A method, system and device for visual quality detection of hollow glass pressing

Abstract

The application discloses a kind of hollow glass pressing visual quality detection method, system and equipment, it is related to image recognition technical field, comprising: according to first image acquisition parameter, the first area of hollow glass is carried out image acquisition, obtains first area image set;According to first pre-processing strategy, first area image set is carried out image enhancement, and glass pressing gap defect detection is carried out, and first area gap defect detection result is output;Second adaptive image acquisition parameter and second adaptive image pre-processing intensity are set, and the second area of hollow glass is carried out image acquisition and image enhancement;Second adaptive determination condition is set, and according to second adaptive determination condition, second enhanced area image set is carried out glass pressing gap defect detection, and second area gap defect detection result is generated.The technical problem that it is solved in prior art relies on artificial visual inspection, ignores the process correlation between sealing layer, leads to low detection efficiency, strong subjectivity, unstable detection precision.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and specifically to a visual quality detection method, system and equipment for the pressing of insulating glass. Background Technology

[0002] Insulating glass is a deep-processed glass product consisting of two or more panes of glass separated by an aluminum spacer frame, sealed with butyl rubber as the first sealant and a two-component sealant as the second sealant, forming a space containing dry gas. In the automated production process of insulating glass, insufficient or uneven pressing pressure during lamination will create gaps between the butyl rubber and the inner surface of the glass, compromising the airtightness of the cavity and affecting the subsequent filling and curing quality of the two-component sealant. Ultimately, this will cause the finished insulating glass to be scrapped due to seal failure.

[0003] Currently, the inspection of the lamination quality of insulating glass mainly relies on manual visual inspection. There is a lack of imaging schemes that differentiate the structural differences and optical properties of different sealing layers, and no linkage detection mechanism based on the process correlation of the two-level sealing layers has been established, making it difficult to achieve accurate and efficient detection of lamination gap defects. Summary of the Invention

[0004] This invention provides a visual quality inspection method, system, and equipment for insulated glass bonding, which solves the technical problems of low inspection efficiency, high subjectivity, and unstable inspection accuracy caused by relying on manual visual inspection and ignoring the process correlation between sealing layers in the prior art.

[0005] In a first aspect, the present invention provides a method for visual quality inspection of insulating glass bonding, the method comprising:

[0006] The first region of the insulating glass is image acquired according to the first image acquisition parameters to obtain the first region image set. The insulating glass is a finished insulating glass, and the first region is the butyl rubber sealing layer between the inner surfaces of the two glass panes and the aluminum spacer frame.

[0007] The first region image set is enhanced according to the first preprocessing strategy, and glass lamination gap defect detection is performed based on the first enhanced region image set. The first region gap defect detection result is output.

[0008] Based on the detection results of the void defects in the first region, the second adaptive image acquisition parameters and the second adaptive image preprocessing intensity are set, and the second region of the insulating glass is image acquired and image enhanced to generate a second enhanced region image set. The second region is the two-component sealant layer between the outer edges of the two glass panes.

[0009] Based on the detection results of void defects in the first region, a second adaptation judgment condition is set, and the glass lamination void defect detection is performed on the image set of the second enhanced region according to the second adaptation judgment condition to generate the void defect detection results of the second region.

[0010] Secondly, the present invention provides a visual quality inspection system for the lamination of insulating glass, the system comprising:

[0011] The first region image acquisition module is used to acquire images of the first region of the insulating glass according to the first image acquisition parameters, and obtain the first region image set. The insulating glass is a finished insulating glass, and the first region is the butyl rubber sealing layer between the inner surfaces of the two glass panes and the aluminum spacer frame.

[0012] The first region void defect detection module is used to perform image enhancement on the first region image set according to the first preprocessing strategy, perform glass lamination void defect detection based on the first enhanced region image set, and output the first region void defect detection result.

[0013] The second region image acquisition module is used to set the second adaptive image acquisition parameters and the second adaptive image preprocessing intensity according to the first region void defect detection result, to perform image acquisition and image enhancement on the second region of the insulating glass, and to generate a second enhanced region image set, wherein the second region is a two-component sealant layer between the outer edges of the two glass panes;

[0014] The second region void defect detection module is used to set a second adaptation judgment condition based on the first region void defect detection result, and to perform glass bonding void defect detection on the second enhanced region image set according to the second adaptation judgment condition, thereby generating a second region void defect detection result.

[0015] Thirdly, the present invention provides a visual quality inspection device for the lamination of insulating glass, wherein the visual quality inspection device is a carrier in the process of performing a visual quality inspection method for lamination of insulating glass.

[0016] One or more technical solutions provided in this invention have at least the following technical effects or advantages:

[0017] This invention provides a visual quality inspection method, system, and device for the lamination of insulating glass. First, image acquisition parameters specifically tailored to the structural characteristics of the first region are set to ensure image quality, enabling effective imaging of the butyl sealant layer between the two glass panes and providing reliable raw data for accurate detection. Second, a first preprocessing strategy is used for denoising, correction, enhancement, and region extraction to eliminate various interference factors during image acquisition, improving the quality and usability of the first region image. Through pixel-by-pixel local grayscale deviation analysis and local adaptive threshold segmentation, high-sensitivity detection of lamination gap defects is achieved, outputting quantitative detection results including defect quantity, area, location coordinates, and grade classification. Third, a correlation is established between the detection results of the first region and the detection parameters of the second region. By adaptively adjusting the image acquisition resolution and preprocessing intensity of the second region according to the severity of defects in the first region, detection resources are optimized, ensuring that the image acquisition and enhancement processing of the second region matches the actual quality condition of the first region, improving the targeting and accuracy of the detection. Finally, the severity of defects in the first region is mapped to a correction coefficient for the judgment threshold of the second region. This avoids the problem of missed detection when the first region defect is severe and false detection when the first region defect is slight when the judgment threshold is fixed. The result of the detection of the lamination defect of the insulating glass is obtained, and the systematic detection of lamination gap defects of the two-level sealing layer is realized. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a visual quality inspection method for insulated glass bonding provided by the present invention.

[0019] Figure 2 This is a logical schematic diagram of a visual quality detection method for insulated glass bonding provided by the present invention;

[0020] Figure 3 This is a schematic diagram of the structure of a visual quality inspection system for insulated glass pressing provided by the present invention.

[0021] In the attached diagram, the components represented by each number are as follows:

[0022] First region image acquisition module 11; first region void defect detection module 12; second region image acquisition module 13; second region void defect detection module 14. Detailed Implementation

[0023] This invention provides a visual quality inspection method, system, and equipment for insulated glass bonding, which solves the technical problems of low inspection efficiency, high subjectivity, and unstable inspection accuracy caused by relying on manual visual inspection and ignoring the process correlation between sealing layers in the prior art.

[0024] The present invention will now be described in detail with reference to the accompanying drawings.

[0025] Example 1, as Figure 1 , Figure 2 As shown, the present invention provides a method for visual quality inspection of insulated glass bonding, the method comprising:

[0026] S100: The first region of the insulating glass is image acquired according to the first image acquisition parameters to obtain the first region image set, wherein the insulating glass is a finished insulating glass and the first region is the butyl rubber sealing layer between the inner surfaces of the two glass panes and the aluminum spacer frame.

[0027] In this embodiment of the invention, the insulated glass is a deep-processed glass product consisting of two or more flat glass panes separated by an aluminum spacer frame and sealed around the perimeter, forming a dry gas space between the glass layers; the first region is the area where the butyl sealant layer is located between the inner surfaces of the two glass panes and the aluminum spacer frame in the insulated glass; the first region is the core area for evaluating the pressing quality of the insulated glass, and if there is a pressing gap between the butyl sealant and the inner surface of the glass, it will directly affect the airtightness of the hollow cavity; the first image acquisition parameters are a set of imaging parameters set for acquiring images of the first region.

[0028] Specifically, butyl sealant is a non-drying sealant composed primarily of butyl rubber and a small amount of polyisobutylene. It is applied to both sides of the aluminum spacer frame before the insulated glass units are laminated. After lamination, the butyl sealant adheres to the inner surfaces of the two glass panes, forming the first sealing barrier of the hollow cavity. Therefore, the first region of the insulated glass unit—the butyl sealant layer between the inner surfaces of the two glass panes and the aluminum spacer frame—is the primary target for inspection. Since the first region is located between the two glass panes and belongs to the internal structural layer of the insulated glass unit, image acquisition requires penetration of the outer glass layer under suitable lighting conditions. Therefore, image acquisition is performed on the first region according to the first image acquisition parameters to obtain an image set reflecting the lamination state of the butyl sealant layer.

[0029] In step S100 of the method of this embodiment of the invention, the automated production process of finished insulating glass includes:

[0030] The glass pieces to be processed are sequentially sent to the cleaning section for cleaning and air drying. After the surface is confirmed to be clean, an aluminum spacer frame coated with butyl rubber is placed on the surface of the first glass piece. Then, the second glass piece is joined to the first glass piece on both sides of the aluminum frame. The pressing section presses the butyl rubber to bond the two glass pieces together to form a hollow cavity. The pressed semi-finished product is then transferred to the sealant applicator to fill the gap between the outer edges of the two glass pieces with two-component sealant. After the sealant has completely cured, the finished insulated glass is formed.

[0031] In this embodiment of the invention, firstly, since the cleanliness of the glass surface directly affects the bonding effect of butyl rubber, any residual contaminants may cause the bonding between the butyl rubber and the glass to fail, thus forming a gap in the bonding. Therefore, the glass to be processed is sent to the cleaning section, where surface contaminants are removed by cleaning, and then air-dried to ensure the surface is dry and clean. Subsequently, the glass to be processed is conveyed by a conveyor mechanism, placing it on the conveyor belt of the conveyor mechanism and conveying it to a specific lamination station. The first piece of glass conveyed is used as the first piece of glass for lamination.

[0032] Secondly, after confirming the glass surface is clean, an automatic adhesive applicator evenly applies a continuous layer of butyl adhesive to both sides of the aluminum spacer frame. The pre-applied butyl adhesive aluminum spacer frame is then precisely placed on the predetermined position on the surface of the first glass piece. Subsequently, the second glass piece to be processed is conveyed to the lamination station by a conveyor mechanism. Under the action of the lamination machine, the second glass piece and the first glass piece are laminationed together on both sides of the aluminum frame, thus initially combining the two glass pieces and the aluminum spacer frame.

[0033] Next, the assembled glass assembly enters the pressing section. The pressing mechanism applies a predetermined pressure evenly according to the preset process parameters, so that the butyl rubber is fully deformed under pressure and tightly adhered to the inner surface of the two glass pieces, forming a sealed hollow cavity between the two glass pieces and the aluminum spacer frame.

[0034] Finally, after pressing, the semi-finished insulating glass is transferred to the sealing station via a conveyor. A two-component sealing machine automatically applies two-component sealant to the gap between the outer edges of the two glass panes according to a set trajectory. After filling, the sealant is cured at room temperature or by heating for a certain period of time. The base adhesive and the curing agent undergo a cross-linking reaction, gradually transforming from a paste state into a highly elastic solid, forming a strong second sealing layer. After the sealant is completely cured, the finished insulating glass is formed. Finally, it is automatically unloaded by the unloading table and transferred to the subsequent inspection or packaging process.

[0035] In step S100 of the method of this embodiment of the invention, the first image acquisition parameters include first light source parameters, first camera parameters, first imaging mode, first acquisition position and angle, and first exposure parameters. The first light source parameters adopt a backlight transmission illumination method, in which the light passes perpendicularly through the thickness direction of the hollow glass along the optical axis, so that the refractive index difference caused by the existence of the pressing gap in the first region is converted into grayscale difference in the transmission image.

[0036] In this embodiment of the invention, the backlight transmission illumination method involves the light source being located behind the object being measured, and the light passing through the object being received by the camera. For transparent or translucent objects, backlight illumination can highlight internal defects and structural features. Grayscale difference refers to the difference in pixel grayscale values ​​between different regions in a digital image. Due to the difference in refractive index, the light propagation path changes in the lamination gap region, resulting in a different grayscale value in the transmission image compared to the surrounding normal area.

[0037] Specifically, the first image acquisition parameters include: first light source parameters, first camera parameters, first imaging mode, first acquisition position and angle, and first exposure parameters. The first light source parameters specify the type of illumination source, illumination temperature, illumination uniformity, and light source power. For example, if the first light source uses a backlit transmission illumination method: the light source is positioned below one side of the insulating glass unit, and the camera is positioned above the other side of the insulating glass unit. The light passes perpendicularly through the thickness direction of the insulating glass unit along the optical axis. Under backlit projection illumination, the working principle is as follows: when the light passes perpendicularly through the insulating glass unit, refraction and reflection occur at the interfaces between the glass and butyl rubber, and between the butyl rubber and air gaps. In well-bonded areas, the transmitted light intensity is uniformly distributed at the glass-butyl rubber interface, resulting in a uniform grayscale. In areas with bonding gaps, the light undergoes additional refraction and reflection at the glass-air interface and the air-butyl rubber interface, causing changes in transmitted light intensity, which are represented in the image as grayscale values ​​different from the surrounding areas. The light source has a color temperature of 4500K~6500K, an illumination uniformity of ≥90%, and a power of 50W~150W.

[0038] The first camera parameter specifies the camera's resolution, frame rate, lens focal length, and other parameters. For example, when using an area array industrial camera, the resolution is 5 megapixels to 20 megapixels, the image acquisition frame rate is 5fps to 30fps, and the lens focal length is 8mm to 25mm. The first imaging mode specifies the camera's operating mode. For example, it may be based on transmitted light imaging, where light passes sequentially through the first glass, the butyl rubber sealing layer, and the second glass before entering the camera. When there are gaps in the butyl rubber sealing layer, the air layer at the gap and the butyl rubber layer create a difference in refractive index, forming grayscale anomaly areas in the transmitted image.

[0039] The first acquisition position and angle are used to specify the spatial position and shooting angle of the camera relative to the insulating glass unit. For example, the camera's optical axis is perpendicular to the surface of the insulating glass unit, the distance between the camera and the glass surface is 200mm~800mm, and the acquisition field of view covers the annular area around the perimeter of the insulating glass unit. The first exposure parameters are used to specify the camera's exposure time, gain, etc. For example, the exposure time is 5ms~50ms, and the gain is 0dB~12dB, so that the gray value of the butyl rubber layer area in the transmitted image is within 40%~80% of the image's gray-level dynamic range. Through the first image acquisition parameters, the transmitted image of the first region is acquired as the first region image set.

[0040] In this embodiment of the invention, by defining the production process of the finished insulating glass, the order and relationship of the formation of the first region and the second region in terms of process are determined; then, the first image acquisition parameters are set respectively, and by adopting a backlight transmission illumination method, the refractive index difference at the pressing gap is converted into the grayscale difference of the transmission image, thereby realizing the effective imaging of the butyl sealant layer located between the two pieces of glass, ensuring the image quality and reliability of the first region detection.

[0041] S200: Perform image enhancement on the first region image set according to the first preprocessing strategy, perform glass lamination gap defect detection based on the first enhanced region image set, and output the first region gap defect detection result.

[0042] In this embodiment of the invention, the first preprocessing strategy is an image preprocessing scheme designed for the characteristics of the first region image set; image enhancement is to transform the original image using digital image processing technology; glass lamination gap defects are local gaps or bubbles formed during the lamination process of insulating glass due to insufficient lamination pressure, uneven lamination, or poor adhesive coating, resulting in the butyl sealant and the inner surface of the glass not being completely bonded.

[0043] Specifically, the image set is enhanced according to the first preprocessing strategy, generating a first enhanced region image set. Glass lamination void defects are detected based on this first enhanced region image set. Local grayscale deviation values ​​are calculated pixel-by-pixel, and candidate defect pixels are segmented using a local adaptive threshold. Morphological processing connects the discrete candidate defect pixels into a complete candidate defect region. Geometric feature parameters of each candidate defect region are extracted, and candidate defect regions that meet the preset first judgment criteria are determined to be lamination void defects. Finally, the detection results for void defects in the first region are summarized and output.

[0044] Step S200 in the method of this embodiment of the invention includes:

[0045] According to the first preprocessing strategy, the first region image set is sequentially subjected to image denoising, illumination correction, contrast enhancement and region of interest extraction to obtain the first enhanced region image set.

[0046] Based on the first enhanced region image set, the local grayscale deviation value is calculated pixel by pixel, and the candidate defect pixels are segmented according to the local adaptive threshold to generate the candidate defect pixel region.

[0047] The candidate defect pixel regions are sequentially subjected to closing and opening operations to generate at least one candidate defect region in the form of a connected component. The geometric feature parameters of each candidate defect region are extracted. Candidate defect regions that meet the preset first judgment conditions are judged as compression gap defects. The preset first judgment conditions include that the area of ​​the candidate defect region is greater than the preset first area threshold, the maximum diameter is greater than the preset first diameter threshold, and the aspect ratio of the minimum bounding rectangle is less than the preset first aspect ratio threshold.

[0048] Summarize all candidate defect regions identified as compression void defects, and output the void defect detection results for the first region. The detection results include the number of void defects, the area of ​​each void defect, its location coordinates, and its classification level.

[0049] In this embodiment of the invention, firstly, the first region image set is sequentially subjected to image denoising, illumination correction, contrast enhancement, and region of interest extraction according to a first preprocessing strategy to obtain a first enhanced region image set. Specifically, image denoising is the process of removing random pixel value anomalies caused by sensor thermal noise, photon noise, circuit noise, etc., in digital images using filtering algorithms; illumination correction is the process of correcting overall or local brightness unevenness in the image caused by factors such as uneven light source and changes in the curvature of the glass surface; contrast enhancement is the process of expanding the grayscale difference between the target area and the background area in the image through grayscale transformation, making the target features clearer and more distinguishable; and region of interest extraction is the process of accurately locating and cropping the local area where the butyl rubber sealing layer is located from the entire image based on prior knowledge, excluding the influence of irrelevant areas such as glass edges and the background on subsequent processing.

[0050] Specifically, the first preprocessing strategy includes the following image enhancement operations performed sequentially:

[0051] First, denoising is performed. Guided filtering or bilateral filtering is used to denoise the first region image set. While effectively filtering out random pixel value anomalies caused by sensor thermal noise, photon noise, and circuit noise, the edge detail information of the butyl rubber sealing layer area is preserved through the edge-preserving characteristics of guided filtering or the gray-level similarity weighting mechanism of bilateral filtering, so as to avoid the edge blurring from interfering with the subsequent detection of void defects. The filtering window size is 3×3 pixels to 7×7 pixels, and the filtering intensity parameter ε is 0.01 to 0.05.

[0052] Secondly, illumination correction is performed. Homomorphic filtering or background subtraction is used to correct for uneven illumination in the denoised image. Homomorphic filtering separates and compresses the illumination and reflection components in the frequency domain to eliminate the influence of uneven overall backlight distribution. Alternatively, background subtraction is used to construct a uniform background reference image to eliminate the light intensity attenuation effect in the glass edge area, making the background grayscale of the butyl sealant layer area more uniform and consistent, thus avoiding the misjudgment of defects due to grayscale deviations caused by illumination changes.

[0053] Next, contrast enhancement is performed. Adaptive histogram equalization or contrast-limited adaptive histogram equalization is used to stretch the contrast of the illumination-corrected image. By dividing the image into several sub-regions and performing histogram equalization on each, local contrast is enhanced. This significantly amplifies the grayscale difference between the low-contrast areas caused by the pressure gaps in the butyl rubber sealant layer and the normally bonded areas. The cropping limit of the contrast-limited adaptive histogram equalization is 0.01 to 0.05 to suppress noise amplification caused by over-enhancement. The grayscale distribution sub-regions are divided into 8×8 to 16×16 blocks to ensure local enhancement accuracy while also considering processing efficiency.

[0054] Finally, precise ROI localization is performed. In the contrast-enhanced image, for example, the outer contour of the insulating glass is extracted using the Sobel edge detection operator. A ring-shaped detection mask is generated based on this outer contour. The inner boundary of this ring-shaped detection mask is 15mm to 40mm away from the edge of the glass outer contour, precisely covering the area where the butyl sealant layer is located between the inner surfaces of the two glass panes and the aluminum spacer frame. Subsequent defect detection is performed on the image area covered by this ring-shaped detection mask. This eliminates interference from non-detection areas at the glass edge, background interference areas, and the area inside the aluminum spacer frame at the source, reducing the computational load of subsequent processing and improving the signal-to-noise ratio of defect determination.

[0055] Secondly, based on the image set of the first enhanced region, the local grayscale deviation value is calculated pixel by pixel. For each pixel, a local neighborhood window of a preset size is selected with the pixel as its center. The size of this window can be set according to the actual detection accuracy requirements and image resolution. For example, in the detection scenario of lamination void defects in the butyl sealant layer of insulating glass, the window size is generally taken as 2 to 3 times the minimum feature size of the defect. Experiments have verified that window sizes in the range of 2×2 pixels to 5×5 pixels can achieve good detection results.

[0056] After determining the local neighborhood window, the mean gray value μ and the standard deviation σ of all pixels within the window are calculated. The mean reflects the gray-level baseline of the area surrounding the pixel. For each pixel, the deviation of its gray value from the average gray value of its local neighborhood is calculated, i.e., local gray-level deviation = pixel gray value - average gray value of pixels in the local neighborhood.

[0057] A local adaptive threshold is then set, employing an adaptive threshold setting strategy based on the local grayscale standard deviation. Specifically, for each pixel, the local adaptive threshold T is determined by the mean and standard deviation of the grayscale values ​​in its local neighborhood. Specifically, T = μ + k × σ, where σ is the grayscale standard deviation and k is the sensitivity coefficient, used to control the sensitivity of defect detection. The value of the sensitivity coefficient k determines the balance between recall and precision in defect detection. The value of k is set according to the actual needs of insulated glass lamination quality inspection and can be dynamically adjusted according to the quality control requirements of the production line in practical applications. For example, during order production with high product quality requirements, the value of k can be appropriately reduced to improve detection sensitivity; during normal production, k = 2 can be used. Then, candidate defect pixels are segmented based on the local adaptive threshold. Specifically, pixels with grayscale values ​​greater than the local adaptive threshold are identified as candidate defect pixels; pixels with grayscale values ​​less than or equal to the local adaptive threshold are identified as normal background pixels. This process is repeated for each pixel to generate the candidate defect pixel region for each first enhancement region image.

[0058] In a feasible embodiment, assuming the local neighborhood grayscale mean μ = 128, the local grayscale standard deviation σ = 15, and the sensitivity coefficient k = 2, then the local adaptive threshold T for this pixel is 128 + 2.0 × 15 = 158. If the grayscale value of this pixel is 175, then its local grayscale deviation value is 175 - 128 = 47. Since the local grayscale deviation value is greater than the local adaptive threshold, this pixel is determined to be a candidate defective pixel.

[0059] Next, closing and opening operations are performed sequentially on the candidate defect pixel regions to generate at least one candidate defect region in the form of a connected component. Geometric feature parameters of each candidate defect region are extracted, and candidate defect regions that meet the preset first judgment conditions are judged as compression gap defects. The preset first judgment conditions include that the area of ​​the candidate defect region is greater than a preset first area threshold, the maximum diameter is greater than a preset first diameter threshold, and the aspect ratio of the minimum bounding rectangle is less than a preset first aspect ratio threshold.

[0060] Specifically, in mathematical morphology, the closing operation involves first dilating an image and then eroding it. The closing operation is used to fill small holes inside foreground objects, connect adjacent fracture areas, and smooth object boundaries. In the detection of lamination voids in insulating glass, a true lamination void defect appears as a continuous region with a certain area and shape in the first enhanced image. However, due to image noise, uneven illumination, and other factors, the same defect region may be segmented into several discrete small regions after initial thresholding. The closing operation can connect these discrete small regions into a complete candidate defect region, and simultaneously fill the tiny holes inside the defect region.

[0061] In a specific embodiment, the structuring element used in the closing operation is a circular structuring element. The size of the structuring element needs to be set according to the connectivity effect of the defect and the detection accuracy. If the structuring element is too small, it cannot effectively connect scattered defect fragments; if the structuring element is too large, it may incorrectly connect originally independent defect areas together, or excessively expand the defect boundary, leading to distorted detection results. The typical size of the lamination gap defect in insulating glass is usually between 0.5mm and 5mm. A circular structuring element with a radius of 2 to 5 pixels can achieve a good connectivity effect. Therefore, the radius of the circular structuring element can be set to 3 pixels.

[0062] After the closing operation, an opening operation is performed on the candidate defect pixel region. The opening operation involves first performing an erosion operation on the image, followed by a dilation operation. The opening operation removes small noise points, isolated small regions, and smooths the contours of objects in the image. Although the candidate defect region is connected after the closing operation, isolated small regions caused by noise may still exist in the image. However, these isolated small regions do not belong to the compression gap defects. The opening operation effectively filters out isolated small regions, purifies the candidate defect region, and reduces interference in subsequent judgments.

[0063] In a specific embodiment, the opening operation uses the same structuring element as the closing operation, namely a circular structuring element with a radius of 3 pixels. After the closing and opening operations, the candidate defect pixel region is transformed into at least one connected component candidate defect region. Each connected component consists of a series of spatially adjacent candidate defect pixels, typically determined using 8-neighborhood connectivity, representing potential candidates for compression gap defects.

[0064] Finally, all candidate defect regions identified as compression void defects are summarized, and the void defect detection results for the first region are output. The detection results include the number of void defects, the area of ​​each void defect, its location coordinates, and its classification.

[0065] In this embodiment of the invention, the image set of the first region is preprocessed by denoising, illumination correction, contrast enhancement, and precise ROI positioning to improve the image quality of the first region; high-sensitivity detection of pressing gaps and effective elimination of noise interference are achieved through local adaptive threshold segmentation and morphological processing; the false detection rate is reduced by comprehensively judging the three geometric constraints of area, diameter, and aspect ratio; and quantitative results including the number, area, and level of defects are output, providing a data basis for the dynamic adjustment of detection parameters in the second region.

[0066] S300: Based on the detection results of the void defects in the first region, set the second adaptation image acquisition parameters and the second adaptation image preprocessing intensity, perform image acquisition and image enhancement on the second region of the insulating glass, and generate a second enhanced region image set, wherein the second region is a two-component sealant layer between the outer edges of the two glass panes;

[0067] In this embodiment of the invention, the second region is the area where the two-component sealant layer is located between the outer edges of the two panes of glass in the insulated glass unit. The second region serves as the second sealing barrier for the insulated glass unit, and its quality directly affects its long-term sealing life. The second adaptive image acquisition parameters are determined after adaptive adjustment based on the void defect detection results of the first region, and are used to acquire the imaging parameter set for the second region image. The second adaptive image preprocessing intensity is determined after adaptive adjustment based on the void defect detection results of the first region, and is used to enhance the image of the second region.

[0068] Specifically, if a lamination gap defect exists in the first region, it indicates an abnormality in the lamination process, which may affect the filling and curing quality of the two-component sealant in the second region. Therefore, the image acquisition parameters and preprocessing intensity of the second region need to be adjusted according to the severity of the defect in the first region. After setting the parameters, images of the second region of the insulating glass are acquired according to the second adapted image acquisition parameters to obtain the second region image set; then, image enhancement processing is performed on the second region image set according to the second adapted image preprocessing intensity to generate the second enhanced region image set.

[0069] Step S300 in the method of this embodiment of the invention includes:

[0070] Configure second image acquisition parameters, which include second light source parameters, second camera parameters, second imaging mode, second acquisition position and angle, and second exposure parameters. The second light source parameters adopt a low-angle reflected light illumination method, with the light illuminating the outer edge surface of the insulating glass at an angle of 15 to 30 degrees, so that surface defects in the second area caused by poor filling or incomplete curing of the two-component sealant are converted into grayscale differences or texture differences in the reflected image. The second camera parameters include a second image resolution.

[0071] The average area of ​​the first void defect and the proportion of the first void defect area are calculated based on the area of ​​each void defect in the void defect detection results of the first region. The severity coefficient of the first void defect is determined by combining the number of void defects.

[0072] The ratio of the severity coefficient of the first void defect to the preset first standard void defect severity coefficient is used as the second correction coefficient.

[0073] The product of the second correction coefficient and the second image resolution is rounded down to obtain the second adapted image resolution, and the second adapted image acquisition parameters are generated by combining the other fixed parameters in the second image acquisition parameters.

[0074] The second region of the insulating glass is image acquired according to the second adaptive image acquisition parameters to obtain the second region image set.

[0075] The second adaptation image preprocessing intensity is set according to the second correction coefficient, and the second region image set is enhanced to generate the second enhanced region image set.

[0076] In this embodiment of the invention, firstly, second image acquisition parameters are configured, including second light source parameters, second camera parameters, second imaging mode, second acquisition position and angle, and second exposure parameters. The second light source parameters adopt a low-angle reflected light illumination method, with the light illuminating the outer edge surface of the insulating glass at an angle of 15 to 30 degrees, so that surface defects in the second area caused by poor filling or incomplete curing of the two-component sealant are converted into grayscale differences or texture differences in the reflected image. The second camera parameters include a second image resolution.

[0077] Specifically, the second image acquisition parameters include the second light source parameters, the second camera parameters, the second imaging mode, the second acquisition position and angle, and the second exposure parameters.

[0078] The second light source can employ a low-angle reflected light illumination method, with a color temperature set within the standard white light range of 4500K to 6500K. Illumination uniformity should be no less than 85% to ensure a consistent illumination baseline across the field of view, avoiding localized underexposure or overexposure due to uneven illumination. An angle exists between the light source's output optical axis and the normal to the illuminated glass surface. When the angle is less than 15 degrees, the light is too close to the glass surface, making it difficult to form an effective reflected light signal on the adhesive layer, resulting in insufficient image contrast. When the angle is greater than 30 degrees, the incident angle of the light is too large, and the reflected light direction deviates from the camera's optical axis, similarly reducing the effective reflected light entering the camera and potentially introducing interference from non-defect factors such as scratches on the glass surface. Therefore, the light source is set to illuminate the outer edge surface of the insulating glass at an angle of 15 to 30 degrees to obtain the optimal reflected light signal intensity and defect contrast on the adhesive layer surface. Multiple LED light source modules are arranged at intervals along the edge of the glass, with a spacing of 100mm to 300mm between adjacent LED modules. The power of a single LED module is 5W to 20W. The combination and arrangement of multiple LED modules achieves uniform illumination of the annular area around the perimeter of the insulating glass, while ensuring that the light source system has sufficient redundancy and maintainability.

[0079] The second camera uses an area array industrial camera with a resolution of 5 to 20 megapixels and an image acquisition frame rate of 5 to 30 fps to adapt to the dynamic acquisition needs of different production line speeds. The lens focal length is a fixed focal length lens in the range of 8mm to 25mm.

[0080] When the two-component sealant layer has defects such as poor filling, curing shrinkage, or bubbles, the local surface orientation of the defective area changes significantly relative to the normal sealant layer surface. This causes the incident light in that area to be deflected or diffusely scattered, resulting in grayscale or texture differences in the reflected image that distinguish it from the surrounding normal sealant layer surface. Therefore, the second imaging mode primarily uses reflected light imaging. That is, when low-angle light shines on the surface of the two-component sealant layer between the outer edges of the two glass panes, reflected light is formed on the sealant layer surface, and the reflected light enters the camera to form a reflected image.

[0081] The second acquisition position and angle settings are: the angle between the camera's optical axis and the normal to the glass surface is 0 to 15 degrees, meaning the camera is approximately perpendicular to the glass surface during acquisition; the acquisition distance between the front face of the camera lens and the glass surface is set to 200mm to 600mm; and the acquisition field of view covers the annular area around the perimeter of the insulating glass. The second exposure parameters are: exposure time set to 3ms to 30ms, and gain set to 0dB to 12dB. By adjusting the exposure time and gain, the grayscale value of the two-component sealant layer area in the reflected image is kept between 40% and 80% of the image's grayscale dynamic range.

[0082] Secondly, the detection results of void defects in the first region include the number of void defects, the area of ​​each void defect, and the location coordinates and classification information of each defect in the area ratio of the void defects. Based on the area of ​​each void defect in the detection results of void defects in the first region, the average area of ​​the void defects and the area ratio of the void defects are calculated. The average area of ​​the void defects is used to reflect the average severity of a single defect. The area ratio of the void defects is the ratio of the area of ​​the void defects to the total area of ​​the butyl rubber sealing layer in the first region, which is used to reflect the coverage density of the defect area in the overall sealing layer.

[0083] The severity coefficient of the first void defect is determined by assessing the number of void defects. Specifically, the severity coefficient of the first void defect is calculated as: average area of ​​the first void defect × percentage of the area of ​​the first void defect × number of void defects. A higher severity coefficient indicates a more severe compression void defect in the first region.

[0084] Next, the ratio of the severity coefficient of the first void defect to the preset first standard void defect severity coefficient is used as the second correction coefficient, i.e., the second correction coefficient = the severity coefficient of the first void defect / the preset first standard void defect severity coefficient. Wherein, when the second correction coefficient = 1, it indicates that no adjustment is needed to the image acquisition parameters of the second region; when the second correction coefficient > 1, it indicates that the defect severity of the first region is higher than the baseline level, and the resolution and preprocessing intensity of the second region detection need to be increased to more precisely capture any accompanying defects in the second region; when the second correction coefficient < 1, it indicates that the defect severity of the first region is lower than the baseline level, and the resolution and preprocessing intensity of the second region detection can be appropriately reduced to improve detection efficiency.

[0085] Specifically, the preset first standard void defect severity coefficient is a pre-set reference standard value, representing the baseline level of the allowable severity of void defects in the first region under normal production process conditions for insulated glass. It can be set according to industry quality standards and internal control regulations, combined with the company's long-term production quality statistics, using the statistical average distribution of severity coefficients from the first region defect detection results of a large batch of qualified products under normal process conditions as a reference. For example, if the first region detection data of 1000 qualified insulated glass units under normal production conditions over the past three months is collected, and the average defect severity coefficient is found to be 0.28, then the preset first standard void defect severity coefficient is set to 0.3.

[0086] Furthermore, the product of the second correction coefficient and the second image resolution is rounded down to obtain the second adapted image resolution, i.e., the second adapted image resolution = ceil(second correction coefficient × second image resolution). Combined with other fixed parameters in the second image acquisition parameters, including the second light source parameters, the second imaging mode, the second acquisition position and angle, the second exposure parameters, etc., the second adapted image acquisition parameters are generated by combination.

[0087] Furthermore, following the acquisition process of the first region image set, and combining the second adaptive image acquisition parameters, images of the second region of the insulating glass are acquired using the same acquisition method to obtain the second region image set.

[0088] Finally, based on the second correction coefficient, a second adaptation image preprocessing intensity is set, and image enhancement is performed on the second region image set to generate a second enhanced region image set. The higher the second adaptation image preprocessing intensity, the richer the defect feature information extracted from the second region image; conversely, the lower the intensity, the less defect feature information is extracted.

[0089] In step S300 of the method of this embodiment of the invention, a second adaptation image preprocessing intensity is set according to the second correction coefficient, wherein the second adaptation image preprocessing intensity is positively correlated with the second correction coefficient, and the second adaptation image preprocessing intensity is mild enhancement, moderate enhancement, and strong enhancement.

[0090] In this embodiment of the invention, when a pressing gap defect exists in the first region, this defect may cause a slight change in the local glass spacing, affecting the curing morphology and surface morphology of the two-component sealant in the second region. The second correction coefficient quantifies the degree of influence; therefore, the intensity of image preprocessing in the second region can be dynamically adjusted according to the second correction coefficient, allowing the image enhancement effect in the second region to match the current detection conditions: when the second correction coefficient is small, the first region has a slight impact on the second region, and mild enhancement is sufficient to meet the detection requirements, resulting in fast processing speed; when the second correction coefficient is large, the first region has a significant impact on the second region, and strong enhancement is used to maximize the extraction of weak defect signals in the second region, avoiding missed detections. This achieves on-demand allocation of image processing resources, balancing detection efficiency and accuracy.

[0091] For example, different processing measures are adopted for different preprocessing intensities of the second adaptation image. During denoising, median filtering is used to denoise the second region image set. The filter window size is determined according to the preprocessing intensity level: for mild enhancement, the filter window is set to 3×3 pixels; for moderate enhancement, the filter window is set to 5×5 pixels; and for strong enhancement, the filter window is set to 7×7 pixels. During illumination correction, homomorphic filtering is used to correct uneven illumination in the denoised image. The filter cutoff frequency is determined according to the preprocessing intensity level; the higher the intensity level, the lower the filter cutoff frequency, and the stronger the smoothing effect. During contrast stretching, contrast-limited adaptive histogram equalization is used to stretch the contrast of the illumination-corrected image. The cropping limit value is determined according to the preprocessing intensity level: for mild enhancement, the cropping limit value is set to 0.03; for moderate enhancement, the cropping limit value is set to 0.02; and for strong enhancement, the cropping limit value is set to 0.01.

[0092] In a specific embodiment, an intensity switching threshold is set. For example, when the second correction coefficient is ≤0.8, a mild enhancement is used, performing only basic image denoising and contrast stretching operations; when 0.8 < the second correction coefficient ≤1.5, a moderate enhancement is used, adding histogram equalization and sharpening filtering on the basis of mild enhancement; when the second correction coefficient >1.5, a strong enhancement is used, adding multi-scale detail enhancement and adaptive gamma correction. For example, when the first void defect severity coefficient = 0.6 and the preset first standard void defect severity coefficient = 0.3, the second correction coefficient = 2, and 2 > 1.5, therefore a strong enhancement of the second adapted image preprocessing intensity is used to ensure that the richest possible defect feature information is extracted from the second region image.

[0093] In this embodiment of the invention, a second correction coefficient is calculated based on the severity of defects in the first region, and the image acquisition resolution and the intensity of the second adaptive image preprocessing in the second region are dynamically adjusted to achieve adaptive optimization of detection accuracy. A three-level second adaptive image preprocessing intensity that is positively correlated with the second correction coefficient is set so that the enhancement intensity increases reasonably with the defect risk level, ensuring that weak defect signals are effectively highlighted when defects are frequent and avoiding over-processing when defects are infrequent, thereby improving detection efficiency.

[0094] S400: Based on the first region void defect detection result, set a second adaptation judgment condition, and perform glass pressing void defect detection on the second enhanced region image set according to the second adaptation judgment condition to generate a second region void defect detection result.

[0095] In this embodiment of the invention, the second adaptation determination condition is determined after adaptive adjustment based on the detection results of the first region void defect, and is a criterion for determining whether the candidate defect region in the second enhanced region image set is a compression void defect; the determination condition is a quantitative standard for distinguishing the authenticity of candidate defect regions in visual defect detection.

[0096] Specifically, the detection results of void defects in the first region are used to adjust the defect judgment conditions for the second region. Based on the severity of defects reflected in the detection results of void defects in the first region, the thresholds in the preset judgment conditions are corrected. When the defects in the first region are severe, it indicates an overall abnormality in the pressing process, and the probability of defects appearing in the second region is higher, and the defects may be more hidden. Therefore, it is necessary to appropriately lower the judgment threshold to improve the detection sensitivity of defects in the second region. When the defects in the first region are minor, a stricter judgment threshold can be used to reduce false detections.

[0097] After setting the second adaptation judgment condition, the glass lamination gap defect detection is performed on the second enhanced region image set according to the condition. Finally, the judgment is made according to the second adaptation judgment condition, and the gap defect detection result of the second region is output.

[0098] Step S400 in the method of this embodiment of the invention includes:

[0099] The ratio of the preset first standard void defect severity coefficient to the first void defect severity coefficient is used as the condition parameter correction coefficient.

[0100] The second area threshold, second diameter threshold, and second aspect ratio threshold in the preset second determination condition are corrected according to the condition parameter correction coefficient to obtain the second adaptation area threshold, second adaptation diameter threshold, and second adaptation aspect ratio threshold, which are used as the second adaptation determination condition.

[0101] In this embodiment of the invention, the first region and the second region form an adjacent, series-sealed structure within the insulating glass. When a compression gap defect exists in the first region, this defect leads to geometrical effects and stress distribution effects. The geometrical effect causes the distance between the two glass panes at the gap in the first region to deviate from the design value, which is transmitted to the outer edge of the glass, resulting in localized changes in the filling space of the two-component sealant in the second region. This may manifest as poor filling or abnormal surface morphology in the second region. The stress distribution effect causes a lack of local structural support in the gap of the first region, redistributing stress at the glass edge and affecting the curing morphology and interfacial bonding quality of the two-component sealant in the second region. Therefore, the more severe the defect in the first region, the greater the impact on the corresponding second region, and the higher the probability of abnormal detection signals in the second region due to the coupling effect of the defect in the first region.

[0102] First, the ratio of the preset first standard void defect severity coefficient to the first void defect severity coefficient is used as the condition parameter correction coefficient, that is, the condition parameter correction coefficient = preset first standard void defect severity coefficient / first void defect severity coefficient.

[0103] Specifically, when the condition parameter correction coefficient is less than 1, the threshold standards for defect judgment in the second region should be lowered to detect minor defects that might otherwise be missed due to excessively high thresholds. This prevents quality issues in the second region caused by severe defects in the first region from being overlooked. Conversely, when the condition parameter correction coefficient is greater than 1, the threshold standards for defect judgment in the second region can be increased to avoid misjudging surface texture differences caused by normal process fluctuations as defects, thus reducing the false alarm rate. When the condition parameter correction coefficient is 1, the judgment conditions for the second region remain unchanged as the preset baseline judgment conditions.

[0104] Secondly, the second area threshold, the second diameter threshold, and the second aspect ratio threshold in the preset second judgment condition are corrected according to the condition parameter correction coefficient. The condition parameter correction coefficient can be multiplied by the second area threshold, the second diameter threshold, and the second aspect ratio threshold in the second judgment condition, and the calculated product is used as the second adaptation area threshold, the second adaptation diameter threshold, and the second adaptation aspect ratio threshold. At the same time, the second adaptation area threshold, the second adaptation diameter threshold, and the second adaptation aspect ratio threshold are used as the second adaptation judgment condition.

[0105] Step S400 in the method of this embodiment of the invention further includes:

[0106] In the second enhanced region image set, the gray-level gradient magnitude is calculated pixel by pixel, and the gray-level gradient magnitude is compared with a preset gradient threshold. Pixels with gray-level gradient magnitude greater than the preset gradient threshold are marked as candidate defect pixels. All marked candidate defect pixels constitute the candidate defect pixel region.

[0107] Morphological processing is performed on the candidate defect pixel regions to generate at least one candidate defect region in the form of a connected component. Geometric feature parameters of each candidate defect region are extracted, including region area, maximum diameter and minimum aspect ratio of the bounding rectangle.

[0108] The candidate defect regions are determined according to the second adaptation determination condition: when a candidate defect region simultaneously satisfies the following conditions: the region area is greater than the second adaptation area threshold, the maximum diameter is greater than the second adaptation diameter threshold, and the aspect ratio of the minimum bounding rectangle is less than the second adaptation aspect ratio threshold, the candidate defect region is determined to be the second region compression void defect.

[0109] Summarize all candidate defect regions identified as second region lamination void defects, and output the second region void defect detection results.

[0110] In this embodiment of the invention, in the second enhanced region image, there are significant differences in surface morphology between the normal area and the defective area of ​​the two-component sealant layer. When the two-component sealant has defects such as poor filling, curing shrinkage, or bubbles, the local surface orientation of the defective area changes. Under low-angle reflected light illumination, the light reflection direction of this area is deflected or diffuse scattering occurs. This is reflected in the reflected image as a drastic change in pixel grayscale value relative to the surrounding normal sealant layer area. This change can be quantified by the grayscale gradient amplitude.

[0111] First, in the second enhanced region image set, the gray-level gradient magnitude is calculated pixel by pixel. The gray-level gradient magnitude is calculated by squaring the first derivative of the pixel in the horizontal and vertical directions, summing them, and then taking the square root. For example, the first derivative of the pixel in the horizontal and vertical directions can be calculated using the Sobel operator.

[0112] The grayscale gradient magnitude is compared with a preset gradient threshold. Pixels with a grayscale gradient magnitude greater than the preset gradient threshold are marked as candidate defect pixels. All marked candidate defect pixels constitute the candidate defect pixel region. The preset gradient threshold is used to distinguish between areas of slow grayscale change on the normal adhesive layer surface and areas of drastic grayscale change caused by defects. It is set according to the specific quality condition of the second region image and the requirements of the second adaptation judgment condition.

[0113] In one embodiment, a statistical analysis of the grayscale gradient amplitude is performed on images of a large number of normal adhesive layer areas. The gradient amplitude of normal areas is usually concentrated in the range of 5 to 15, while the gradient amplitude of defective areas is usually greater than 20. Therefore, the midpoint between the mean of the gradient amplitude of defective areas minus 1.5 times the standard deviation and the mean of defective areas plus 1 times the standard deviation can be taken as the preset gradient threshold, for example, set to 20.

[0114] Secondly, morphological processing is performed on the candidate defect pixel regions, including closing operations to connect adjacent break regions and fill internal holes, and opening operations to remove isolated noise regions. Subsequently, candidate connected components are extracted to generate at least one candidate defect region in the form of a connected component. For each candidate defect region, geometric feature parameters are extracted, including region area, maximum diameter, and minimum aspect ratio of the bounding rectangle.

[0115] Next, each candidate defect region is judged according to the second adaptation judgment condition: when a candidate defect region simultaneously meets the following conditions, the region area is greater than the second adaptation area threshold, the maximum diameter is greater than the second adaptation diameter threshold, and the aspect ratio of the minimum bounding rectangle is less than the second adaptation aspect ratio threshold, the candidate defect region is judged to be a second region pressing void defect. Conversely, if none of the three conditions are met simultaneously, it is not judged to be a second region pressing void defect, but classified as negligible minor surface texture fluctuations.

[0116] Finally, all candidate defect areas identified as lamination void defects in the second region are summarized, and the detection results of the second region void defects are output, including the number of defects, the area of ​​each defect, the location coordinates and the level classification, etc., to complete the visual inspection of the lamination quality of the insulating glass.

[0117] In this embodiment of the invention, a condition parameter correction coefficient is calculated using a preset first standard void defect severity coefficient and a first void defect severity coefficient. The second adaptation area threshold, the second adaptation diameter threshold, and the second adaptation aspect ratio threshold are dynamically adjusted to obtain the second adaptation judgment condition, so that the judgment standard adapts to the defect risk. By detecting grayscale gradient amplitude and morphological processing, candidate defect areas are extracted and comprehensively judged according to the three judgment conditions, thereby improving the detection accuracy of the second region compression void defect, reducing the false detection rate, and ensuring the consistency of the two-level sealing layer detection standards.

[0118] Through the above specific implementation methods, the embodiments of the present invention achieve the following technical effects:

[0119] In this embodiment of the invention, firstly, the production process of the finished insulating glass is used to determine the order and relationship between the formation of the first region and the second region in terms of process; then, the first image acquisition parameters are set respectively, and by using backlight transmission illumination, the refractive index difference at the pressing gap is converted into the grayscale difference of the transmission image, thereby realizing effective imaging of the butyl sealant layer located between the two pieces of glass, ensuring the image quality and reliability of the first region detection.

[0120] Secondly, the image set of the first region is preprocessed by denoising, illumination correction, contrast enhancement, and precise ROI positioning to improve the image quality of the first region. Through local adaptive threshold segmentation and morphological processing, high-sensitivity detection of pressing gaps is achieved and noise interference is effectively eliminated. By comprehensively judging the three geometric constraints of area, diameter, and aspect ratio, the false detection rate is reduced. The output includes quantitative results containing the number, area, and grade of defects, providing a data basis for the dynamic adjustment of detection parameters in the second region.

[0121] Furthermore, the second correction coefficient is calculated based on the severity of defects in the first region, and the image acquisition resolution and the intensity of the second adaptive image preprocessing in the second region are dynamically adjusted to achieve adaptive optimization of detection accuracy. Three levels of the second adaptive image preprocessing intensity—mild, moderate, and strong—are set, which are positively correlated with the second correction coefficient, so that the enhancement intensity increases reasonably with the defect risk level. This ensures that weak defect signals are effectively highlighted when defects are frequent, and avoids over-processing when defects are infrequent, thereby improving detection efficiency.

[0122] Finally, by using the preset first standard void defect severity coefficient and the first void defect severity coefficient, the condition parameter correction coefficient will be calculated, and the second adaptation area threshold, the second adaptation diameter threshold, and the second adaptation aspect ratio threshold will be dynamically adjusted to obtain the second adaptation judgment condition, so that the judgment standard adapts to the defect risk. Through grayscale gradient amplitude detection and morphological processing, candidate defect areas will be extracted and comprehensively judged according to the three judgment conditions, thereby improving the detection accuracy of the second region compression void defect, reducing the false detection rate, and ensuring the consistency of the two-level sealing layer detection standards.

[0123] Example 2, as Figure 3 As shown, based on the same inventive concept as the visual quality inspection method for insulated glass bonding provided in Embodiment 1, this embodiment of the invention also provides a visual quality inspection system for insulated glass bonding, the system comprising:

[0124] The first region image acquisition module 11 is used to acquire images of the first region of the insulating glass according to the first image acquisition parameters, and obtain the first region image set. The insulating glass is a finished insulating glass, and the first region is the butyl rubber sealing layer between the inner surfaces of the two glass panes and the aluminum spacer frame.

[0125] The first region void defect detection module 12 is used to perform image enhancement on the first region image set according to the first preprocessing strategy, perform glass lamination void defect detection based on the first enhanced region image set, and output the first region void defect detection result.

[0126] The second region image acquisition module 13 is used to set the second adaptation image acquisition parameters and the second adaptation image preprocessing intensity according to the first region void defect detection result, to perform image acquisition and image enhancement on the second region of the insulating glass, and to generate a second enhanced region image set, wherein the second region is a two-component sealant layer between the outer edges of the two glass panes.

[0127] The second region void defect detection module 14 is used to set a second adaptation judgment condition based on the first region void defect detection result, perform glass bonding void defect detection on the second enhanced region image set according to the second adaptation judgment condition, and generate a second region void defect detection result.

[0128] In one embodiment, the first region image acquisition module 11 is used for:

[0129] The glass pieces to be processed are sequentially sent to the cleaning section for cleaning and air drying. After the surface is confirmed to be clean, an aluminum spacer frame coated with butyl rubber is placed on the surface of the first glass piece. Then, the second glass piece is joined to the first glass piece on both sides of the aluminum frame. The pressing section presses the butyl rubber to bond the two glass pieces together to form a hollow cavity. The pressed semi-finished product is then transferred to the sealant applicator to fill the gap between the outer edges of the two glass pieces with two-component sealant. After the sealant has completely cured, the finished insulated glass is formed.

[0130] The first image acquisition parameters include first light source parameters, first camera parameters, first imaging mode, first acquisition position and angle, and first exposure parameters. The first light source parameters adopt a backlight transmission illumination method, in which light passes perpendicularly through the thickness direction of the hollow glass along the optical axis, so that the refractive index difference caused by the existence of the compression gap in the first region is converted into grayscale difference in the transmission image.

[0131] In one embodiment, the first region void defect detection module 12 is used for:

[0132] According to the first preprocessing strategy, the first region image set is sequentially subjected to image denoising, illumination correction, contrast enhancement and region of interest extraction to obtain the first enhanced region image set.

[0133] Based on the first enhanced region image set, the local grayscale deviation value is calculated pixel by pixel, and the candidate defect pixels are segmented according to the local adaptive threshold to generate the candidate defect pixel region.

[0134] The candidate defect pixel regions are sequentially subjected to closing and opening operations to generate at least one candidate defect region in the form of a connected component. The geometric feature parameters of each candidate defect region are extracted. Candidate defect regions that meet the preset first judgment conditions are judged as compression gap defects. The preset first judgment conditions include that the area of ​​the candidate defect region is greater than the preset first area threshold, the maximum diameter is greater than the preset first diameter threshold, and the aspect ratio of the minimum bounding rectangle is less than the preset first aspect ratio threshold.

[0135] Summarize all candidate defect regions identified as compression void defects, and output the void defect detection results for the first region. The detection results include the number of void defects, the area of ​​each void defect, its location coordinates, and its classification level.

[0136] In one embodiment, the second region image acquisition module 13 is used for:

[0137] Configure second image acquisition parameters, which include second light source parameters, second camera parameters, second imaging mode, second acquisition position and angle, and second exposure parameters. The second light source parameters adopt a low-angle reflected light illumination method, with the light illuminating the outer edge surface of the insulating glass at an angle of 15 to 30 degrees, so that surface defects in the second area caused by poor filling or incomplete curing of the two-component sealant are converted into grayscale differences or texture differences in the reflected image. The second camera parameters include a second image resolution.

[0138] The average area of ​​the first void defect and the proportion of the first void defect area are calculated based on the area of ​​each void defect in the void defect detection results of the first region. The severity coefficient of the first void defect is determined by combining the number of void defects.

[0139] The ratio of the severity coefficient of the first void defect to the preset first standard void defect severity coefficient is used as the second correction coefficient.

[0140] The product of the second correction coefficient and the second image resolution is rounded down to obtain the second adapted image resolution, and the second adapted image acquisition parameters are generated by combining the other fixed parameters in the second image acquisition parameters.

[0141] The second region of the insulating glass is image acquired according to the second adaptive image acquisition parameters to obtain the second region image set.

[0142] The second adaptation image preprocessing intensity is set according to the second correction coefficient, and the second region image set is enhanced to generate the second enhanced region image set.

[0143] The second adaptation image preprocessing intensity is set according to the second correction coefficient, wherein the second adaptation image preprocessing intensity is positively correlated with the second correction coefficient, and the second adaptation image preprocessing intensity is mild enhancement, moderate enhancement, and strong enhancement.

[0144] In one embodiment, the second region void defect detection module 14 is used for:

[0145] The ratio of the preset first standard void defect severity coefficient to the first void defect severity coefficient is used as the condition parameter correction coefficient.

[0146] The second area threshold, second diameter threshold, and second aspect ratio threshold in the preset second determination condition are corrected according to the condition parameter correction coefficient to obtain the second adaptation area threshold, second adaptation diameter threshold, and second adaptation aspect ratio threshold, which are used as the second adaptation determination condition.

[0147] Specifically, based on the second adaptation determination condition, glass lamination gap defect detection is performed on the second enhanced region image set to generate a second region gap defect detection result, including:

[0148] In the second enhanced region image set, the gray-level gradient magnitude is calculated pixel by pixel, and the gray-level gradient magnitude is compared with a preset gradient threshold. Pixels with gray-level gradient magnitude greater than the preset gradient threshold are marked as candidate defect pixels. All marked candidate defect pixels constitute the candidate defect pixel region.

[0149] Morphological processing is performed on the candidate defect pixel regions to generate at least one candidate defect region in the form of a connected component. Geometric feature parameters of each candidate defect region are extracted, including region area, maximum diameter and minimum aspect ratio of the bounding rectangle.

[0150] The candidate defect regions are determined according to the second adaptation determination condition: when a candidate defect region simultaneously satisfies the following conditions: the region area is greater than the second adaptation area threshold, the maximum diameter is greater than the second adaptation diameter threshold, and the aspect ratio of the minimum bounding rectangle is less than the second adaptation aspect ratio threshold, the candidate defect region is determined to be the second region compression void defect.

[0151] Summarize all candidate defect regions identified as second region lamination void defects, and output the second region void defect detection results.

[0152] Compared to existing technologies, this invention firstly uses a first-region image acquisition module 11 to set image acquisition parameters specifically tailored to the structural characteristics of the first region, ensuring image quality and enabling effective imaging of the butyl sealant layer located between the two glass panes, providing reliable raw data for accurate detection. Secondly, the first-region void defect detection module 12 utilizes a first preprocessing strategy for denoising, correction, enhancement, and region extraction, eliminating various interference factors during image acquisition and improving the quality and usability of the first-region image. Through pixel-by-pixel local grayscale deviation analysis and local adaptive threshold segmentation, it achieves high-sensitivity detection of lamination void defects, outputting quantitative detection results including defect quantity, area, location coordinates, and grade classification. Thirdly, the second-region image acquisition module 13 establishes a correlation between the first-region detection results and the second-region detection parameters. By adaptively adjusting the image acquisition resolution and preprocessing intensity of the second region according to the severity of defects in the first region, it optimizes detection resources, ensuring that the image acquisition and enhancement processing of the second region matches the actual quality condition of the first region, improving the targeting and accuracy of the detection. Finally, the second region void defect detection module 14 maps the severity of the first region defect to the correction coefficient of the second region judgment threshold, avoiding the problem of missed detection when the first region defect is severe and false detection when the first region defect is slight when the fixed judgment threshold is used. The result of the detection of the insulating glass lamination defect is obtained, and the systematic detection of lamination void defects of the two-level sealing layer is realized.

[0153] Example 3: Based on the same inventive concept as the visual quality inspection method for insulating glass pressing provided in Example 1, this embodiment of the invention also provides a visual quality inspection device for insulating glass pressing, wherein the visual quality inspection device is a carrier in the execution process of the visual quality inspection method for insulating glass pressing.

Claims

1. A method for visual quality inspection of insulated glass bonding, characterized in that, The methods include: The first region of the insulating glass is image acquired according to the first image acquisition parameters to obtain the first region image set. The insulating glass is a finished insulating glass, and the first region is the butyl rubber sealing layer between the inner surfaces of the two glass panes and the aluminum spacer frame. The first region image set is enhanced according to the first preprocessing strategy, and glass lamination gap defect detection is performed based on the first enhanced region image set. The first region gap defect detection result is output. Based on the detection results of the void defects in the first region, the second adaptive image acquisition parameters and the second adaptive image preprocessing intensity are set, and the second region of the insulating glass is image acquired and image enhanced to generate a second enhanced region image set. The second region is the two-component sealant layer between the outer edges of the two glass panes. Based on the detection results of void defects in the first region, a second adaptation judgment condition is set, and the glass lamination void defect detection is performed on the image set of the second enhanced region according to the second adaptation judgment condition to generate the void defect detection results of the second region.

2. The visual quality inspection method for insulating glass lamination according to claim 1, characterized in that, The automated production process of finished insulating glass includes: the glass to be processed is sequentially sent to the cleaning section for cleaning and air drying. After the surface is confirmed to be clean, an aluminum spacer frame coated with butyl sealant is placed on the surface of the first glass. Then, the second glass is joined with the first glass on both sides of the aluminum frame. The pressing section presses the butyl sealant to bond the two glass pieces together to form an insulating cavity. The pressed semi-finished product is then transferred to a sealant applicator to fill the gap between the outer edges of the two glass pieces with two-component sealant. After the sealant is completely cured, the finished insulating glass is formed.

3. The visual quality inspection method for insulating glass lamination according to claim 1, characterized in that, The first image acquisition parameters include first light source parameters, first camera parameters, first imaging mode, first acquisition position and angle, and first exposure parameters. The first light source parameters adopt a backlight transmission illumination method, in which the light passes perpendicularly through the thickness direction of the hollow glass along the optical axis, so that the refractive index difference caused by the existence of the compression gap in the first region is converted into grayscale difference in the transmission image.

4. The visual quality inspection method for insulating glass lamination according to claim 1, characterized in that, Image enhancement is performed on the first region image set according to the first preprocessing strategy. Glass lamination void defect detection is performed based on the first enhanced region image set, and the void defect detection result of the first region is output, including: According to the first preprocessing strategy, the first region image set is sequentially subjected to image denoising, illumination correction, contrast enhancement and region of interest extraction to obtain the first enhanced region image set. Based on the first enhanced region image set, the local grayscale deviation value is calculated pixel by pixel, and the candidate defect pixels are segmented according to the local adaptive threshold to generate the candidate defect pixel region. The candidate defect pixel regions are sequentially subjected to closing and opening operations to generate at least one candidate defect region in the form of a connected component. The geometric feature parameters of each candidate defect region are extracted. Candidate defect regions that meet the preset first judgment conditions are judged as compression gap defects. The preset first judgment conditions include that the area of ​​the candidate defect region is greater than the preset first area threshold, the maximum diameter is greater than the preset first diameter threshold, and the aspect ratio of the minimum bounding rectangle is less than the preset first aspect ratio threshold. Summarize all candidate defect regions identified as compression void defects, and output the void defect detection results for the first region. The detection results include the number of void defects, the area of ​​each void defect, its location coordinates, and its classification level.

5. The visual quality inspection method for insulating glass lamination according to claim 1, characterized in that, Based on the detection results of void defects in the first region, the second adaptive image acquisition parameters and the second adaptive image preprocessing intensity are set. Image acquisition and image enhancement are performed on the second region of the insulating glass to generate a second enhanced region image set, including: Configure second image acquisition parameters, which include second light source parameters, second camera parameters, second imaging mode, second acquisition position and angle, and second exposure parameters. The second light source parameters adopt a low-angle reflected light illumination method, with the light illuminating the outer edge surface of the insulating glass at an angle of 15 to 30 degrees, so that surface defects in the second area caused by poor filling or incomplete curing of the two-component sealant are converted into grayscale differences or texture differences in the reflected image. The second camera parameters include a second image resolution. The average area of ​​the first void defect and the proportion of the first void defect area are calculated based on the area of ​​each void defect in the void defect detection results of the first region. The severity coefficient of the first void defect is determined by combining the number of void defects. The ratio of the severity coefficient of the first void defect to the preset first standard void defect severity coefficient is used as the second correction coefficient. The product of the second correction coefficient and the second image resolution is rounded down to obtain the second adapted image resolution, and the second adapted image acquisition parameters are generated by combining the other fixed parameters in the second image acquisition parameters. The second region of the insulating glass is image acquired according to the second adaptive image acquisition parameters to obtain the second region image set. The second adaptation image preprocessing intensity is set according to the second correction coefficient, and the second region image set is enhanced to generate the second enhanced region image set.

6. The visual quality inspection method for insulating glass lamination according to claim 5, characterized in that, The second adaptation image preprocessing intensity is set according to the second correction coefficient, wherein the second adaptation image preprocessing intensity is positively correlated with the second correction coefficient, and the second adaptation image preprocessing intensity is mild enhancement, moderate enhancement, and strong enhancement.

7. The visual quality inspection method for insulating glass lamination according to claim 5, characterized in that, Based on the detection results of void defects in the first region, a second adaptation judgment condition is set, including: The ratio of the preset first standard void defect severity coefficient to the first void defect severity coefficient is used as the condition parameter correction coefficient. The second area threshold, second diameter threshold, and second aspect ratio threshold in the preset second determination condition are corrected according to the condition parameter correction coefficient to obtain the second adaptation area threshold, second adaptation diameter threshold, and second adaptation aspect ratio threshold, which are used as the second adaptation determination condition.

8. The visual quality inspection method for insulating glass lamination according to claim 7, characterized in that, Based on the second adaptation determination condition, glass lamination gap defects are detected in the second enhanced region image set, generating a second region gap defect detection result, including: In the second enhanced region image set, the gray-level gradient magnitude is calculated pixel by pixel, and the gray-level gradient magnitude is compared with a preset gradient threshold. Pixels with gray-level gradient magnitude greater than the preset gradient threshold are marked as candidate defect pixels. All marked candidate defect pixels constitute the candidate defect pixel region. Morphological processing is performed on the candidate defect pixel regions to generate at least one candidate defect region in the form of a connected component. Geometric feature parameters of each candidate defect region are extracted, including region area, maximum diameter and minimum aspect ratio of the bounding rectangle. The candidate defect regions are determined according to the second adaptation determination condition: when a candidate defect region simultaneously satisfies the following conditions: the region area is greater than the second adaptation area threshold, the maximum diameter is greater than the second adaptation diameter threshold, and the aspect ratio of the minimum bounding rectangle is less than the second adaptation aspect ratio threshold, the candidate defect region is determined to be the second region compression void defect. Summarize all candidate defect regions identified as second region lamination void defects, and output the second region void defect detection results.

9. A visual quality inspection system for the pressing of insulating glass, characterized in that, A visual quality inspection method for implementing the insulating glass lamination method according to any one of claims 1-8, the system comprising: The first region image acquisition module is used to acquire images of the first region of the insulating glass according to the first image acquisition parameters, and obtain the first region image set. The insulating glass is a finished insulating glass, and the first region is the butyl rubber sealing layer between the inner surfaces of the two glass panes and the aluminum spacer frame. The first region void defect detection module is used to perform image enhancement on the first region image set according to the first preprocessing strategy, perform glass lamination void defect detection based on the first enhanced region image set, and output the first region void defect detection result. The second region image acquisition module is used to set the second adaptive image acquisition parameters and the second adaptive image preprocessing intensity according to the first region void defect detection result, to perform image acquisition and image enhancement on the second region of the insulating glass, and to generate a second enhanced region image set, wherein the second region is a two-component sealant layer between the outer edges of the two glass panes; The second region void defect detection module is used to set a second adaptation judgment condition based on the first region void defect detection result, and to perform glass bonding void defect detection on the second enhanced region image set according to the second adaptation judgment condition, thereby generating a second region void defect detection result.

10. A visual quality inspection device for the pressing of insulating glass, characterized in that, The visual quality inspection equipment includes a visual quality inspection system for insulating glass lamination, serving as a carrier for performing the visual quality inspection method for insulating glass lamination as described in any one of claims 1-8.

Images