) Fine adaptive threshold binarization: In view of the characteristics of the Poweroff image, such as black background, faint dust and low contrast, the local adaptive threshold binarization scheme can extremely finely separate the light dust that is difficult to distinguish with the naked eye from the black background, ensuring the integrity and accuracy of the dust information.

[0135] c) Multi-parameter Contour Shape Precision Matching: By comparing the geometric features of contour 1 in the Poweroff image and contour 2 in the Poweron image, it is possible to accurately determine whether suspected defects are caused by dust. This multi-dimensional matching mechanism avoids the limitations of single-feature judgment, making dust filtering more robust and accurate. This significantly reduces false alarms caused by dust, improves the accuracy of defect identification, and enables the AOI system to more accurately reflect the quality of the panel, thereby improving the final product yield.

[0136] Figure 8 The structure block diagram of the panel dust and defect detection device provided in an embodiment of the present application is shown, and the device includes:

[0137] The suspected point identification module 810 is configured to capture a point illumination image of the panel under test under a test light source using an automated optical inspection (AOI) machine, and obtain a point-light-free image of the panel under test without a light source when identifying suspected defect points in the point illumination image.

[0138] A contour extraction module 820 is configured to clip corresponding illuminated sub-images to be inspected and dark sub-images to be inspected from the reported point illumination image and the reported point dark sub-image based on the suspected defect point, perform threshold binarization processing on the illuminated sub-images to be inspected and the dark sub-images to be inspected, respectively, and extract the defect contour from the binarized images;

[0139] The defect confirmation module 830 is used to screen suspected dust points based on the pixel area of ​​the defect contour in the light-free sub-image to be inspected, and perform similarity matching on the defect contours corresponding to the suspected dust points in the illuminated sub-image to be inspected and the light-free sub-image to be inspected, and determine the dust points and the real defect points based on the matching results.

[0140] It should be noted that the panel dust and defect detection device provided in the embodiment of the present application is only illustrated by the division of the above-mentioned functional modules / functional units. In actual applications, the above-mentioned functions can be assigned to different functional modules / functional units as needed, that is, the internal structure of the panel dust and defect detection device can be divided into different functional modules / functional units to complete all or part of the functions described above. In addition, the implementation method of the panel dust and defect detection method provided in the above-mentioned method embodiment and the implementation method of the panel dust and defect detection device provided in this embodiment belong to the same concept. The specific implementation process of the panel dust and defect detection device provided in this embodiment is detailed in the above-mentioned method embodiment and will not be repeated here.

[0141] Figure 9 The following is a block diagram of the structure of a computer device provided by an exemplary embodiment of the present application. The computer device is a desktop computer, a laptop computer, a PDA, a cloud server, and the like. The computer device may include, but is not limited to, a processor and a memory. The processor and the memory may be connected via a bus or other means. The processor may be a central processing unit (CPU). The processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, graphics processing units (GPU), embedded neural network processors (NPU) or other dedicated deep learning coprocessors, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.

[0142] The processor may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor may be implemented in at least one of the following hardware forms: a DSP (Digital Signal Processing), an FPGA (Field-Programmable Gate Array), or a PLA (Programmable Logic Array). The processor may also include a main processor and a coprocessor. The main processor is used to process data while awake, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor used to process data while in standby mode. In some embodiments, the processor may integrate a graphics processing unit (GPU), which is responsible for rendering and drawing content displayed on the display. In some embodiments, the processor may also include an artificial intelligence (AI) processor, which is used to handle computational operations related to machine learning.

[0143] As a non-transient computer-readable storage medium, the memory can be used to store non-transient software programs, non-transient computer executable programs and modules, such as program instructions / modules corresponding to the methods in the above-mentioned embodiments of the present application. The processor executes various functional applications and data processing of the processor by running the non-transient software programs, instructions and modules stored in the memory, that is, the method in the above-mentioned method embodiment is implemented. The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created by the processor, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0144] In some embodiments, the computer device may optionally include a peripheral device interface and at least one peripheral device. The processor, memory, and peripheral device interface may be connected via a bus or signal lines. Each peripheral device may be connected to the peripheral device interface via a bus, signal lines, or circuit boards. Specifically, the peripheral device includes at least one of a radio frequency circuit, a display screen, and a keyboard.

[0145] The peripheral device interface can be used to connect at least one I / O (Input / Output)-related peripheral device to the processor and memory. In some embodiments, the processor, memory, and peripheral device interface are integrated on the same chip or circuit board. In other embodiments, any one or two of the processor, memory, and peripheral device interface can be implemented on separate chips or circuit boards, although this embodiment is not limited to this.

[0146] The display screen is used to display a UI (User Interface). The UI may include graphics, text, icons, videos, or any combination thereof. When the display screen is a touch screen, it also has the ability to capture touch signals on or above the surface of the display screen. The touch signals can be input as control signals to a processor for processing. In this case, the display screen can also be used to provide virtual buttons and / or virtual keyboards, also known as soft buttons and / or soft keyboards. In some embodiments, there can be one display screen, disposed on the front panel of the computer device; in other embodiments, there can be at least two display screens, disposed on different surfaces of the computer device or in a foldable design; in still other embodiments, the display screen can be a flexible display screen, disposed on a curved or foldable surface of the computer device. Furthermore, the display screen can be configured as a non-rectangular, irregular shape, also known as a special-shaped screen. The display screen can be made of materials such as LCD (Liquid Crystal Display) and OLED (Organic Light-Emitting Diode).

[0147] A power supply is used to power various components in computer equipment. The power supply can be AC, DC, disposable batteries, or rechargeable batteries. When the power supply includes a rechargeable battery, it can be wired or wirelessly rechargeable. A wired rechargeable battery is charged via a wired line, while a wireless rechargeable battery is charged via a wireless coil. The rechargeable battery can also support fast charging technology.

[0148] Those skilled in the art will understand that the structure shown in this embodiment does not constitute a limitation on the computer device, and may include more or fewer components than shown in the figure, or combine certain components, or adopt a different component arrangement.

[0149] The embodiment of the present application also discloses a computer-readable storage medium. Specifically, the computer-readable storage medium is used to store a computer program, and when the computer program is executed by the processor, the method in the above-mentioned method implementation is implemented. Those skilled in the art will understand that all or part of the processes in the above-mentioned method implementation of the present application can be completed by instructing the relevant hardware through a computer program. The program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the implementation of the above-mentioned methods. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory (Flash Memory), a hard disk drive (HDD) or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above-mentioned types of memory.

[0150] This specific embodiment is merely an explanation of the present invention and is not intended to limit the present invention. After reading this specification, those skilled in the art may make non-creative modifications to this embodiment as needed. However, as long as such modifications are within the scope of the claims of the present invention, they are protected by patent law.

Claims

1. A panel dust and defect detection method, characterized in that: The method comprises: Using an automated optical inspection (AOI) machine to capture a point-reporting illumination image of the panel under a test light source, and when identifying suspected defect points in the point-reporting illumination image, obtaining a point-reporting dark image of the panel under no light source; Obtain a point-reporting illumination map of the panel under test under different test light sources, poll all suspected defect points therein, and establish a mapping relationship between the suspected defect points and their corresponding point-reporting illumination maps; poll the point-reporting illumination map during the recognition phase, and mark all confirmed dust points and true defect points in all unidentified point-reporting illumination maps; filter suspected defect points at the same position in the unidentified point-reporting illumination map based on the confirmed dust and defect coordinates, and crop corresponding illuminated sub-maps and no-light sub-maps based on the remaining suspected defect points; performing threshold binarization processing on the illuminated sub-image to be inspected and the dark sub-image to be inspected respectively, and extracting defect contours from the binarized images; Suspected dust spots are screened based on the pixel area of ​​the defect contour in the light-free sub-image to be inspected, and similarity matching is performed on the defect contours corresponding to the suspected dust spots in the illuminated sub-image to be inspected and the light-free sub-image to be inspected, including calculating the similarity value between the contour shape parameters and the area parameters, and determining the dust spots and the real defect points based on the similarity value matching results; the contour shape parameters include the Hu moment value.

2. The method according to claim 1, characterized in that The threshold binarization process includes: Performing grayscale processing on the illuminated sub-image to be inspected and the dark sub-image to be inspected respectively to obtain a grayscale illuminated sub-image and a grayscale dark sub-image; All pixels in the grayscale non-photon sub-image and the grayscale illuminated sub-image are selected by polling respectively; for the selected target pixel, a local pixel image is intercepted with the target pixel as the center; Calculating a Gaussian weighted average of all pixels in the local pixel map; Determine a pixel threshold of the target pixel based on the Gaussian weighted average; Comparing all original pixel values ​​of the grayscale photoless sub-image and the grayscale illuminated sub-image with corresponding pixel thresholds to determine the image foreground and image background; An image binarization operation is performed on the grayscale non-photon sub-image and the grayscale illuminated sub-image based on the image foreground and the image background to form a binary image.

3. The method according to claim 1, characterized in that The process of extracting defect contours includes: Scanning the binary image line by line from the edge of the image inward to obtain foreground pixels; The first scanned foreground pixel is used as the boundary starting point. The boundary is tracked according to the set tracking direction, and the boundary pixels are scanned one by one. Each pixel has up, down, left, right, and four diagonal lines. When tracking the boundary, starting from the current pixel, the eight directions are checked in sequence to find the next boundary pixel belonging to the contour. The coordinates of the boundary pixels are recorded to form a contour point sequence of the defect contour.

4. The method according to claim 1, wherein The screening of suspected dust spots according to the pixel area of ​​the defect outline in the dark sub-image to be inspected includes: Calculating the area of ​​each defect contour of the dark sub-image to be inspected, or counting the number of foreground pixels in the defect contour; When the defect contour area exceeds a set threshold, or the number of foreground pixels exceeds a set number, it is determined to be the suspected dust point; When the defect contour area does not exceed a set threshold, or the number of foreground pixels does not exceed a set number, it is determined to be the real defect point.

5. The method according to claim 4, characterized in that The contour shape parameters also include at least one of a perimeter ratio and a roundness difference, and the area parameters include at least one of an area ratio and a bounding box ratio. The smaller the Hu moment value, the more similar the contour shapes are. The area ratio is used to represent the relative difference in area between two contours. The perimeter ratio is used to represent the relative difference in perimeter between two contours. The bounding box ratio is used to represent the relative difference in width and height of the two minimum circumscribed rectangles used to calculate the two contours. The higher the roundness value, the closer the contour shape is to a circle. Performing similarity matching on the defect outlines corresponding to the suspected dust spots in the illuminated sub-image to be inspected and the dark sub-image to be inspected, including: Calculating at least one set of contour shape parameters and area parameters of the defect contours in the two images respectively, and calculating corresponding weighted similarity values ​​according to the weight ratios of the different contour shape parameters and area parameters of the panel to be tested; The contour shape parameters include Hu moment values.

6. The method according to claim 5, characterized in that The determining of the dust point and the real defect point based on the similarity value matching result includes: When the weighted similarity value of the illuminated sub-image to be inspected and the dark sub-image to be inspected is lower than a set threshold, the suspected defect point in the reported illumination image is determined to be the real defect point; when the weighted similarity value is not lower than the set threshold, the suspected defect point is determined to be the dust point.

7. A panel dust and defect detection device, characterized in that: The device comprises: The suspected point recognition module is used to capture a point illumination image of the panel under test under a test light source using an automated optical inspection (AOI) machine, and obtain a point-light-free image of the panel under test without a light source when identifying suspected defect points in the point illumination image; A contour extraction module is configured to obtain the reported point illumination map of the panel under test under different test light sources, poll all suspected defect points therein, and establish a mapping relationship between the suspected defect points and the corresponding reported point illumination map; poll the reported point illumination map during the recognition phase, and mark all confirmed dust points and true defect points in all unidentified reported point illumination maps; filter suspected defect points at the same position in the unidentified reported point illumination map based on the confirmed dust and defect coordinates, and crop the corresponding illuminated sub-map to be inspected and the dark sub-map to be inspected based on the remaining suspected defect points; performing threshold binarization processing on the illuminated sub-image to be inspected and the dark sub-image to be inspected respectively, and extracting defect contours from the binarized images; The defect confirmation module is used to screen suspected dust points based on the pixel area of ​​the defect contour in the light-free sub-image to be inspected, and perform similarity matching on the defect contours corresponding to the suspected dust points in the illuminated sub-image to be inspected and the light-free sub-image to be inspected, including calculating the similarity value between the contour shape parameters and the area parameters, and determining the dust points and the real defect points based on the similarity value matching results; the contour shape parameters include the Hu moment value.

8. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the panel dust and defect detection method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that The readable storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the panel dust and defect detection method as described in any one of claims 1 to 6.

Images