Method, Device and Storage Medium for Detecting Appearance Defects of Battery Chip Screen Printing

Abstract

The present invention provides a method for detecting appearance defects in screen printing of solar cells, which performs rough positioning, fine positioning, and brightness correction on the pattern to be detected; uses a standard model to compare the brightness values and brightness variances of each pixel in the image after brightness correction to obtain abnormal regions; and performs defect detection based on the abnormal regions. The standard model is obtained by the following method: Select a feature region and a search region for the screen printing boundary in the template image; use the feature region as a matching object to perform rough positioning on the image to be located; use the affine transformation and sub-pixel edge fitting methods for fine positioning; calculate the shape features and brightness features of the image after brightness correction after brightness correction. The present invention can quickly obtain a standard model for different screen patterns, so that the position and size of the abnormal region can be quickly obtained by means of difference comparison during defect detection, and the detection speed is fast and accurate.

Description

Technical Field

[0001] The present invention belongs to the field of screen printing, and particularly relates to a method, device and storage medium for detecting appearance defects of screen printing on battery wafers. Background Art

[0002] The appearance detection of screen printing for photovoltaic cells is performed after screen printing, and is used to detect printing defects on the surface of battery wafers after screen printing, so as to inspect the printing quality.

[0003] Existing screen printing detection algorithms adopt edge detection methods to extract parts such as main grids and fine grids, and judge whether each part is abnormal, with a relatively slow detection speed. Moreover, for different screen patterns, it is necessary to use complex methods to label different parts of the screen in order to correctly extract each component, which is time-consuming and laborious. In addition, there are also methods for detecting the appearance printing of battery wafers using deep learning or neural networks, but this method requires collecting a large number of samples, and the algorithm is complex, with low practical value for some scenarios. Summary of the Invention

[0004] The purpose of the present invention is to provide a method, device and storage medium for detecting appearance defects of screen printing on battery wafers, which can quickly obtain the position and size of abnormal areas, with a simple method and not affected by screen patterns.

[0005] The technical solution adopted by the present invention is as follows: A method for obtaining a standard model for detecting appearance defects of screen printing on battery wafers, comprising the following steps:

[0006] S1. Coarse positioning

[0007] Obtain a template image and an image to be located by photographing a standard product, and select a feature area with local uniqueness and a search area for the screen printing boundary in the template image;

[0008] Use the feature area with local uniqueness in the template image as a matching object, and use a shape-based template matching method to perform coarse positioning on the image to be located, and obtain the coordinates and angle of the corresponding feature area in the image to be located;

[0009] S2. Fine positioning

[0010] Use the coordinates and angle of the corresponding feature area in the image to be located to affine-transform the image to be located to a position matching the template image; in the screen printing boundary search area, use a sub-pixel edge fitting method to perform fine positioning on the screen printing boundary, and obtain the coordinates of the screen printing boundary in the image to be located, so as to achieve fine positioning of the image to be located;

[0011] S3. Brightness correction

[0012] Perform brightness correction on the image after fine positioning in step S2 using a brightness normalization algorithm;

[0013] S4. Obtain a standard model

[0014] Using the coordinates of the silk screen boundary, align multiple brightness-corrected images, and calculate the shape features and brightness features of the brightness-corrected images. The shape features are the coordinates of the silk screen boundary, and the brightness features include the average brightness and brightness variance of each pixel, as well as the brightness distribution interval of each pixel. The shape features and brightness features constitute the standard model.

[0015] According to the above method, the feature region with local uniqueness is the region in the template image where the edge position and gradient direction have local uniqueness features.

[0016] According to the above method, the shape-based template matching method is specifically: select the feature region with local uniqueness in the template image as the matching object, and search for the feature region with the same features in the image to be located.

[0017] According to the above method, the silk screen boundary is the edge of the silk screen graphic area; the search area of the silk screen boundary is the area range where the silk screen boundary may exist.

[0018] According to the above method, the brightness normalization algorithm in S3 is specifically:

[0019] Taking the convolution kernel size as the sliding window size, perform sliding window calculation on the located image through convolution operation to obtain the brightness weighted value of each local image block, thereby obtaining the brightness distribution map of the entire located image; perform weighted calculation on the original image of the located image and the brightness distribution map to compensate for the brightness difference between each local image block, thereby obtaining an image with a unified brightness distribution as the brightness-corrected image.

[0020] A method for detecting appearance defects of battery cell screen printing, including:

[0021] Perform rough positioning, fine positioning, and brightness correction on the image to be detected;

[0022] Adopt a standard model to compare the brightness value and brightness variance of each pixel of the image to be detected after brightness correction, and obtain the abnormal area; the abnormal area is: at least one of the brightness features or shape features, and the area where the difference from the standard model exceeds the set threshold range;

[0023] Perform defect detection based on the abnormal area;

[0024] The described standard model is obtained through the described standard model acquisition method; the method for rough positioning and fine positioning of the image to be detected is the same as that for the image to be positioned; the method for brightness correction of the image to be detected is the same as that for the brightness correction of the image after fine positioning.

[0025] Further, the defect detection based on the abnormal area is specifically as follows:

[0026] Perform morphological opening and closing operations on the obtained abnormal area, remove the abnormal areas smaller than the set pixel size, and connect the adjacent abnormal areas to obtain a complete defect area; the adjacent abnormal areas refer to the abnormal areas within a certain pixel range.

[0027] Furthermore, after obtaining the complete defect area, judge the defect type, specifically as follows:

[0028] Judge whether the defect belongs to the printing position or on the battery cell according to the gray value and position of the defect area.

[0029] If the defect belongs to the printing position, then compare the direction of the defect area with the preset grid line direction to judge which grid line the defect is located on.

[0030] Combined with the brightness, shape and area of the defect area, judge the specific type of the defect.

[0031] A battery cell screen printing appearance defect detection device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the above-mentioned battery cell screen printing appearance defect detection method is realized.

[0032] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned battery cell screen printing appearance defect detection method is realized.

[0033] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0034] The method of the present invention adopts a registration method based on a standard model, without the need to prepare defective samples. Only a small number of defect-free printed images need to be collected. One of them is used as a template image, and the other several are used as images to be located for training. Then, a standard model can be quickly obtained for different screen patterns. Thus, during defect detection, by means of difference comparison, the position and size of the abnormal area can be quickly obtained, and the detection speed is fast and accurate. During the process of obtaining the standard model and defect detection, both the rough and fine positioning methods are used, which can quickly and accurately locate the product area in a sub-pixel manner and make the printing areas accurately aligned. By adding brightness correction, the problem of poor background uniformity caused by differences in the reflection characteristics of different product surfaces is solved, making the detection results more accurate. By using the method of taking the mean and variance of the standard model, the reasonable variation range of the product can be fully obtained and fault tolerance processing can be carried out, avoiding misjudgment caused by directly using the threshold judgment method. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a flowchart for model training.

[0036] Figure 2 It is a flowchart for defect detection.

[0037] Figure 3 It is a schematic diagram of the template image for rough positioning.

[0038] Figure 4 It is a schematic diagram of the template image after fine positioning.

[0039] In the figure: 1 - template image, 2 - feature area, 3 - screen printing boundary. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0041] The edge detection method is easily affected by the uniformity of the optical imaging and the consistency of the brightness distribution of the product itself, and it is easy to produce mis-extraction when extracting the edge, and the extraction algorithm is affected by different screen patterns. In order to solve the problem of image brightness consistency and quickly adapt to the switching of different screens, the present invention adopts a method based on standard model registration, uses a small amount of defect-free images to train standard model samples, and uses a brightness uniformization algorithm to correct the brightness of the image, thereby avoiding mis-detection caused by brightness changes; in order to solve the influence of different screen patterns on the switching type, this method uses the screen pattern boundary for positioning, alignment and model making, thereby avoiding the use of different grid line extraction methods for different patterns and speeding up the production of different screen models.

[0042] Generally speaking, the printed pattern of a battery cell includes main grid lines and fine grid lines. The main grid lines and the fine grid lines are perpendicular to each other, and the grid lines are perpendicular or parallel to the boundary of the battery cell product.

[0043] The method for detecting appearance defects of screen printing of a battery cell of the present invention comprises two steps: model training and defect detection.

[0044] The first step is model training, the purpose of which is to obtain a standard model for cell screen printing appearance defect detection. The basic process of model training is as follows: Figure 1 As shown, including:

[0045] 1. Make a template image:

[0046] 1) Photograph multiple standard products to obtain multiple images, one of which is used as a template image, and the others are used as images to be positioned. Perform horizontal correction on the template image according to the direction of the grid lines of the battery cell. The grid lines can be either main grid lines or auxiliary grid lines. The purpose is to improve the speed and effect of subsequent preprocessing. Among them, standard products are products that have been printed and are defect-free. It should be noted that it is more troublesome to obtain defective samples, and it is more troublesome to obtain defective samples for each screen for different screens. The present invention only obtains printed defect-free images, and the method is simpler.

[0047] 2) Select a feature region with local uniqueness and a search region for the silk-screen boundary in the template image. In this embodiment, the feature region with local uniqueness is a region in the template image whose edge position and gradient direction have local uniqueness characteristics. The search region for the silk-screen boundary is a region range where the silk-screen boundary may exist, and is composed of a coordinate range. The purpose of setting the search region is to reduce the search range, thereby accelerating the edge search speed. Usually, a coordinate system is established with the center of the template image as the coordinate origin, and the coordinate ranges of the feature region with local uniqueness and the search region for the silk-screen boundary are known.

[0048] 2. Rough positioning:

[0049] Use the feature region with local uniqueness in the template image as the matching object, and use the shape-based template matching method to roughly locate the image to be located, and obtain the coordinates and angles of the corresponding feature region in the image to be located.

[0050] As Figure 3 shown, select the feature region 2 with local uniqueness in the template image as the matching object, and use the shape-based template matching method to search for the feature region with the same features in the image to be located, and obtain the coordinates and angles of the corresponding feature region in the image to be located.

[0051] 3. Fine positioning:

[0052] As Figure 4 shown, use the coordinates and angles of the corresponding feature region in the image to be located to affine-transform the image to be located to a position matching the template image 1; in the screen printing boundary search region, use the sub-pixel edge fitting method to finely position the screen printing boundary, and obtain the coordinates of the screen printing boundary 3 in the image to be located, so as to realize the fine positioning of the image to be located. The screen printing boundary 3 is the edge of the screen printing graphic region, that is, the edge of the region with grid lines.

[0053] Through the method of affine transformation, it can be ensured that the printed patterns on the battery cells are completely aligned. The coordinates of the obtained screen printing boundary are exactly the same in the template image and the image to be located.

[0054] 4. Brightness correction: Use the brightness normalization algorithm to perform brightness correction on the finely positioned image. The method is to estimate the overall brightness distribution map of the image through convolution operation, so as to adjust the background brightness of the image to be consistent according to the brightness distribution map. Specifically:

[0055] Use the convolution kernel size as the sliding window size, perform sliding window calculation on the positioned image through convolution operation, obtain the brightness weighted value of each local image block, and thus obtain the brightness distribution map of the entire positioned image; perform weighted calculation on the original image of the positioned image and the brightness distribution map to compensate for the brightness difference between each local image block, so as to obtain an image with a unified brightness distribution as the image after brightness correction.

[0056] 5. Obtain the standard model

[0057] Use the coordinates of the screen printing boundary to align multiple images after brightness correction (such as by perspective transformation), calculate the shape features and brightness features of the images after brightness correction. The shape features are the coordinates of the screen printing boundary, and the brightness features include the brightness mean and variance of each pixel, as well as the brightness distribution interval of each pixel. The shape features and brightness features constitute the standard model.

[0058] The second step is defect detection. The basic process of defect detection is as Figure 2 shown, including:

[0059] 1. Load the image to be detected (take an image of the battery cell to be detected, which can show the outer contour boundary of the entire battery cell, and then load the image). Use the template image made in the first step to perform rough positioning on the image to be detected. The rough positioning method is the same as the second step of the first step.

[0060] 2. Perform fine positioning on the roughly positioned image to be detected. The fine positioning is the same as the third step of the first step.

[0061] 3. Perform brightness correction on the finely positioned image to be detected to make the background brightness distribution of each part of the image consistent. The brightness correction is the same as the fourth step of the first step.

[0062] 4. Use the standard model trained in the first step to compare the brightness values and brightness variances of each pixel of the image to be detected after brightness correction to obtain the abnormal area; the abnormal area is: at least one of the brightness feature or the shape feature, and the area where the difference from the standard model is greater than the set threshold range.

[0063] 5. Perform defect detection based on the abnormal area. Specifically:

[0064] For the obtained abnormal area, first use morphological opening and closing operations to remove the abnormal areas smaller than the set pixel size (within the allowed range), and connect the adjacent abnormal areas to obtain a complete defect area.

[0065] After obtaining the complete defect area, judge the defect type. Specifically: according to the gray value and position of the defect area, judge whether the defect belongs to the printing position or on the battery cell; if the defect belongs to the printing position, then compare the direction of the defect area with the preset grid line direction to judge which grid line the defect is located on; then combine the brightness, shape and area of the defect area to judge the specific type of the defect; further judge the defect level according to the difference in local contrast between the defect area and the surrounding area.

[0066] In this embodiment, for the detected abnormal gray values and their position information, it is determined whether they belong to the printing position (i.e., the grid line position of the battery cell) or on the battery cell (the printing area is a bright area on the standard model, and the battery cell is a dark area on the standard model), and then the defect type is determined. Through the direction, it can be known whether the defect is located on the main grid line or the secondary grid line (it is set in advance which direction corresponds to the main grid or the secondary grid). Among them, the defect types are divided into multiple types, and the specific defect type can be judged through information such as brightness, shape, area, and position. When the defect is located on the grid line, it includes broken grid (the broken grid position becomes darker), misaligned grid line (the area that should be bright becomes darker, and the area that should be dark becomes brighter), over-wide grid line (the brightness width increases), etc.; when the defect is located in the non-printing area on the battery cell, possible defects include multiple leaks (an additional bright area), in addition, the entire battery cell may have a situation where the chip is broken and part of the battery cell is missing (the brightness is the same as the background brightness). Finally, according to the difference between the defect and the local contrast of its surrounding area, the defect level is further judged.

[0067] The present invention also provides a battery cell screen printing appearance defect detection device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the above-mentioned battery cell screen printing appearance defect detection method is implemented.

[0068] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned battery cell screen printing appearance defect detection method is implemented.

[0069] In summary, the present invention uses the method of model registration to detect screen printing appearance defects, which can quickly and accurately detect printing defects. By using the method of rough positioning combined with fine positioning, it can quickly and accurately locate the product area in a sub-pixel manner, and make the printing area accurately aligned. By adding a brightness correction preprocessing method, the problem of poor background uniformity caused by differences in the surface reflection characteristics of different products can be solved, making the detection results more accurate. By using the method of training with a standard model, the normal change range of the printed matter can be learned, avoiding misjudgment caused by directly using the threshold judgment method.

[0070] Since the detection method of the present application is not only accurate but also has a particularly fast detection speed, this method is particularly suitable for the preliminary defect inspection after the grid line printing on one side of the battery cell is completed, facilitating subsequent processing immediately after printing defects are quickly discovered in the initial stage.

[0071] It should be noted that according to the needs of implementation, each step / component described in the present application can be split into more steps / components, or two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0072] Those skilled in the art can easily understand that the above are only the preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for obtaining a standard model for detecting appearance defects in screen printing of solar cells, characterized in that: It includes the following steps: S1. Coarse positioning Obtain a template image and an image to be positioned by photographing a standard product. Select a feature area with local uniqueness and a search area for the screen printing boundary in the template image; the standard product is a solar cell without defects after printing. Use the feature area with local uniqueness in the template image as the matching object, and use the shape-based template matching method to perform coarse positioning on the image to be positioned, and obtain the coordinates and angle of the corresponding feature area in the image to be positioned. S2. Fine positioning Use the coordinates and angle of the corresponding feature area in the image to be positioned to affine-transform the image to be positioned to a position matching the template image; in the search area of the screen printing boundary, use the sub-pixel edge fitting method to perform fine positioning on the screen printing boundary, and obtain the coordinates of the screen printing boundary in the image to be positioned, so as to realize the fine positioning of the image to be positioned. S3. Brightness correction Use the brightness normalization algorithm to perform brightness correction on the image after fine positioning in step S2; the brightness normalization algorithm is specifically: Taking the convolution kernel size as the sliding window size, perform sliding window calculation on the positioned image through convolution operation to obtain the brightness weighted value of each local image block, so as to obtain the brightness distribution map of the entire positioned image; perform weighted calculation on the original image of the positioned image and the brightness distribution map to compensate for the brightness difference between each local image block, so as to obtain an image with a unified brightness distribution as the image after brightness correction. S4. Obtain the standard model Use the coordinates of the screen printing boundary to align multiple images after brightness correction, and calculate the shape features and brightness features of the images after brightness correction. The shape features are the coordinates of the screen printing boundary, and the brightness features include the average brightness and brightness variance of each pixel, as well as the brightness distribution interval of each pixel. The shape features and brightness features constitute the standard model.

2. The standard model obtaining method according to claim 1, characterized in that: The feature area with local uniqueness is an area in the template image where the edge position and gradient direction have local uniqueness features.

3. The standard model obtaining method according to claim 1 or 2, characterized in that: The shape-based template matching method is specifically: select the feature area with local uniqueness in the template image as the matching object, and search for the feature area with the same features in the image to be positioned.

4. The standard model obtaining method according to claim 1, characterized in that: The screen printing boundary is the edge of the screen printing graphic area; the search area of the screen printing boundary is the area range where the screen printing boundary may exist.

5. A method for detecting appearance defects in screen printing of solar cells, characterized in that: It includes: Perform coarse positioning, fine positioning and brightness correction on the image to be detected; Adopt the standard model to compare the brightness value and brightness variance of each pixel of the image to be detected after brightness correction, and obtain the abnormal area; the abnormal area is: at least one of the brightness features or shape features, and the area where the standard model has a deviation greater than the set threshold range; Perform defect detection according to the abnormal area; The standard model described above is obtained by the standard model acquisition method described in any one of claims 1 to 4; the method for rough positioning and fine positioning of the image to be detected is the same as that for the image to be positioned; the method for brightness correction of the image to be detected is the same as that for the brightness correction of the image after fine positioning.

6. The method for detecting the appearance defects of screen printing on solar cells according to claim 5, characterized in that: the defect detection according to the abnormal area is specifically: using morphological opening operation and closing operation on the obtained abnormal area, removing the abnormal areas smaller than the set pixel size, and connecting the adjacent abnormal areas to obtain a complete defect area; the adjacent abnormal areas refer to the abnormal areas within a certain pixel range.

7. The method for detecting the appearance defects of screen printing on solar cells according to claim 6, characterized in that: after obtaining the complete defect area, judging the defect type, specifically: judging whether the defect belongs to the printing position or on the solar cell according to the gray value and position of the defect area; if the defect belongs to the printing position, then comparing the direction of the defect area with the preset grid line direction to judge which grid line the defect is located on; further combining the brightness, shape and area of the defect area to judge the specific type of the defect.

8. A device for detecting the appearance defects of screen printing on solar cells, characterized in that: comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the computer program, it implements the method for detecting the appearance defects of screen printing on solar cells according to any one of claims 5-7 above.

9. A computer-readable storage medium, on which a computer program is stored, characterized in that: when the computer program is executed by the processor, it implements the method for detecting the appearance defects of screen printing on solar cells according to any one of claims 5-7 above.

Images