Appearance inspection method for photovoltaic glass modules after second laying

Abstract

The present application discloses a method for inspecting the appearance of photovoltaic glass components after two-pass laying, comprising the following steps: along the conveying direction of the assembly line, a plurality of images of photovoltaic glass components are sequentially acquired by a plurality of cameras on the two-pass laying assembly line, and two adjacent images have overlapping areas; the images are cropped using a straight line perpendicular to the conveying direction at a certain point in the overlapping area as a cropping reference line, and then the cropped images are spliced ​​in the order of the cameras to form a large image; the large image is split according to a predetermined step size using a pixel frame of a predetermined size to form small images with a field of view overlapping by 100 pixels, and then these small images are sequentially fed into a defect detection algorithm, i.e., Fai, for abnormality detection, wherein the abnormalities include foreign matter, dirt, and breakage. The method for inspecting the appearance of photovoltaic glass components after two-pass laying provided by the present application can ensure the quality of inspection without affecting the production efficiency of photovoltaic glass components, so that the missed detection rate and the false detection rate are less than 1‰.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic component detection, and in particular to a method for detecting the appearance of a photovoltaic glass component after second-layer paving. Background Art

[0002] There is a two-step laying process in the processing of photovoltaic glass components, that is, first the glass components are put into the EVA punching, cutting and laying machine, and then the EVA film with holes punched and cut to a fixed length is clamped by the robotic arm in the EVA punching, cutting and laying machine and laid on the surface of the glass components. Then the glass components with EVA film laid flow into the backboard punching, cutting and laying machine, and the TPT backboard with holes punched and cut to a fixed length is laid on the surface of the glass components pre-laid with EVA film by the robotic arm in the backboard punching, cutting and laying machine.

[0003] Currently, there are many quality inspection requirements in the secondary laying process of photovoltaic glass modules, such as:

[0004] Since EVA is a flexible material, if the clamped EVA film is not flat, the cut will be uneven and roughly jagged, which does not meet the quality requirements;

[0005] When EVA and TPT are laid on the surface of the glass assembly, the busbars on the glass assembly are not perforated;

[0006] The size of EVA and TPT should be the same as that of the glass component, or there is only a very small error. However, due to personnel settings, the cutting machine may cut the EVA and TPT to an inappropriate size;

[0007] In addition, there are also problems such as EVA and TPT being laid unevenly, and the presence of foreign matter, oil stains, and damage in EVA and TPT, all of which do not meet quality requirements.

[0008] If manual inspection is used to check quality, it will not only require a lot of manpower and time, affecting production efficiency and increasing inspection costs, but there will also be problems of missed inspections and false inspections. Summary of the Invention

[0009] One advantage of the present invention is that it provides a method for detecting the appearance of photovoltaic glass components after the second laying process. After the photovoltaic glass components are laid with EVA and TPT, an artificial intelligence-based quality inspection is performed. This method not only does not affect the normal processing rhythm of the photovoltaic glass components, but also ensures production efficiency. At the same time, it can also ensure the detection quality, making the missed detection rate and false detection rate less than 1‰, and can detect defects as small as 0.5mm. 2 defects, can significantly reduce the cost of detection.

[0010] To achieve at least one of the above advantages of the present invention, the present invention provides a method for inspecting the appearance of a photovoltaic glass assembly after second-layer laying, comprising the following steps:

[0011] S10, along the conveying direction of the assembly line, using multiple cameras to sequentially obtain multiple images of photovoltaic glass components on the second laying assembly line, and two adjacent images have overlapping areas;

[0012] S20, cropping the image using a straight line perpendicular to the conveying direction at a point in the overlapping area as a cropping reference line, and then stitching the cropped images together in the order of the cameras to form a large image;

[0013] S30, using a pixel frame of a predetermined size, splitting the image according to a predetermined step size, splitting the large image into small images with a field of view overlapping by 100 pixels, and then feeding these small images into a defect detection algorithm, i.e., Fai, to detect abnormalities, where the abnormalities include foreign matter, dirt, and damage;

[0014] For foreign body detection, the Fai adds a corner point detection algorithm to search for the number of corner points in the outline of the small image. When the number of corner points is greater than a first threshold, it is determined that there is a foreign body in the photovoltaic glass component and an NG result is output; otherwise, an OK result is output;

[0015] For dirt detection, the thumbnail image is binarized to find the dirt pixels. If the area of ​​the dirt pixels is greater than a second threshold, the photovoltaic glass module is judged to be dirty and an NG result is output; otherwise, an OK result is output.

[0016] For damage detection, based on the corner point detection algorithm, the number of corner points is searched in the outline of the small image. If the total number of corner points in the first and last images in each row of the small image is greater than a third threshold, the photovoltaic glass component is determined to be damaged and an NG result is output; otherwise, an OK result is output;

[0017] S40, transmit the test result to the pipeline PLC for diversion operation: when the PLC receives an OK result, it flows normally to the next process; when the PLC receives an NG result, it is diverted to the abnormal buffer rack and waits for processing.

[0018] According to an embodiment of the present invention, when performing foreign matter detection and damage detection, the thumbnail is first scaled, the image gradient is calculated, and then a contour image of the thumbnail is obtained based on the image gradient.

[0019] According to an embodiment of the present invention, when performing dirt detection, the thumbnail is first scaled, and then a binarization operation is performed based on the dirt grayscale interval on the scaled thumbnail.

[0020] According to one embodiment of the present invention, based on Fai, the abnormal situation also includes: punching position detection, three ROI coordinates are preset, and the preset ROI coordinates are used as the base point to radiate to the outer peripheral side to form a punching detection range, and are preliminarily cut into three punching detection images, and then the punching position is detected by edge detection, and the detected punching position is compared with the standard position preset by the system. If it exceeds the fourth threshold, it is determined that the punching position is deviated and an NG result is output, otherwise an OK result is output.

[0021] According to an embodiment of the present invention, in punching position detection, the punching detection image is scaled, and then the image gradient is calculated to obtain a contour image of the punching detection image, wherein the contour with the largest area is the contour of the detected hole.

[0022] According to one embodiment of the present invention, after Fai detection, the defect name and the prediction frame coordinate information corresponding to the defect name are obtained, and the prediction frame coordinates are converted to form the large image prediction frame coordinates, wherein the defect name corresponds one-to-one to the large image prediction frame coordinates.

[0023] According to an embodiment of the present invention, the method of converting the prediction frame coordinates into the large image prediction frame coordinates is:

[0024] Pori_x=i*1380

[0025] Pori_y=j*1380

[0026] Pori=(Pori_x, Pori_y)

[0027] Among them, Pori_x is the x coordinate of the large image corresponding to the origin of the small image, Pori_y is the y coordinate of the large image corresponding to the origin of the small image, Pori is the x and y coordinates of the large image corresponding to the origin of the small image, where i is the column number, from 0 to n, where j is the row number, from 0 to n, and the pixels of the predetermined size pixel frame are 1380*1380.

[0028] According to an embodiment of the present invention, step S30 further includes detecting the deviation of the TPT and EVA positions, specifically including:

[0029] S311, calibrating the millimeter pixel ratio, obtaining the actual length of the photovoltaic glass assembly and the pixel length in the image of the photovoltaic glass assembly obtained by the camera, and calculating the millimeter pixel ratio;

[0030] S312, calculate the offset, capture the image of the edge of the photovoltaic glass component through ROI, then perform contrast enhancement, filtering and image gradient calculation on the ROI screenshot to obtain an edge map, and then use the length threshold to filter out the miscellaneous lines to obtain the glass edge and the EVA or TPT edge, and then calculate and determine whether the distance between the glass edge and the points on the EVA edge or TPT edge exceeds the offset threshold in the x direction or y direction. If it exceeds the threshold, it is determined that there is a paving deviation defect and an NG result is output, otherwise an OK result is output.

[0031] According to an embodiment of the present invention, step S30 further includes detecting whether a busbar is not perforated, specifically including:

[0032] Set vertical lighting facing the production line and analyze the grayscale value within a predetermined circular range with the punching hole as the center. If the imaging grayscale value is lower than the imaging threshold, it is determined that the photovoltaic glass component has a busbar unperforated defect and an NG result is output; otherwise, an OK result is output.

[0033] According to an embodiment of the present invention, analyzing the grayscale values ​​within a predetermined circular range with the punch hole as the center specifically includes:

[0034] S321, find the locations of multiple punching holes through Fai;

[0035] S322, expanding outward by 100 pixels with the punch hole as the center point to form a range frame corresponding to each punch hole;

[0036] S323, sorting the grayscale values ​​of the pixels within each range box, removing the smallest 10% of the values ​​and the largest 10% of the values, and then outputting the average of the remaining values;

[0037] S324, analyzing the mean grayscale value of each pixel in the range frame to determine whether it is lower than the imaging threshold; if it is lower than the imaging threshold, determining that the photovoltaic glass assembly has a busbar non-perforated defect.

[0038] These and other objects, features and advantages of the present invention will be fully reflected in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A flow chart of a method for inspecting the appearance of photovoltaic glass components after second-layer laying according to a preferred embodiment of the present application is shown. DETAILED DESCRIPTION

[0040] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are for illustrative purposes only, and those skilled in the art will readily appreciate other obvious variations. The basic principles of the present invention defined in the following description may be applied to other embodiments, variations, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the present invention.

[0041] Those skilled in the art should understand that, in the disclosure of the specification, the terms "longitudinal", "transverse", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like to indicate the orientation or position relationship are based on the orientation or position relationship shown in the accompanying drawings, which are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the above terms should not be understood as limiting the present invention.

[0042] It is to be understood that the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element may be one, while in another embodiment, the number of the elements may be multiple, and the term "one" should not be understood as a limitation on the quantity.

[0043] During the processing of photovoltaic glass modules, problems with incoming materials or laying can cause substandard quality. For example, EVA and TPT may contain foreign matter, damage, oil stains, and dirt. Low-precision laying equipment can also lead to problems such as laying offsets and unperforated busbars. Relying on manual inspection to ensure processing quality will not only significantly reduce work efficiency and increase inspection costs, but also pose risks of missed inspections and false detections.

[0044] In addition, if a machine vision inspection system is installed in the laying machine to perform photo quality inspection after EVA or TPT laying is completed, the laying machine will wait for the inspection system to give a judgment result before proceeding to the next step, which will increase the production cycle by an additional 3s to 8s and also reduce work efficiency.

[0045] Therefore, a quality inspection method for photovoltaic glass components after the second laying is proposed, that is, the appearance quality inspection is carried out after the EVA and TPT laying are completed at the same time. In this way, the normal production cycle will not be occupied, and the quality of photovoltaic glass components can be ensured on the basis of ensuring production efficiency. At the same time, the inspection cost is low, the missed detection rate and the false detection rate can be reduced to less than 1‰, and the detection rate as small as 0.5mm can be detected. 2 defects.

[0046] Specifically, refer to Figure 1A preferred embodiment of the present application provides a method for inspecting the appearance of a photovoltaic glass assembly after two layers of paving, comprising the following steps:

[0047] S10, along the conveying direction of the assembly line, multiple cameras are used to sequentially obtain multiple images of photovoltaic glass components on the second-laid assembly line, and adjacent images have overlapping areas. The camera brackets that fix the cameras are connected to objects with little or no vibration as much as possible to ensure that the cameras do not come into contact with the assembly line with severe shaking. In addition, the need to add sunshades is determined based on the lighting conditions at the project site. At the same time, during the camera installation process, it is ensured that the images obtained by adjacent cameras have a partial overlapping field of view, so that the complete glass components can be captured when the puzzle is assembled. A long tape measure can be pulled along the installation direction of the several cameras, and the camera positions can be adjusted by looking at the tape measure scale to ensure that the adjacent cameras have a partial overlapping field of view.

[0048] S20, cropping the image using a straight line perpendicular to the conveying direction at a certain point (this point is a custom point) in the overlapping area as a cropping reference line, and then stitching the cropped multiple images together in the order of the cameras to form a large image. The pixels of the stitched large image are approximately 14000*11000;

[0049] S30: Since image processing on a large image takes a very long time, the large image is first split into multiple small images. The images are split using a predetermined pixel frame size and a predetermined step size. For example, a 1380*1380 pixel frame is split using a step size of 1280. The large image is split into dozens of small images with a field of view overlapping by 100 pixels. These small images are then sequentially fed into a defect detection algorithm, i.e., Fai, for abnormality detection, where the abnormalities include foreign matter, dirt, and damage.

[0050] Regarding foreign object detection, in general, these foreign objects are caused by the outer shell paper of EVA or TPT packaging being accidentally rolled into the material and torn into shreds. Shredded paper has many corners, while the surface of a normal glass component is a smooth plane without corners. Therefore, foreign object detection can be performed based on this feature. That is, the Fai has added a corner detection algorithm. To reduce the amount of calculation, the image is first scaled to a smaller image of 640*640 size, the image gradient is obtained, and a contour map is obtained. The number of corner points is then searched in the contour map of the smaller image. When the number of corner points is greater than a first threshold, it is determined that the photovoltaic glass component has foreign objects. Fai will frame the foreign object position according to the corner point coordinates and output an NG result. Otherwise, it will output an OK result.

[0051] For dirt detection, dirt is generally small black dots, which may be caused by oil on EVA or TPT during the loading process, or other reasons. However, they are all dark in the image. Therefore, the image is first scaled to 640*640 to reduce the amount of calculation. Then, the dirt grayscale interval on the thumbnail image or the scaled thumbnail is binarized to find the dirt pixel. When the area of ​​the dirt pixel is greater than the second threshold, it is determined that the photovoltaic glass module is dirty. Fai box out the dirt area and output NG result, otherwise it outputs OK result.

[0052] Among them, for damage detection, most of the damage is caused by the jagged shape when the robot arm clamps EVA or TPT for cutting due to uneven cuts. Because the feeding direction is the short side, the cutting position is also the short side, so this type of defect is at the short side of EVA or TPT. In the image, it is the first and last pictures in each row of the split small pictures. If the cut is uneven, there will be a large number of jagged edges, and the number of corner points will be much more than normal. Therefore, based on the corner point detection algorithm, the image is first scaled to 640*640 size, the image gradient is obtained, and the contour map is obtained. Then, the number of corner points is searched in the contour map of the small picture. If the total number of corner points in the first and last pictures in each row of the small picture is greater than the third threshold, it is determined that the photovoltaic glass component is damaged, and Fai outputs an NG result, otherwise it outputs an OK result;

[0053] S40, transmit the test results to the assembly line PLC for diversion operation: when the PLC receives an OK result, the photovoltaic glass components flow normally into the next process; when the PLC receives an NG result, the photovoltaic glass components are diverted to the abnormal buffer rack and wait for processing.

[0054] In one embodiment, based on Fai, the abnormal situation also includes: punching position detection. Since the punching position is relatively fixed, which is three holes in the middle of the glass component, three ROI coordinates can be preset, and the preset ROI coordinates are used as the base point to radiate to the outer peripheral side to form a punching detection range, and preliminarily cut into three 1280*1280 punching detection images, and then the position of the punching is accurately detected and found by edge detection. The detected position of the punching is then compared with the standard position preset by the system. If it exceeds the fourth threshold, the punching position is determined to be deviated and an NG result is output, otherwise an OK result is output. Since the middle of the photovoltaic glass component is a smooth surface, and there will be an obvious boundary line at the position of the hole, the position of the punching can be found by edge detection.

[0055] Further preferably, in the punching position detection, in order to further reduce the amount of calculation and improve the detection accuracy, the punching detection image is scaled to a size of 640*640, and then the image gradient is calculated to obtain a contour image of the punching detection image, where the contour with the largest area is the contour of the detected hole.

[0056] In addition, after Fai detection, the defect name and the prediction frame coordinate information corresponding to the defect name are obtained. Since the prediction is performed on the split small image, it is necessary to convert the prediction frame coordinates of the small image into the prediction frame coordinates of the large image, where the defect name corresponds one-to-one with the prediction frame coordinates of the large image, so that it is convenient to view the position of the defect corresponding to the defect name in the image, and then facilitate the corresponding processing of the NG photovoltaic glass component in the subsequent cache position.

[0057] Further preferably, the method of converting the prediction frame coordinates into the large image prediction frame coordinates is:

[0058] Pori_x=i*1380

[0059] Pori_y=j*1380

[0060] Pori=(Pori_x, Pori_y)

[0061] Among them, Pori_x is the x coordinate of the large image corresponding to the origin of the small image, Pori_y is the y coordinate of the large image corresponding to the origin of the small image, Pori is the x and y coordinates of the large image corresponding to the origin of the small image, where i is the column number, from 0 to n, where j is the row number, from 0 to n, and the pixels of the predetermined size pixel frame are 1380*1380.

[0062] As a preferred embodiment, step S30 further includes detecting the position deviation of the TPT and EVA, thereby determining whether the TPT and EVA are actually deviated by calculating the offset of the TPT and EVA, wherein the detecting the position deviation of the TPT and EVA specifically includes:

[0063] S311, calibrating the millimeter pixel ratio, obtaining the actual length of the photovoltaic glass assembly and the pixel length in the image of the photovoltaic glass assembly obtained by the camera, and calculating the millimeter pixel ratio;

[0064] In one embodiment, the calculation formula is as follows:

[0065] Rate=Lmm / Lpix

[0066] Where Rate is the millimeter pixel ratio, Lmm is the component size in millimeters, and Lpix is ​​the component pixel length in pixels;

[0067] Other fixed-size calibration objects can also be used instead;

[0068] S312, calculate the offset. First, in order to reduce the amount of calculation, the image of the edge of the photovoltaic glass component is captured through ROI. Then, the ROI screenshot is contrast enhanced, filtered to remove noise, and the image gradient is calculated to obtain an edge map. Then, the length threshold is used to filter out the miscellaneous lines. After processing, the glass edge and the EVA or TPT edge are obtained. Then, the distance between the glass edge and the points on the EVA edge or TPT edge is calculated and judged whether it exceeds the offset threshold in the x direction or y direction. If the edge is in the left and right directions of the image, the distance in the x direction is compared. If the edge is in the up and down directions of the image, the distance in the y direction is compared. If it exceeds the threshold, it is determined that there is a paving deviation defect and an NG result is output. Otherwise, an OK result is output.

[0069] In the process of calculating the offset in step S312, the offset threshold can be adjusted so as to be suitable for different judgment criteria. In addition, if the cutting size of EVA and TPT is inconsistent with the glass component, one end of the TPT or EVA will definitely be offset, and such defects can also be treated as offset defects.

[0070] As a preferred embodiment, step S30 also includes busbar non-perforation detection. The busbar without punching will push the EVA or TPT into a curved surface. If it is a normally perforated EVA or TPT, the area around the punching is flat, the imaging is bright and the light is uniform. If it is a curved surface, the imaging of the curved surface is dark and the light is uneven. Therefore, the presence of a busbar non-perforation defect can be distinguished by analyzing the grayscale value around the punching. The busbar non-perforation detection specifically includes:

[0071] Set vertical lighting facing the production line and analyze the grayscale value within a predetermined circular range with the punching hole as the center. If the imaging grayscale value is lower than the imaging threshold, it is determined that the photovoltaic glass component has a busbar unperforated defect and an NG result is output; otherwise, an OK result is output.

[0072] Further preferably, the analyzing the grayscale value within a predetermined circular range with the punching hole as the center specifically includes:

[0073] S321, find the positions of multiple punching holes through Fai, such as the positions of three punching holes;

[0074] S322, expanding outward by 100 pixels with the punch hole as the center point to form a range frame corresponding to each punch hole;

[0075] S323, sorting the grayscale values ​​of the pixels within each range box, removing the smallest 10% of the values ​​and the largest 10% of the values, and then outputting the average of the remaining values, wherein the range of the removed values ​​is adjustable, and may be 5% or 20%, etc.;

[0076] S324, analyzing the mean grayscale value of each pixel in the range frame to determine whether it is lower than the imaging threshold; if it is lower than the imaging threshold, determining that the photovoltaic glass assembly has a busbar non-perforated defect.

[0077] In addition, for the appearance inspection after the second paving, the calculation results, inspection diagram and detailed description of the inspection content can be displayed on the software display interface.

[0078] It should be noted that the terms "first, second, third and fourth" in the present invention are only used for descriptive purposes and do not represent any order. They cannot be understood as indicating or implying relative importance. These terms can be interpreted as names.

[0079] Those skilled in the art will appreciate that the embodiments of the present invention described above and shown in the accompanying drawings are intended to be illustrative only and are not intended to limit the present invention. The advantages of the present invention have been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and illustrated in the embodiments. Any variations or modifications may be made to the embodiments of the present invention without departing from the principles described.

Claims

1. A method for inspecting the appearance of photovoltaic glass components after second-layer laying, characterized in that: The following steps are involved: S10, along the conveying direction of the assembly line, using multiple cameras to sequentially obtain multiple images of photovoltaic glass components on the second laying assembly line, and two adjacent images have overlapping areas; S20, cropping the image using a straight line perpendicular to the conveying direction at a point in the overlapping area as a cropping reference line, and then stitching the cropped images together in the order of the cameras to form a large image; S30, using a pixel frame of a predetermined size, splitting the image according to a predetermined step size, splitting the large image into small images with a field of view overlapping by 100 pixels, and then feeding these small images into a defect detection algorithm, i.e., Fai, to detect abnormalities, where the abnormalities include foreign matter, dirt, and damage; For foreign body detection, the Fai adds a corner point detection algorithm to search for the number of corner points in the outline of the small image. When the number of corner points is greater than a first threshold, it is determined that there is a foreign body in the photovoltaic glass component and an NG result is output; otherwise, an OK result is output; For dirt detection, the thumbnail image is binarized to find the dirt pixels. If the area of ​​the dirt pixels is greater than a second threshold, the photovoltaic glass module is judged to be dirty and an NG result is output; otherwise, an OK result is output. For damage detection, based on the corner point detection algorithm, the number of corner points is searched in the outline of the small image. If the total number of corner points in the first and last images in each row of the small image is greater than a third threshold, the photovoltaic glass component is determined to be damaged and an NG result is output; otherwise, an OK result is output; S40, transmit the test results to the assembly line PLC for diversion operation: when the PLC receives the OK result, it flows normally to the next process; When the PLC receives an NG result, it is diverted to the abnormal buffer rack and waits for processing; After Fai detection, the defect name and the prediction frame coordinate information corresponding to the defect name are obtained, and the prediction frame coordinates are converted to form the large image prediction frame coordinates, where the defect name corresponds to the large image prediction frame coordinates one by one; The method of converting the predicted frame coordinates into the large image predicted frame coordinates is: Pori_x=i*1380 Pori_y=j*1380 Pori=(Pori_x, Pori_y) Among them, Pori_x is the x coordinate of the large image corresponding to the origin of the small image, Pori_y is the y coordinate of the large image corresponding to the origin of the small image, Pori is the x and y coordinates of the large image corresponding to the origin of the small image, where i is the column number, from 0 to n, where j is the row number, from 0 to n, and the pixels of the predetermined size pixel frame are 1380*1380.

2. The method for inspecting the appearance of a photovoltaic glass assembly after second-layer laying according to claim 1, characterized in that: When performing foreign matter detection and damage detection, the small image is first scaled, the image gradient is calculated, and then the contour image of the small image is obtained based on the image gradient.

3. The method for inspecting the appearance of a photovoltaic glass assembly after second-layer laying according to claim 1, characterized in that: When performing dirt detection, the small image is first scaled, and then a binarization operation is performed based on the dirt grayscale interval on the scaled small image.

4. The method for detecting the appearance of a photovoltaic glass assembly after second-layer laying according to claim 1, characterized in that: Based on Fai, the abnormal situation also includes: punching position detection, three ROI coordinates are preset, and the punching detection range is formed by radiating to the outer peripheral side with the preset ROI coordinates as the base point, and preliminarily cutting into three punching detection images, and then detecting the position of the punching by edge detection, comparing the detected punching position with the standard position preset by the system. If it exceeds the fourth threshold, it is determined that the punching position is deviated and the NG result is output, otherwise the OK result is output.

5. The method for inspecting the appearance of a photovoltaic glass assembly after second-layer laying as claimed in claim 4, characterized in that: In the punching position detection, the punching detection image is scaled, and then the image gradient is calculated to obtain a contour image of the punching detection image, wherein the contour with the largest area is the contour of the detected hole.

6. The method for inspecting the appearance of a photovoltaic glass assembly after second-layer laying according to claim 1, characterized in that: In step S30, the TPT and EVA position deviation detection is also included, specifically including: S311, calibrating the millimeter pixel ratio, obtaining the actual length of the photovoltaic glass assembly and the pixel length in the image of the photovoltaic glass assembly obtained by the camera, and calculating the millimeter pixel ratio; S312, calculate the offset, capture the image of the edge of the photovoltaic glass component through ROI, then perform contrast enhancement, filtering and image gradient calculation on the ROI screenshot to obtain an edge map, and then use the length threshold to filter out the miscellaneous lines to obtain the glass edge and the EVA or TPT edge, and then calculate and determine whether the distance between the glass edge and the points on the EVA edge or TPT edge exceeds the offset threshold in the x direction or y direction. If it exceeds the threshold, it is determined that there is a paving deviation defect and an NG result is output, otherwise an OK result is output.

7. The method for inspecting the appearance of a photovoltaic glass assembly after second-layer laying according to claim 1, characterized in that: In step S30, busbar non-perforation detection is also included, specifically including: Set vertical lighting facing the production line and analyze the grayscale value within a predetermined circular range with the punching hole as the center. If the imaging grayscale value is lower than the imaging threshold, it is determined that the photovoltaic glass component has a busbar unperforated defect and an NG result is output; otherwise, an OK result is output.

8. The method for inspecting the appearance of a photovoltaic glass assembly after second-layer laying as claimed in claim 7, characterized in that: The analyzing of the grayscale values ​​within a predetermined circular range with the punch hole as the center specifically includes: S321, find the locations of multiple punching holes through Fai; S322, expanding outward by 100 pixels with the punch hole as the center point to form a range frame corresponding to each punch hole; S323, sorting the grayscale values ​​of the pixels within each range box, removing the smallest 10% of the values ​​and the largest 10% of the values, and then outputting the average of the remaining values; S324, analyzing the mean grayscale value of each pixel in the range frame to determine whether it is lower than the imaging threshold; if it is lower than the imaging threshold, determining that the photovoltaic glass assembly has a busbar non-perforated defect.

Images