A glass hole and hidden crack detection method and system using image processing

Abstract

The present application relates to the technical field of image processing, and more particularly to a glass punching hidden crack detection method and system using image processing. It includes: obtaining point cloud data of the target glass, calculating the curvature of each region, and generating a curvature distribution map; based on the distribution map, simulating the light propagation path to obtain the light intensity adjustment coefficient, optimizing the lighting conditions through the adjustable light source device, and collecting the image data set; using Gaussian filtering to process the image data, and then obtaining the hidden crack binary image through binary processing; calculating the length and width indicators of the crack according to the hidden crack binary image, determining the hidden crack position set, and aligning with the curvature distribution map to generate the data; finally, the hidden crack position is fused with the curvature distribution map to generate a preliminary report map, and the final hidden crack recognition result is generated through image synthesis. The present application effectively improves the detection accuracy of hidden cracks, and is particularly suitable for detecting hidden crack defects generated in the laser punching process of photovoltaic panels.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and system for detecting hidden cracks in glass through drilling using image processing. Background Technology

[0002] In the manufacturing process of photovoltaic panels, glass raw materials are melted in a furnace and then undergo multiple processes such as roll forming, cooling and cutting, edge grinding, coating, and screen printing to finally form a photovoltaic panel. In these processes, the panel structure consists of a front panel and a back panel. The back panel requires laser drilling to create lead-out holes for wiring. However, during laser drilling, microscopic cracks are easily generated on the hole walls. These microcracks are usually very small and difficult to detect using conventional testing methods, but they may further expand during subsequent processing or use, leading to photovoltaic panel breakage and seriously affecting product quality and lifespan. Therefore, timely and accurate detection and identification of these microcracks during the manufacturing process is crucial for improving the quality and performance of photovoltaic panels.

[0003] Most existing detection methods rely on visual inspection or conventional crack detection techniques. However, due to the unique characteristics of photovoltaic panels, particularly the curved shape of the glass and the complexity of the drilling process, traditional detection methods often fall short of meeting the requirements for efficient and accurate detection. Especially on curved glass, microcrack features are easily masked by background noise or uneven lighting, leading to missed detections or misjudgments. While existing image processing technologies can detect planar cracks to a certain extent, significant technical challenges remain in detecting complex curved structures and microscopic microcracks.

[0004] Therefore, there is an urgent need for a new detection method that can combine image processing with multiple factors such as illumination and curvature to achieve accurate identification of hidden cracks in curved glass, especially for efficient detection of hidden cracks during the glass drilling process. Summary of the Invention

[0005] This invention provides a method and system for detecting microcracks in glass through drilling using image processing. By combining curvature distribution with image data, the detection accuracy and reliability of microcracks in curved glass are improved, the microcrack identification process is optimized, and the final microcrack identification result is ensured to have high accuracy and clarity.

[0006] In a first aspect, the present invention provides a method for detecting microcracks in glass through drilling using image processing, mainly comprising: Step S1: Acquire point cloud data of the target glass using a 3D scanner, divide the target glass into multiple regions, calculate the curvature of the point cloud data for each region, and obtain the curvature distribution map of the target glass based on the curvature of each region. Step S2: Simulate the light propagation path according to the curvature distribution map, obtain the light intensity adjustment coefficient, adjust the illumination intensity based on the light intensity adjustment coefficient, and collect the image dataset; Step S3: Smooth the noise in the high curvature region of the image dataset using Gaussian filtering to obtain a denoised image set; perform binarization processing on the denoised image set to obtain a hidden crack binary map; Step S4: Calculate the length and width indices of the cracks based on the binary image of the hidden cracks, and determine the set of hidden crack locations based on the length and width indices and align it with the curvature distribution map to obtain superimposed data; Step S5: Based on the superimposed data, fuse the hidden crack location set with the curvature distribution map to obtain a preliminary report map; perform image synthesis on the preliminary report map to generate the final hidden crack identification result.

[0007] As a preferred embodiment of the present invention, step S1 involves acquiring point cloud data of the curved glass and processing the point cloud data using a curvature calculation algorithm to obtain a curvature distribution map, including: Point cloud data of curved glass surface is collected by a 3D scanner. Curvature value of each region is quantified by curvature calculation algorithm. The curved glass is divided into multiple regions according to position coordinate mapping. If the curvature value of a region exceeds a preset threshold, it is marked as a high curvature region. The curvature classification of the region is determined and the curvature distribution map is generated.

[0008] As a preferred embodiment of the present invention, in step S2, the light propagation path is simulated according to the curvature distribution map to obtain the light intensity adjustment coefficient, and an image dataset of the curved glass is collected, including: Based on the curvature classification of the regions in the curvature distribution map, an illumination model is used to simulate the light propagation path based on the curvature classification, and an illumination intensity adjustment coefficient for each region is obtained. The curved glass is then illuminated using an adjustable light source device with the illumination intensity adjustment coefficient, and the original image sequence of the corresponding region is collected to generate the image dataset.

[0009] As a preferred embodiment of the present invention, step S3 involves denoising the image dataset to obtain a denoised image set, including: Based on the curvature classification in the image dataset, a Gaussian filtering algorithm is used to smooth noise in high curvature regions. If the pixel intensity after denoising is lower than a preset threshold, regional connectivity analysis is applied to enhance features, determine the location of hidden crack edges, and generate the denoised image set.

[0010] As a preferred embodiment of the present invention, step S3 involves binarizing the denoised image set to obtain a hidden crack binary image, including: For the denoised image set, binarization is used to separate the hidden crack region from the background. Morphological dilation is used to expand the range of hidden crack texture extraction. If the background noise suppression is insufficient after edge smoothing filtering, the segmentation accuracy is evaluated again to generate the hidden crack binary map.

[0011] As a preferred embodiment of the present invention, in step S4, crack indices are calculated based on the binary image of the hidden crack, the set of hidden crack locations is determined and aligned with the curvature distribution map to obtain superimposed data, including: The crack length and width indices are calculated using the binary image of the hidden crack. If the indices exceed the warning threshold, they are marked as defect areas. The set of hidden crack locations is determined, and the set of hidden crack locations is aligned with the curvature distribution map using position coordinate mapping to generate the superimposed data.

[0012] As a preferred embodiment of the present invention, in step S5, the preliminary report map is obtained by fusing the hidden crack location set and the curvature distribution map based on the superimposed data, including: Based on the superimposed data, the fusion of the hidden crack location set and the curvature distribution map is processed using curvature value overlay layer. Defect labels highlighting high curvature areas are represented by color coding. The overlay transparency is adjusted to optimize the visualization of curvature gradient, and the preliminary report map is generated.

[0013] As a preferred embodiment of the present invention, step S5 involves image synthesis of the preliminary report image to generate the final hidden crack identification result, including: For the preliminary report image, an image synthesis method is used to integrate all the details of the hidden crack location set. If the defect label is not clear in the curvature gradient visualization, background noise suppression and enhancement contrast are applied to determine the generation status of the complete detection report image and generate the final hidden crack identification result.

[0014] Secondly, the present invention also provides a glass drilling microcrack detection system employing image processing to implement the above-mentioned method, the system comprising: The curvature calculation unit is used to acquire point cloud data of the target glass through a 3D scanner, divide the target glass into multiple regions, calculate the curvature of the point cloud data of each region, and obtain the curvature distribution map of the target glass based on the curvature of each region. The image acquisition unit is used to simulate the light propagation path according to the curvature distribution map, obtain the light intensity adjustment coefficient, and the adjustable light source device adjusts the illumination intensity based on the light intensity adjustment coefficient and acquires the image dataset. The image processing unit is used to smooth the noise in the high curvature region of the image dataset using Gaussian filtering to obtain a denoised image set; and to perform binarization processing on the denoised image set to obtain a hidden crack binary image. The alignment processing unit is used to calculate the length and width indices of the cracks based on the binary image of the hidden cracks, and determine the set of hidden crack locations based on the length and width indices and align it with the curvature distribution map to obtain superimposed data. The hidden crack identification unit is used to fuse the hidden crack location set and the curvature distribution map according to the superimposed data to obtain a preliminary report map; and to perform image synthesis on the preliminary report map to generate the final hidden crack identification result.

[0015] Thirdly, the present invention also provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the above-described method.

[0016] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention acquires point cloud data of a glass surface using a 3D scanner and generates a curvature distribution map using a curvature calculation algorithm. This map accurately reflects the geometric features of various regions of the glass, providing reliable data support for subsequent hidden crack detection. Based on the curvature distribution map, the illumination intensity is adjusted using an illumination model to optimize image acquisition conditions, especially in high curvature areas, ensuring consistent and accurate image quality. Next, after denoising the image dataset, binarization is used to clearly separate the hidden crack area from the background, effectively enhancing the identifiability of the hidden crack. Furthermore, the length and width of the crack are calculated to accurately locate the hidden crack position and align it with the curvature distribution map to generate overlay data. This data integrates the geometric features of the hidden crack with the curvature information of the glass surface, ensuring spatial matching between the hidden crack and curvature, thus providing a foundation for generating a high-precision hidden crack report map. Finally, image synthesis technology is used to fuse the hidden crack position with the curvature distribution map, optimizing the visualization effect of defect annotation. Transparency adjustment and noise suppression techniques are used to ensure the clarity and accuracy of the report map. The combination of the above technical solutions improves the accuracy and robustness of microcrack detection in curved glass, enabling accurate identification of minute cracks generated during glass drilling, thus meeting the quality control requirements in the photovoltaic panel manufacturing process. Attached Figure Description

[0017] Figure 1 This is a flowchart of a glass drilling microcrack detection method using image processing in an embodiment of the present invention; Figure 2 This is a structural diagram of a glass drilling microcrack detection system using image processing, as described in an embodiment of the present invention. Detailed Implementation

[0018] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0019] like Figure 1 This embodiment describes a method for detecting microcracks in glass through drilling using image processing, which may specifically include: Point cloud data of the target glass is acquired using a 3D scanner. The target glass is divided into multiple regions, and the curvature of the point cloud data for each region is calculated. Based on the curvature of each region, a curvature distribution map of the target glass is obtained. Specifically, this includes: Point cloud data of the target glass surface is collected by a 3D scanner. The curvature value of each region is quantified by a curvature calculation algorithm. The curved glass is divided into multiple regions according to the position coordinate mapping. If the curvature value of a region exceeds a preset threshold, it is marked as a high curvature region. The curvature classification of the regions is determined, and the curvature distribution map is generated.

[0020] Specifically, in the photovoltaic panel manufacturing process, glass raw materials are melted in a kiln and then sequentially processed through roll forming, cooling, cutting, edge grinding, coating, and screen printing to form photovoltaic panels. The panel structure includes two parts: the front panel and the back panel. The back panel needs to be formed with lead wire holes through laser drilling. However, microscopic crack defects are easily generated on the hole walls during laser drilling, especially in curved photovoltaic glass. These defects may expand during subsequent processing or use, leading to panel breakage and seriously affecting product quality and service life. This embodiment provides a glass drilling crack detection method using image processing. It adopts a combination of image processing and curvature calculation techniques, which is particularly suitable for specific scenarios of glass drilling crack detection. In this method, point cloud data of the curved glass is first collected by a 3D scanner to ensure that high-precision geometric information of the glass surface can be obtained.

[0021] Specifically, a 3D scanner deployed around the perforated area of ​​the glass projects light using laser or structured light to collect point cloud data of the target curved glass surface. The obtained point cloud data provides the basic information for subsequent curvature calculation. Then, the collected point cloud data is processed using a curvature calculation algorithm to quantify the curvature value of each region. Specifically, the least squares method is used to fit the local surface to remove noise points in the point cloud data, and the Gaussian curvature of each point is further calculated. This is done by fitting a quadratic surface and calculating the product of the principal curvatures to quantify the degree of curvature of each region, thus obtaining the Gaussian curvature value, which can effectively reflect the local geometric characteristics of the curved glass.

[0022] Subsequently, the curved glass is divided into multiple regions based on the position coordinate system, and the point cloud data in each region is classified based on the curvature value. According to the calculated curvature distribution map, each region is classified. If the curvature value of a region exceeds a preset threshold, the region is marked as a high curvature region. This threshold can be set according to the material properties of the glass and the design standards of the curved surface. High curvature regions usually mean that there may be a large stress concentration, so they need to be paid special attention to.

[0023] After marking the high curvature regions, the entire curved glass surface is then divided into regions. The point cloud data is projected onto a two-dimensional plane using a Cartesian coordinate system, and the regions are divided according to a preset grid size. This transforms the complex three-dimensional data into more manageable and processable two-dimensional data, and provides a clear structure for subsequent curvature classification. Then, based on the classification results among the marked high curvature, medium curvature, and low curvature regions, a curvature distribution map can be generated. This curvature distribution map uses color gradients to represent different curvature regions. High curvature regions are usually highlighted with dark or red colors, while low curvature regions are represented by lighter colors.

[0024] The above technical solution can obtain a curvature distribution map that clearly reflects the geometric characteristics of different regions of curved glass. Combining the geometric data and image data of curved glass, it not only provides accurate curvature information for subsequent microcrack detection, but also provides data basis for analyzing the relationship between microcracks and curvature distribution.

[0025] Step S2: Simulate the light propagation path based on the curvature distribution map, obtain the illumination intensity adjustment coefficient, adjust the illumination intensity of the adjustable light source device based on the illumination intensity adjustment coefficient, and collect the image dataset; specifically including: Based on the curvature classification of the regions in the curvature distribution map, an illumination model is used to simulate the light propagation path based on the curvature classification, and an illumination intensity adjustment coefficient for each region is obtained. The curved glass is then illuminated using an adjustable light source device with the illumination intensity adjustment coefficient, and the original image sequence of the corresponding region is collected to generate the image dataset.

[0026] Specifically, the propagation path of light is simulated based on the curvature distribution map, and an illumination intensity adjustment coefficient is obtained to optimize the image dataset acquisition process. Firstly, based on the curved glass point cloud data acquired by a 3D scanner, the curvature value of each region is calculated, and the regions are classified into low, medium, and high curvature regions according to a preset curvature threshold. Next, based on these curvature classifications, an illumination model is used to simulate the propagation path of light in each region. Specifically, the Phong illumination model is used to calculate the incident angle of light in different curvature regions, and the illumination intensity is adjusted according to the curvature changes. Especially for high curvature regions, the specular reflection component increases to simulate the bending effect of light, ensuring illumination uniformity. Through iterative calculations, deviations in the light path are simulated, and the illumination intensity attenuation rate of each region is calculated accordingly, thereby obtaining the illumination intensity adjustment coefficient for each region. This ensures that the illumination intensity in high curvature regions is improved to compensate for possible scattering losses during light propagation.

[0027] After obtaining the illumination intensity adjustment coefficient, an adjustable light source device, such as an LED array light source, is used to illuminate each area of ​​the curved glass according to the calculated adjustment coefficient. Specifically, the intensity of the light source is adjusted in real time according to the adjustment coefficient of each area to ensure that the illumination intensity of each area matches its curvature value, thereby optimizing the image acquisition effect. At this time, a camera is used to acquire image sequences of each area and form a dataset. The image sequence includes multiple images to capture dynamic changes under different illumination conditions, which can significantly improve the brightness uniformity of images in high curvature areas and effectively reduce image noise. This provides higher quality image data for subsequent denoising processing and hidden crack detection. Furthermore, by analyzing the image dataset, potential hidden crack areas can be detected, especially in high curvature areas, improving the accuracy and efficiency of hidden crack detection.

[0028] The above technical solution ensures that the image quality of each area can be effectively improved during the detection of microcracks in curved glass. Furthermore, by dynamically adjusting the lighting conditions, the problem of uneven lighting in high curvature areas can be reduced, providing more reliable data support for subsequent microcrack detection, noise reduction, and other image processing.

[0029] Step S3: Denoise the image dataset to obtain a denoised image set; binarize the denoised image set to obtain a hidden crack binary map; In step S3, the image dataset is denoised to obtain a denoised image set, including: Based on the curvature classification in the image dataset, a Gaussian filtering algorithm is used to smooth noise in high curvature regions. If the pixel intensity after denoising is lower than a preset threshold, regional connectivity analysis is applied to enhance features, determine the location of hidden crack edges, and generate the denoised image set.

[0030] Specifically, in this embodiment, in order to achieve accurate detection of microcracks caused by drilling in photovoltaic curved glass, the obtained image dataset is first denoised. Based on the curvature classification in the image dataset, a Gaussian filtering algorithm is used to smooth the noise in high curvature regions. Specifically, the Gaussian filtering algorithm performs a weighted average of the pixels in the image through convolution operations. The standard deviation of the Gaussian function used determines the degree of smoothing. Especially for high curvature regions, these regions usually amplify noise due to uneven light reflection. Therefore, by selecting a Gaussian kernel with an appropriate standard deviation for filtering, important edge details can be preserved while smoothing noise, reducing the impact of noise and ensuring the accuracy of subsequent microcrack identification.

[0031] After Gaussian filtering, if the pixel intensity after denoising is lower than a preset threshold, it indicates that the image may have excessively smoothed features. In this case, region connectivity analysis is applied to further enhance the features. Region connectivity analysis uses the flood fill algorithm to check the neighborhood connections between pixels in the image, identify and enhance isolated feature points. For smaller connected regions, morphological operations such as dilation are used to enhance them, thereby connecting discontinuous edges and ensuring that the hidden crack features in high curvature regions can be effectively extracted and presented in a complete form, reducing the possibility of missed detections.

[0032] The above technical solution, through precise noise reduction and feature enhancement, combined with regional connectivity analysis and edge detection technology, ensures the efficiency and reliability of the microcrack detection method in complex curved glass, and is especially suitable for the accurate identification of small cracks such as microcracks caused by glass drilling.

[0033] Further, in step S3, the denoised image set is binarized to obtain a hidden crack binary image, including: For the denoised image set, binarization is used to separate the hidden crack region from the background. Morphological dilation is used to expand the range of hidden crack texture extraction. If the background noise suppression is insufficient after edge smoothing filtering, the segmentation accuracy is evaluated again to generate the hidden crack binary map.

[0034] Specifically, in the detection of microcracks in curved glass, denoised image sets often suffer from background noise and low contrast, making it difficult to effectively distinguish the microcrack area from the background. To accurately extract the microcrack texture, traditional segmentation methods may not be effective in identifying minute cracks on complex curved surfaces. Therefore, this paper proposes a binarization process to separate the microcrack area from the background, simplifying subsequent processing steps. Binarization can divide pixel values ​​in the image into foreground and background by setting a threshold, thus clearly identifying the microcrack area. For the detection method of microcracks caused by glass drilling, the denoised image set is first binarized to effectively separate the microcrack area from the background. Specifically, a suitable threshold is used to compare the pixel values ​​in each denoised image. If the pixel value is higher than the threshold, it is determined to be the foreground, i.e., the hidden crack region; if the pixel value is lower than the threshold, it is determined to be the background region, thus achieving preliminary separation of the hidden crack region. However, the curvature of the curved glass and uneven lighting may cause edge blurring, affecting the binarization effect. Therefore, morphological dilation operation is used to expand the extraction range of the hidden crack texture and connect the broken texture fragments, thereby improving the integrity and accuracy of texture extraction. This dilation operation uses structuring elements such as circular or square kernels to process the binarized image. By placing the structuring elements around each foreground pixel and filling its neighborhood, the boundary of the hidden crack region is expanded, thereby improving the coverage and continuity of the hidden crack texture. Especially under the complex geometry of photovoltaic curved glass, it effectively connects those broken texture fragments caused by the curved shape.

[0035] Furthermore, to further optimize the detection effect of hidden crack regions, if the background noise suppression effect is insufficient after image edge smoothing filtering following dilation, repeated segmentation accuracy evaluation of the image is required. During implementation, edge smoothing filtering techniques, such as median filtering, are first applied to suppress noise and smooth edges, thereby reducing false detections. Next, the pixel variance of the background region of the filtered image is calculated. If this variance value is higher than a preset threshold, the noise suppression effect is deemed insufficient. The entire segmentation process is repeatedly optimized by adjusting the threshold of binarization processing and the parameters of the dilation operation, such as the number of dilation iterations. Through the above iterative optimization, the noise variance is further reduced, and image clarity and the positioning accuracy of hidden cracks are improved. Uneven illumination or other factors may increase noise; therefore, during optimization, the evaluation criteria are adjusted by combining the curvature distribution map to ensure the smoothness and segmentation accuracy of high-curvature regions, significantly improving detection robustness and reducing the marking of false positive defects.

[0036] The above technical solution can accurately highlight the hidden crack area through the generated binary image of the hidden crack, providing a reliable basis for subsequent hidden crack location and report generation.

[0037] Step S4: Calculate crack indices based on the binary image of the hidden crack, determine the set of hidden crack locations, and align it with the curvature distribution map to obtain superimposed data; specifically including: The crack length and width indices are calculated using the binary image of the hidden crack. If the indices exceed the warning threshold, they are marked as defect areas. The set of hidden crack locations is determined, and the set of hidden crack locations is aligned with the curvature distribution map using position coordinate mapping to generate the superimposed data.

[0038] Specifically, in this embodiment, crack indices are calculated based on the binary image of the hidden crack and aligned with the curvature distribution map to determine the set of hidden crack locations and generate overlay data, thereby achieving accurate detection of hidden cracks in curved glass. First, the geometric features of the hidden crack region are extracted through the binary image of the hidden crack. Specifically, this includes calculating the length and width indices of each hidden crack, that is, extracting the pixel chain of each crack from the binary image, calculating the pixel length of the crack chain as the crack length index, and measuring the average pixel width perpendicular to the chain as the width index. If the calculated length and width indices exceed a preset warning threshold, the corresponding area is marked as a defect area, further determining the set of hidden crack locations, thereby providing accurate location information for subsequent defect analysis.

[0039] A positional coordinate mapping technique is employed to precisely align the set of hidden crack locations with a pre-calculated curvature distribution map. First, the two-dimensional coordinates of the hidden crack location set are converted into three-dimensional spatial coordinates. An affine transformation matrix is ​​then used to ensure that the hidden crack locations are accurately superimposed on the high curvature regions of the curvature distribution map. Specifically, for areas with high curvature, such as the curved edges of glass, the curvature gradient vector is calculated and the projection angle of the hidden crack locations is adjusted to align the two-dimensional coordinates of the hidden cracks with the three-dimensional coordinate system of the curvature distribution map. These processing steps effectively reduce mismatch errors caused by the geometric characteristics of the curved surface and improve the accuracy of the correlation analysis between the hidden crack locations and the curved surface features. This is particularly helpful in identifying high-risk areas in scenarios with complex curved surfaces.

[0040] For low curvature regions, if the microcrack features are relatively small, a linear interpolation method is used to directly map the coordinates of the microcrack location set to the corresponding grid in the curvature distribution map, ensuring that the alignment accuracy is within the set range. This not only improves the recognition accuracy in high curvature regions but also expands the applicability of the detection method and enhances its universality in glass with different curvatures. Finally, through the alignment results, the microcrack location set and the curvature distribution map are fused to form an overlay data structure containing location and curvature labels, which facilitates subsequent visualization processing and analysis.

[0041] The above technical solution not only ensures the integrity of microcrack detection, but also provides reliable data support for the subsequent generation of microcrack reports.

[0042] Step S5: Based on the superimposed data, fuse the hidden crack location set with the curvature distribution map to obtain a preliminary report map; perform image synthesis on the preliminary report map to generate the final hidden crack identification result.

[0043] In step S5, the hidden crack location set and the curvature distribution map are fused based on the superimposed data to obtain a preliminary report map, including: Based on the superimposed data, the fusion of the hidden crack location set and the curvature distribution map is processed using curvature value overlay layer. Defect labels highlighting high curvature areas are represented by color coding. The overlay transparency is adjusted to optimize the visualization of curvature gradient, and the preliminary report map is generated.

[0044] Specifically, in this embodiment, the method for detecting microcracks in glass drilling utilizes image processing technology combined with curvature analysis to achieve efficient and accurate microcrack identification. Based on the aforementioned overlaid data, the microcrack location set and curvature distribution map are fused to generate a preliminary report image. Specifically, a curvature value overlay layer is first used to process the fusion of the microcrack location set and the curvature distribution map. In this process, the curvature value overlay layer, as an image processing layer, multiplies the binary markers of the microcrack locations with the curvature data. Through weighted processing, the microcrack locations in high curvature areas are highlighted. The generation of this weighted layer makes the markings of microcracks in high curvature areas more prominent, thus facilitating the identification and location of potential microcrack defects. Next, to further enhance the visualization effect of the report image, a color coding scheme is used to highlight defect labels in high curvature areas, particularly using red coding to display high curvature areas with curvature values ​​exceeding a threshold, and overlaying microcrack location markers on these areas. To ensure the visualization of defects, the average curvature value of each high curvature area is further calculated, and the saturation of the label is adjusted according to the preset threshold to visually enhance the defect identification of high curvature areas, making these defects more conspicuous in the final report image.

[0045] Furthermore, to optimize the visualization of curvature gradients, the transparency of the overlay layers was adjusted. Specifically, the transparency value for each region was determined by calculating the curvature gradient difference. Transparency is proportional to the reciprocal of the curvature gradient, ensuring that lower transparency is set in high-gradient areas to reveal underlying details. This process effectively prevents high-curvature areas from obscuring underlying details, making hidden cracks visually clearer. The transparency adjustment ensures sharper edges of hidden cracks in curved areas with significant curvature changes, thereby improving detection efficiency. In the final preliminary report image, the curvature gradient with optimized transparency clearly displays the defect area, while enhancing contrast, making the location, size, and severity of hidden cracks more explicit.

[0046] The above technical solution integrates defect annotations for all hidden crack locations and high curvature areas in the final generated preliminary report image and optimizes the curvature gradient, which significantly improves the accuracy of hidden crack identification and the visualization effect of the report image. This provides accurate and clear data support for subsequent hidden crack analysis and quality control. This comprehensive method, through the combination of image processing and curvature analysis, can effectively solve the difficulties in the detection of hidden cracks in curved glass and provides an efficient and reliable technical solution for the detection of hidden cracks in glass drilling.

[0047] Further, in step S5, the preliminary report image is synthesized to generate the final hidden crack identification result, including: For the preliminary report image, an image synthesis method is used to integrate all the details of the hidden crack location set. If the defect label is not clear in the curvature gradient visualization, background noise suppression and enhancement contrast are applied to determine the generation status of the complete detection report image and generate the final hidden crack identification result.

[0048] Specifically, in this embodiment, for the preliminary report image, an image synthesis method is used to integrate all details of the hidden crack location set to generate the final hidden crack identification result. First, a pixel-level fusion algorithm is used to overlay the crack length, width, and location coordinates from the hidden crack location set with the curvature distribution layer of the preliminary report image to ensure that details of each defect area, such as edge contours and curvature values, are completely preserved, forming a synthesized image. During this process, the hidden crack data is matched pixel-by-pixel with the corresponding positions in the curvature distribution map, thereby achieving high-precision image synthesis. To ensure that the detail integration does not lose the original curvature gradient information, an averaging method is used to further optimize the fusion effect of the image data after overlay.

[0049] During the synthesis process, if the defect labels in the curvature gradient visualization are unclear, it may be due to noise interference. Therefore, background noise suppression methods are applied to enhance image contrast. Here, a median filtering algorithm is used to smooth noise, specifically by replacing each pixel value with the median of its neighboring pixel values, thereby effectively removing isolated noise points. Furthermore, to improve the clarity of the labels in the image, a histogram equalization method is used to enhance contrast. This process involves statistically analyzing the gray-level histogram of the image and calculating the cumulative distribution function to map it to the range of 0-255, enhancing the distinction between defect labels and the background. Through these processing steps, the visualization effect of defect areas can be significantly improved, especially in high curvature areas, avoiding misjudgments caused by noise.

[0050] After image synthesis and noise suppression, it is determined whether the synthesized image contains all hidden crack details and clear labels. Specifically, the coverage of the hidden crack location set in the synthesized image is verified, and the label sharpness variance is checked. If the coverage reaches 100% and the label sharpness variance exceeds a preset threshold, the synthesized image is considered complete, and the final hidden crack identification result is generated. If unclear or missing defect labels are detected, the process returns to the iterative enhancement step, where noise suppression and contrast enhancement are applied again to ensure the accuracy and completeness of the final report image.

[0051] In scenarios involving the inspection of diverse curved glass surfaces, such as low-curvature planar regions and medium-curvature curved regions, targeted noise suppression parameter adjustments are employed. Small-neighborhood filtering is used in low-curvature regions to preserve more subtle details, while large-neighborhood filtering is used in medium-curvature regions to enhance the algorithm's robustness. Through these technical measures, the final output of the microcrack identification result can adapt to glass samples with different curvatures, improving the versatility and accuracy of the inspection, and supporting subsequent quality control and decision analysis.

[0052] This invention also provides a glass drilling microcrack detection system using image processing to implement the above-mentioned method, such as... Figure 2 As shown, the system includes: The curvature calculation unit is used to acquire point cloud data of the target glass through a 3D scanner, divide the target glass into multiple regions, calculate the curvature of the point cloud data of each region, and obtain the curvature distribution map of the target glass based on the curvature of each region. The image acquisition unit is used to simulate the light propagation path according to the curvature distribution map, obtain the light intensity adjustment coefficient, and the adjustable light source device adjusts the illumination intensity based on the light intensity adjustment coefficient and acquires the image dataset. The image processing unit is used to smooth the noise in the high curvature region of the image dataset using Gaussian filtering to obtain a denoised image set; and to perform binarization processing on the denoised image set to obtain a hidden crack binary image. The alignment processing unit is used to calculate the length and width indices of the cracks based on the binary image of the hidden cracks, and determine the set of hidden crack locations based on the length and width indices and align it with the curvature distribution map to obtain superimposed data. The hidden crack identification unit is used to fuse the hidden crack location set and the curvature distribution map according to the superimposed data to obtain a preliminary report map; and to perform image synthesis on the preliminary report map to generate the final hidden crack identification result.

[0053] The present invention also provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the above-described method.

[0054] In summary, this invention's 3D scanner accurately acquires geometric data of the target glass surface and quantifies the curvature value of each region. Analysis of the curvature distribution map identifies high-curvature areas, which are typically hotspots for microcracks. Furthermore, by adjusting the illumination intensity coefficient based on the curvature distribution map, the image acquisition quality is optimized, especially in high-curvature areas, ensuring the uniformity and accuracy of the image data and laying a solid foundation for subsequent image processing. The image dataset is also denoised using Gaussian filtering to smooth noise, particularly in high-curvature areas. Further, region connectivity analysis is employed to enhance the visibility of microcrack edges. The denoised image set is then binarized to obtain a microcrack binary map, ensuring effective separation of the microcrack area from the background and clearly presenting the microcrack features. The length, width, and other indicators of the crack are calculated from the microcrack binary map and marked as defect areas. These microcrack location sets are aligned with the curvature distribution map to generate overlay data, tightly integrating the microcrack location with curvature information. This process ensures that the spatial location of the microcrack matches the curvature variation of the glass surface, thus providing accurate data support for subsequent microcrack labeling and analysis. The overlay data is fused with the curvature value overlay layer and the microcrack location set to generate a preliminary report image. Through image synthesis technology, the location and details of the microcrack are accurately overlaid on the curvature distribution map, further optimizing the image visualization. If the defect label is unclear in the curvature gradient visualization, background noise suppression and contrast enhancement are used to ensure the clarity and accuracy of the final report image. The synergy between these technical solutions effectively improves the accuracy and visualization quality of microcrack detection in curved glass.

[0055] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the concept of this application. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for detecting microcracks in glass through drilling using image processing, characterized in that, include: Step S1: Obtain point cloud data of the target glass using a 3D scanner, divide the target glass into multiple regions, calculate the curvature of the point cloud data of each region, and obtain the curvature distribution map of the target glass based on the curvature of each region. Step S2: Use the Phong lighting model to calculate the incident angle of light in different curvature regions, and adjust the lighting intensity according to the curvature changes. For high curvature regions, the specular reflection component increases. Through iterative calculation, simulate the deviation of the light path, and calculate the lighting intensity attenuation rate of each region accordingly to obtain the lighting intensity adjustment coefficient for each region. Use an adjustable light source device to illuminate each region of the curved glass according to the calculated adjustment coefficient. Use a camera to collect image sequences of each region and form an image dataset. The image sequence includes multiple images to capture dynamic changes under different lighting conditions. Step S3: The image dataset is smoothed by Gaussian filtering to remove noise in high curvature regions, resulting in a denoised image set. For the denoised image set, binarization is used to separate the hidden crack region from the background. Morphological dilation is used to expand the range of hidden crack texture extraction. If the background noise suppression is insufficient after edge smoothing filtering, the segmentation accuracy is evaluated again to generate a hidden crack binary map. Step S4: Calculate the length and width indices of the crack based on the binary image of the hidden crack, and determine the set of hidden crack locations based on the length and width indices and align them with the curvature distribution map to obtain superimposed data; by converting the two-dimensional coordinates of the hidden crack location set into three-dimensional spatial coordinates, an affine transformation matrix is ​​used to ensure that the hidden crack locations can be accurately superimposed on the high curvature region of the curvature distribution map. For regions with high curvature, the curvature gradient vector is calculated and the projection angle of the hidden crack locations is adjusted so that the two-dimensional coordinates of the hidden cracks are aligned with the three-dimensional coordinate system of the curvature distribution map. Step S5: Based on the superimposed data, the curvature value superimposed layer is used to fuse the hidden crack location set and the curvature distribution map to generate a preliminary report map; for the preliminary report map, an image synthesis method is used to integrate all the details of the hidden crack location set; if the defect label in the curvature gradient visualization is not clear, background noise suppression and enhancement contrast are applied to determine the generation status of the complete detection report map and generate the final hidden crack identification result.

2. The method as described in claim 1, characterized in that, In step S1, point cloud data of the curved glass is acquired, and the point cloud data is processed using a curvature calculation algorithm to obtain a curvature distribution map, including: Point cloud data of curved glass surface is collected by a 3D scanner. Curvature value of each region is quantified by curvature calculation algorithm. The curved glass is divided into multiple regions according to position coordinate mapping. If the curvature value of a region exceeds a preset threshold, it is marked as a high curvature region. The curvature classification of the region is determined and the curvature distribution map is generated.

3. The method as described in claim 1, characterized in that, In step S3, the image dataset is denoised to obtain a denoised image set, including: Based on the curvature classification in the image dataset, a Gaussian filtering algorithm is used to smooth noise in high curvature regions. If the pixel intensity after denoising is lower than a preset threshold, regional connectivity analysis is applied to enhance features, determine the location of hidden crack edges, and generate the denoised image set.

4. The method as described in claim 1, characterized in that, In step S5, the hidden crack location set and the curvature distribution map are fused based on the superimposed data to obtain a preliminary report map, including: Based on the superimposed data, the fusion of the hidden crack location set and the curvature distribution map is processed using curvature value superimposition layer. Defect labels highlighting high curvature areas are represented by color coding. The superimposition transparency is adjusted to optimize the visualization of curvature gradient, and the preliminary report map is generated.

5. A glass drilling microcrack detection system employing image processing, for implementing the method as described in any one of claims 1-4, characterized in that, The system includes: The curvature calculation unit is used to acquire point cloud data of the target glass through a 3D scanner, divide the target glass into multiple regions, calculate the curvature of the point cloud data of each region, and obtain the curvature distribution map of the target glass based on the curvature of each region. The image acquisition unit is used to simulate the light propagation path according to the curvature distribution map, obtain the light intensity adjustment coefficient, and the adjustable light source device adjusts the illumination intensity based on the light intensity adjustment coefficient and acquires the image dataset. The image processing unit is used to smooth the noise in the high curvature region of the image dataset using Gaussian filtering to obtain a denoised image set; and to perform binarization processing on the denoised image set to obtain a hidden crack binary image. The alignment processing unit is used to calculate the length and width indices of the cracks based on the binary image of the hidden cracks, and determine the set of hidden crack locations based on the length and width indices and align it with the curvature distribution map to obtain superimposed data. The hidden crack identification unit is used to fuse the hidden crack location set and the curvature distribution map according to the superimposed data to obtain a preliminary report map; and to perform image synthesis on the preliminary report map to generate the final hidden crack identification result.

6. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the method as described in any one of claims 1-4.

Images