Die cutting sheet online AOI detection method based on AI vision sorting

Abstract

The present application relates to the technical field of image data processing, and more particularly to a die-cutting sheet continuous line AOI detection method based on AI vision sorting, which obtains texture interval and texture gray scale to generate texture characteristic coefficient based on the collected texture image of the surface of the die-cuting sheet, compares the pixel features of the texture image and the preset standard texture image to determine whether there is distortion tendency in the AOI detection; obtains the pixel difference coefficient of the die-cutting sheet with weak deviation tendency; collects the scratch image of the die-cutting sheet with existing parallel texture scratches and the die-cutting sheet with strong deviation tendency; respectively calculates the similarity of the scratch feature vector and a plurality of historical feature vectors to determine whether the die-cuting sheet is a first-grade die-cutting sheet or a second-grade die-cuting sheet; calculates the simulated scratch density coefficient to correct the historical feature vector and sorts the next batch of die-cuting sheets. The present application improves the accuracy of the die-cutting sheet continuous line AOI detection method based on AI vision sorting.

Description

Technical Field

[0001] This invention relates to the field of image data processing technology, and in particular to an AOI detection method for connected wafer materials based on AI visual sorting. Background Technology

[0002] In practical applications, the surface of die-cut materials often has regular textures. Scratches along the texture direction are easily confused with the texture itself, leading to misjudgment or missed detection. Factors such as changes in lighting and equipment vibration can cause image acquisition distortion, resulting in inaccurate pixel comparison and affecting the reliability of detection results. Therefore, there is an urgent need for an AI vision and AOI detection method that can accurately identify scratches along the texture, effectively deal with detection distortion, and achieve dynamic and accurate grading, so as to improve the detection level and product consistency of die-cut material production lines.

[0003] Chinese Patent Application Publication No. CN116542976A discloses an invention relating to the field of image data processing technology, specifically a visual detection system for defects in die-cut parts. The system includes: an image acquisition and processing module that performs binarization to obtain a suspected indentation region and a background region; an indentation probability calculation module that calculates the approximation of the average gradient direction of the suspected indentation region boundary based on the gradient direction of pixels on the suspected indentation region boundary, thereby obtaining the anomaly probability of the suspected indentation region; a pixel enhancement necessity module that determines the enhancement necessity of pixels in the suspected indentation region based on the anomaly probability and grayscale value fluctuations of the suspected indentation region; and an image enhancement module that enhances the grayscale image of the die-cut part according to enhancement weights. This invention utilizes the characteristics of grayscale changes and continuity near the indentation, enhancing different pixel regions of the image to varying degrees, making the indentation more clearly visible in the image, which is beneficial for the identification and segmentation of the indentation.

[0004] The existing technology also has the following problems: the existing technology only improves the detection accuracy through image enhancement, without taking into account the distortion of the detection equipment, so as to improve the detection accuracy by selectively sorting the sample material for re-inspection. Summary of the Invention

[0005] To address this issue, the present invention provides an AI-based visual sorting method for AOI detection of die-cutting materials, which overcomes the problem in existing technologies that only improve detection accuracy through image enhancement without considering the distortion of the detection equipment, so as to improve detection accuracy by selectively sorting die-cutting materials for re-inspection.

[0006] To achieve the above objectives, the present invention provides an AOI detection method for die-cutting material interconnection based on AI visual sorting, comprising:

[0007] Based on the texture image of the collected die-cut material surface, the texture spacing and texture gray level are obtained to generate texture characterization coefficients, so as to preliminarily determine whether there are scratches along the texture of the die-cut material;

[0008] AI vision inspection is performed on the die-cutting material that is initially determined to have scratches along the grain. The texture image of the die-cutting material that is initially determined not to have scratches along the grain is compared with the preset standard texture image to determine whether there is a tendency for distortion in AOI detection.

[0009] In response to the distortion tendency of AOI detection, a texture deviation coefficient is calculated based on the texture characterization coefficient and texture characterization threshold of each die slice material that is initially determined to have no scratches along the texture to determine the deviation trend of each die slice material, and the pixel difference coefficient of the die slice material with weak deviation trend is obtained.

[0010] Based on the pixel difference coefficient, determine whether to change the deviation trend of the die-cut material with a weak deviation trend to a strong deviation trend, and collect scratch images of the die-cut material with preliminarily determined to have scratches along the grain and the die-cut material with a strong deviation trend.

[0011] Scratch edge features and scratch sharpness features are extracted from the scratch image to calculate scratch feature vectors. The similarity between the scratch feature vectors and several historical feature vectors is calculated to determine whether the die-cutting material is a first-grade die-cutting material or a second-grade die-cutting material.

[0012] Based on the sorting ratio of the second-grade die-cutting material within a preset period, it is determined whether to energize the second-grade die-cutting material. The simulated scratch density coefficient is calculated by combining the current and voltage of the second-grade die-cutting material with the scratch image to correct the historical feature vector and sort the next batch of die-cutting material.

[0013] Furthermore, the process of obtaining texture spacing and texture grayscale to generate texture representation coefficients includes,

[0014] The texture image is divided into several acquisition regions, and the texture spacing and texture grayscale of each acquisition region are extracted.

[0015] The standard deviation of several texture spacings is determined as a spacing factor, and the standard deviation of several texture grayscales is determined as a grayscale factor;

[0016] The weighted sum of the spacing factor and the grayscale factor is determined to be the texture characterization coefficient.

[0017] Furthermore, based on the acquired texture image of the die-cut material surface, texture spacing and texture grayscale are obtained to generate texture characterization coefficients, in order to preliminarily determine whether there are scratches along the grain of the die-cut material.

[0018] If the texture characterization coefficient is less than or equal to the texture characterization threshold, it is determined that the die-cut material does not have scratches along the grain.

[0019] If the texture characterization coefficient is greater than the texture characterization threshold, it is determined that the die-cut material has scratches along the grain.

[0020] Furthermore, the process of comparing the pixel features of the texture image of the preliminarily determined non-scratched material with a preset standard texture image to determine whether AOI detection has a distortion tendency includes,

[0021] Alignment processing is performed on the texture image and the preset standard texture image respectively. The texture image is converted into a grayscale image to determine the grayscale value of each pixel as the texture grayscale. The specific values ​​of L, a, and b of each pixel in the texture image are obtained. The grayscale image is converted into a black and white binary image, and pixels with abrupt grayscale changes are selected to generate continuous and closed contour regions.

[0022] The average value of the texture grayscale in the texture image is determined to be the brightness feature factor;

[0023] The average value of the Lab color in the texture image is determined as the color feature factor;

[0024] The area of ​​the contour region in the texture image is determined as the region feature factor;

[0025] Based on the comparison results of the brightness feature factor, the color feature factor, and the region feature factor with the brightness standard feature factor, color standard feature factor, and region standard feature factor of the preset standard texture image, it is determined whether there is a tendency for AOI detection to be distorted.

[0026] The pixel features include brightness features, color features, and display area features. The preset standard texture image is a texture image of a die-cutting material with the same texture characterization coefficient as the die-cutting material in historical detection data, provided that there is no distortion tendency in AOI detection.

[0027] Furthermore, in response to the existence of any feature factor less than the corresponding standard feature factor, it is determined that AOI detection has a tendency to be distorted, wherein,

[0028] The feature factors include the luminance feature factor, the color feature factor, and the region feature factor, and the standard feature factors include the luminance standard feature factor, the color standard feature factor, and the region standard feature factor.

[0029] Furthermore, the process of calculating the texture deviation coefficient based on the texture characterization coefficient and texture characterization threshold of each die-cut material that is initially determined to have no along-grain scratches to determine the deviation trend of each die-cut material includes,

[0030] The difference between the texture representation threshold and the texture representation coefficient is determined as the texture deviation coefficient;

[0031] In response to the texture deviation coefficient being greater than or equal to the texture deviation threshold, the deviation trend of the die-cut material is determined to be a weak deviation trend;

[0032] In response to the texture deviation coefficient being less than the texture deviation threshold, the deviation trend of the die-cut material is determined to be a strong deviation trend.

[0033] Furthermore, the process of determining whether to change the deviation trend of the die-cut material with a weak deviation trend to a strong deviation trend based on the pixel difference coefficient includes,

[0034] Given that at least one feature factor is less than its corresponding standard feature factor, the ratio of the difference between each feature factor and its corresponding standard feature factor is determined as the pixel difference coefficient.

[0035] If all the pixel difference coefficients are greater than or equal to the pixel difference threshold, it is determined that the deviation trend of the die-cut material will be changed to a strong deviation trend.

[0036] Furthermore, the process of calculating the similarity between the scratch feature vector and several historical feature vectors to determine whether the die-cutting material is a first-grade die-cutting material or a second-grade die-cutting material includes,

[0037] The scratch edge features and scratch sharpness features are input into the learning model for feature encoding to generate scratch feature vectors;

[0038] The first similarity is calculated as the similarity between the scratch feature vector and the historical feature vector of the first-grade die-cut material;

[0039] The similarity between the scratch feature vector and the historical feature vector of the second-grade die-cut material is calculated as the second similarity.

[0040] If the first similarity is greater than or equal to the second similarity, the die-cutting material is determined to be first-grade die-cutting material.

[0041] If the first similarity is less than the second similarity, the die-cutting material is determined to be second-grade die-cutting material.

[0042] The scratch edge features include edge continuity and edge smoothness, and the scratch sharpness features include tip sharpness and tip convergence. The scratch severity of the first-grade die-cutting material is lower than that of the second-grade die-cutting material.

[0043] Furthermore, the process of determining whether to energize the second-grade die-cutting material based on the sorting ratio of the second-grade die-cutting material within a preset period includes,

[0044] The sorting percentage is determined as the ratio of the number of second-grade die-cut materials identified within a preset period to the total number of sorting times of the die-cut materials.

[0045] If the sorting ratio is greater than the sorting ratio threshold, it is determined that the second-level die-cut material is energized.

[0046] The total number of sorting times is the total number of first-grade die-cut materials and second-grade die-cut materials.

[0047] Furthermore, the process of calculating the simulated scratch density coefficient by combining the current and voltage of the second-level die-cutting material with the scratch image to correct the historical feature vector includes,

[0048] The mean of the ratios of the maximum scratch length to the texture length of the die-cut material in each scratch image is determined as the scratch length factor;

[0049] The ratio of the scratch area to the total area of ​​each scratch image is determined as the scratch surface factor;

[0050] The reciprocal of the ratio of the current to the voltage is determined as the conductivity factor;

[0051] The product of the scratch length factor, the scratch surface factor, and the conductivity factor is determined to be the simulated scratch density coefficient;

[0052] The variance of several simulated scratch density coefficients is determined as correction coefficients, and the historical feature vector is corrected based on the correction coefficients.

[0053] Compared with the prior art, the beneficial effects of the present invention are as follows: by combining the stability of texture spacing and the consistency of texture grayscale, the present invention can effectively distinguish between the natural texture and the scratch defects along the texture of the die-cutting material. When the local texture spacing or grayscale changes abruptly, the texture characterization coefficient will increase, thereby accurately determining the existence of scratches. By comparing the texture image without scratches along the texture with the preset standard texture image in multiple dimensions, the distortion tendency of AOI detection can be effectively identified. For die-cutting materials that are initially determined to have no scratches along the texture but have a distortion tendency, the pixel difference coefficient is calculated to reflect the degree of distortion of AOI detection. By quantifying the weak deviation trend and combining it with the pixel difference coefficient, it is determined whether to correct it to a strong deviation trend, thereby achieving accurate identification of critical defects, and further improving the accuracy of the AOI detection method for die-cutting material connection based on AI vision sorting.

[0054] Furthermore, this invention calculates a texture deviation coefficient when AOI detection is deemed to have a distortion tendency, reflecting the degree to which the die-cut material deviates from the normal texture under the current distortion environment, thereby determining its deviation trend. Since overall image distortion can mask minor scratches, it clearly distinguishes between weak and strong deviation trends. For die-cut materials with a slight deviation, a pixel difference coefficient is calculated; for die-cut materials with a severe deviation, AI visual detection is performed directly. This hierarchical processing strategy achieves precise allocation of computing resources, effectively improving the system's online detection speed while ensuring detection accuracy. If the pixel difference coefficient of a die-cut material with a weak deviation trend is small, it indicates that the AOI detection distortion is small and the texture consistency of the die-cut material is good, and the system will maintain its judgment of a weak deviation trend. If the pixel difference coefficient is large, it indicates that the AOI detection distortion is large and the texture of the die-cut material is abnormal, and the system will reclassify it as a strong deviation trend, correcting the interfered data and effectively reducing the false negative rate under complex working conditions, thereby further improving the accuracy of the die-cut material connection AOI detection method based on AI visual sorting.

[0055] Furthermore, this invention extracts the continuity and smoothness of the scratch edges, as well as the sharpness and convergence of the scratch tips from scratch images, and fuses them into a scratch feature vector to reflect the severity of the scratches. Sharp, discontinuous scratches are potential sources of crack initiation and are easily propagated under stress or environmental erosion. The grade is determined by calculating the similarity between the current scratch feature vector and several historical feature vectors. This allows for the direct screening of die-cut materials with good light transmittance, haze, and visual uniformity. The grading standard is based on the statistical accumulation of a large amount of historical data. Die-cut materials with strong deviation trends and die-cut materials with directly detected grain-side scratches are collected side-by-side and graded. Dividing the die-cut materials into first and second grades enables the production line to achieve refined sorting of die-cut materials, thereby further improving the accuracy of the AI ​​vision sorting-based die-cut material AOI detection method.

[0056] Furthermore, this invention combines visual inspection of die-cut materials with electrical performance. When the proportion of the second-grade materials reaches a certain threshold, an electrical test is triggered on the second-grade die-cut materials. The physical electrical performance is used to verify the visual grading results. If the electrical test results and grading results are inconsistent, it indicates that the sorting proportion is unreasonable and the grading of the die-cut materials is too strict, requiring correction of the historical feature vector. If the electrical test results and grading results are consistent, it indicates that the processing technology of the die-cut materials is abnormal, requiring an alarm signal to be issued to inspect the processing workshop. The historical feature vector is corrected based on the simulated scratch density coefficient, avoiding misjudgment of grading caused by the use of inaccurate historical feature vectors. At the same time, it realizes the automatic optimization of the inspection standards for the next batch of die-cut materials, thereby further improving the accuracy of the AI ​​visual sorting-based die-cut material AOI inspection method. Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating the steps of the AI-based visual sorting method for detecting the connection lines of cut materials according to an embodiment of the present invention.

[0058] Figure 2 This is a logic diagram for determining whether there are grain-side scratches on the die-cut material in an embodiment of the present invention.

[0059] Figure 3 A logic diagram for determining the deviation trend of each die-cutting material in an embodiment of the present invention;

[0060] Figure 4 This is a logic diagram for determining whether to energize the die-cut material in an embodiment of the present invention. Detailed Implementation

[0061] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0062] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0063] It should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0064] Please see Figure 1 The diagram shows a flowchart of the AOI detection method for die-cutting material connections based on AI visual sorting according to an embodiment of the present invention. The AOI detection method for die-cutting material connections based on AI visual sorting according to an embodiment of the present invention includes:

[0065] Step S1: Based on the collected texture image of the die-cut material surface, obtain the texture spacing and texture grayscale to generate texture characterization coefficients, so as to preliminarily determine whether there are scratches along the texture of the die-cut material.

[0066] Step S2: Perform AI visual inspection on the die-cutting material that is initially determined to have scratches along the grain, and compare the pixel features of the texture image of the die-cutting material that is initially determined not to have scratches along the grain with the preset standard texture image to determine whether there is a tendency for distortion in AOI detection.

[0067] Step S3: In response to the distortion tendency of AOI detection, the texture deviation coefficient is calculated based on the texture characterization coefficient and texture characterization threshold of each die slice material that is initially determined to have no scratches along the texture to determine the deviation trend of each die slice material, and the pixel difference coefficient of the die slice material with weak deviation trend is obtained.

[0068] Step S4: Determine whether to change the deviation trend of the die-cut material with a weak deviation trend to a strong deviation trend based on the pixel difference coefficient, and collect scratch images of the die-cut material with preliminarily determined to have scratches along the grain and the die-cut material with a strong deviation trend.

[0069] Step S5: Extract scratch edge features and scratch sharpness features from the scratch image to calculate scratch feature vectors. Calculate the similarity between the scratch feature vectors and several historical feature vectors to determine whether the die-cutting material is a first-grade die-cutting material or a second-grade die-cutting material.

[0070] Step S6: Based on the sorting ratio of the second-grade die-cutting material within a preset period, determine whether to energize the second-grade die-cutting material. Combine the current, voltage and scratch image of the second-grade die-cutting material to calculate the simulated scratch density coefficient to correct the historical feature vector and sort the next batch of die-cutting materials.

[0071] It is understood that this invention is applied to die-cut materials made of brushed metal, such as brushed stainless steel sheets and brushed aluminum panels. Since such die-cut materials have brushed textures, the brushed textures and the morphology of scratches along the grain are highly similar. Considering that brushed textures and scratches along the grain are difficult to distinguish during inspection, this invention uses AOI and AI vision to perform multiple inspections on the die-cut materials to improve the inspection accuracy of the die-cut materials.

[0072] Specifically, the process of obtaining texture spacing and texture grayscale to generate texture representation coefficients includes,

[0073] The texture image is divided into several acquisition regions, and the texture spacing and texture grayscale of each acquisition region are extracted.

[0074] Determine the standard deviation of several texture spacings as the spacing factor, and determine the standard deviation of several texture grayscales as the grayscale factor;

[0075] The weighted sum of the spacing factor and the gray factor is determined to be the texture characterization coefficient.

[0076] Specifically, when calculating the standard deviation of texture spacing and texture grayscale, it is necessary to first normalize the texture spacing and texture grayscale to convert them to a uniform range, for example, to the interval [0, 1], to eliminate the influence of numerical differences. This will not be elaborated further here.

[0077] Specifically, the sum of the weighting coefficients of the spacing factor and the grayscale factor is 1. Since the scratch defects along the grain mainly manifest as the destruction of the periodic spacing structure of the brushed texture and have a significant impact on the spacing distribution, the weighting coefficient of the spacing factor is 0.7 and the weighting coefficient of the grayscale factor is 0.3.

[0078] Specifically, for each acquisition area, the average pixel distance between adjacent extreme points is obtained by detecting the grayscale profile extreme points perpendicular to the main direction of the brushed texture, and is used as the texture spacing. The average grayscale value of the bright and dark texture areas is extracted as the texture grayscale.

[0079] Please see Figure 2 As shown, this is a logic diagram for determining whether there are grain-side scratches on the die-cutting material according to an embodiment of the present invention. Based on the collected texture image of the die-cutting material surface, texture spacing and texture grayscale are obtained to generate texture characterization coefficients, in order to initially determine whether there are grain-side scratches on the die-cutting material.

[0080] If the texture characterization coefficient is less than or equal to the texture characterization threshold, it is determined that the template material does not have scratches along the grain.

[0081] If the texture characterization coefficient is greater than the texture characterization threshold, it is determined that the die-cut material has scratches along the grain.

[0082] Specifically, the preset texture characterization threshold is the product of the texture characterization coefficient reference value and the texture characterization factor. The texture characterization coefficient reference value is the average value of the texture characterization coefficient detected under the same working conditions in historical data. The texture characterization factor can be set by those skilled in the art according to the accuracy requirements of the test material detection. The higher the accuracy requirement, the smaller the value should be. Optionally, the value range can be [0.05, 0.15]. Preferably, it can be 0.1.

[0083] Understandably, determining whether AOI inspection has a tendency to be distorted is a matter of whether the accuracy of AOI inspection meets the standards. That is, under the condition that AOI inspection has a tendency to be distorted, the inspection of die-cut materials with excessively fine scratches along the grain may miss some cases. However, if the AOI inspection equipment can still detect the scratches along the grain of the die-cut material even when the accuracy is low, it means that the scratches along the grain are relatively large. Therefore, the determination that the die-cut material has scratches along the grain is reliable.

[0084] Specifically, the process of comparing the pixel features of the texture image of the die-cut material that is initially determined to have no scratches along the grain with a preset standard texture image to determine whether AOI detection has a tendency to be distorted includes,

[0085] Alignment processing is performed on the texture image and the preset standard texture image respectively. The texture image is converted into a grayscale image to determine the grayscale value of each pixel as the texture grayscale. The specific values ​​of L, a, and b of each pixel in the texture image are obtained. The grayscale image is converted into a black and white binary image, and pixels with abrupt grayscale changes are selected to generate continuous and closed contour regions.

[0086] The average value of the texture grayscale in the texture image is determined as the brightness feature factor;

[0087] The average value of Lab color in the texture image is determined as the color feature factor;

[0088] The area of ​​the contour region in the texture image is determined as the region feature factor;

[0089] The AOI detection is judged based on the comparison results of the brightness feature factor, color feature factor and region feature factor with the brightness standard feature factor, color standard feature factor and region standard feature factor of the preset standard texture image, respectively.

[0090] The pixel features include brightness features, color features, and display area features. The preset standard texture image is the texture image of the template material with the same texture characterization coefficient as the template material in the historical detection data under the condition that there is no distortion tendency in AOI detection.

[0091] Specifically, when aligning a texture image with a preset standard texture image, feature points are first extracted from the standard texture image, such as key corner points of die-cut edges or brushed textures. Then, the SIFT algorithm is used to match the same feature points in the actual image. The actual texture image is corrected to a coordinate system that completely overlaps with the standard texture image through perspective transformation. Finally, the brightness of the two images is normalized with the same parameters.

[0092] Specifically, Lab is a uniform color space specifically designed to accurately describe color and color difference. In practice, the arithmetic mean of each pixel in the L, a, and b channels is first calculated, and then the mean of the three arithmetic means is determined to be the average value of the Lab color.

[0093] Specifically, in response to all feature factors being greater than or equal to the corresponding standard feature factors, it is determined that AOI detection does not have a tendency to be distorted.

[0094] In response to the existence of any feature factor less than the corresponding standard feature factor, it is determined that AOI detection has a tendency to be distorted.

[0095] The characteristic factors include luminance characteristic factors, color characteristic factors, and region characteristic factors, while the standard characteristic factors include luminance standard characteristic factors, color standard characteristic factors, and region standard characteristic factors.

[0096] In a specific embodiment, the standard feature factor of brightness in the preset standard texture image is 185, the standard feature factor of color is 35.7, and the standard feature factor of region is 500 pixels. The brightness feature factor of the texture image is detected to be 180, and the L=85, a=11, and b=6 of Lab color are calculated to be 34 and the region feature factor is 480 pixels. Since the feature factors are all less than the corresponding standard feature factors, it is determined that the AOI detection has a tendency to be distorted.

[0097] Specifically, this invention effectively distinguishes between natural textures and grain-following scratches in die-cutting materials by combining the stability of texture spacing and the consistency of texture grayscale. When a sudden change occurs in local texture spacing or grayscale, the texture characterization coefficient increases, thus accurately determining the presence of scratches. By comparing the texture image without grain-following scratches with a preset standard texture image using multi-dimensional pixel features, the AOI detection distortion tendency can be effectively identified. For die-cutting materials initially determined to have no grain-following scratches but exhibiting distortion tendency, the pixel difference coefficient is calculated to reflect the degree of AOI detection distortion. By quantifying weak deviation trends and combining them with the pixel difference coefficient, it is determined whether to correct them to strong deviation trends, thus achieving accurate identification of critical defects and further improving the accuracy of the AI ​​vision sorting-based die-cutting material connection AOI detection method.

[0098] Please see Figure 3 As shown, this is a logic diagram for determining the deviation trend of each die-cut material in an embodiment of the present invention. The process of calculating the texture deviation coefficient based on the texture characterization coefficient and texture characterization threshold of each die-cut material that is initially determined to have no along-grain scratches to determine the deviation trend of each die-cut material includes:

[0099] The difference between the texture representation threshold and the texture representation coefficient is determined as the texture deviation coefficient;

[0100] In response to a texture deviation coefficient greater than or equal to a texture deviation threshold, the deviation trend of the die-cut material is determined to be a weak deviation trend.

[0101] When the texture deviation coefficient is less than the texture deviation threshold, the deviation trend of the die-cut material is determined to be a strong deviation trend.

[0102] Specifically, the purpose of setting the texture deviation threshold is to characterize the consistency between the surface texture of the die-cut material and the surface texture of the die-cut material without scratches along the grain under the same working conditions. Optionally, the texture deviation threshold can be in the range of [0.1, 0.3], and preferably, it can be 0.2.

[0103] Specifically, the process of determining whether to change the deviation trend of a weak deviation trend in a die-cut material to a strong deviation trend by combining the pixel difference coefficient includes,

[0104] Given that at least one feature factor is less than its corresponding standard feature factor, the ratio of the difference between each feature factor and its corresponding standard feature factor is determined as the pixel difference coefficient.

[0105] If all pixel difference coefficients are greater than or equal to the pixel difference threshold, then the deviation trend of the die-cut material will be changed to a strong deviation trend.

[0106] If any pixel difference coefficient is less than the pixel difference threshold, it is determined that the deviation trend of the die-cut material will not be changed to a strong deviation trend.

[0107] Specifically, since the magnitude difference between the feature factors is too large, when calculating the pixel difference coefficient, it is necessary to first normalize each feature factor and its corresponding standard feature factor, and transform each feature factor and its corresponding standard feature factor into a unified range, for example, into the interval [0, 1], to eliminate the influence of numerical differences. This will not be elaborated further here.

[0108] Specifically, the difference ratio is the ratio of the difference between the standard characteristic factor and the characteristic factor to the standard characteristic factor.

[0109] Specifically, the purpose of setting the pixel difference threshold is to characterize the deviation between the degree of distortion of AOI detection and the situation where AOI detection has no tendency to be distorted. Optionally, the value range of the pixel difference threshold can be [0.15, 0.25], and preferably, it can be 0.2.

[0110] Specifically, this invention calculates a texture deviation coefficient when AOI detection is deemed to have a distortion tendency, reflecting the degree to which the die-cut material deviates from the normal texture under the current distortion environment, thereby determining its deviation trend. Since overall image distortion can mask minor scratches, it clearly distinguishes between weak and strong deviation trends. For die-cut materials with a slight deviation, a pixel difference coefficient is calculated; for die-cut materials with a severe deviation, AI visual detection is performed directly. This hierarchical processing strategy achieves precise allocation of computing resources, effectively improving the system's online detection speed while ensuring detection accuracy. If the pixel difference coefficient of a die-cut material with a weak deviation trend is small, it indicates that the AOI detection distortion is small and the texture consistency of the die-cut material is good, and the system will maintain its judgment of a weak deviation trend. If the pixel difference coefficient is large, it indicates that the AOI detection distortion is large and the texture of the die-cut material is abnormal, and the system will reclassify it as a strong deviation trend, correcting the interfered data and effectively reducing the false negative rate under complex working conditions, thereby further improving the accuracy of the die-cut material connection AOI detection method based on AI visual sorting.

[0111] Specifically, the process of calculating the similarity between the scratch feature vector and several historical feature vectors to determine whether the die-cutting material is a first-grade die-cutting material or a second-grade die-cutting material includes,

[0112] The scratch edge features and scratch sharpness features are input into the learning model for feature encoding to generate scratch feature vectors;

[0113] The first similarity is calculated by comparing the scratch feature vector with the historical feature vector of the first-grade die-cut material.

[0114] The similarity between the scratch feature vector and the historical feature vector of the second-grade die-cut material is calculated as the second similarity.

[0115] If the first similarity is greater than or equal to the second similarity, the die-cutting material is determined to be first-grade die-cutting material.

[0116] If the first similarity is less than the second similarity, the die-cutting material is determined to be second-grade die-cutting material.

[0117] Among them, the scratch edge characteristics include edge continuity and edge smoothness, and the scratch sharpness characteristics include tip sharpness and tip convergence. The scratch degree of the first-grade die-cutting material is lower than that of the second-grade die-cutting material.

[0118] Specifically, calculating the scratch feature vector With several historical feature vectors similarity ;in, ;

[0119] Similarity :

[0120] .

[0121] Specifically, the first-grade die-cutting material has a uniform brushed texture and no excessively deep or extremely irregular scratches along the grain. It can be used for high-end appearance parts, such as laptop shells and high-end home appliance panels, or sold directly to the market. The second-grade die-cutting material has slight scratches along the brushed direction, but these do not penetrate the brushed texture layer and are not noticeable under non-focused light conditions. It can be used for internal support and connecting parts of products where appearance is not a requirement.

[0122] Specifically, edge continuity refers to the continuity of pixels at the edge of the scratch, edge smoothness refers to the uniformity of curvature at the edge of the scratch, tip sharpness refers to the angle of the scratch tip region, and tip convergence refers to the compactness of the scratch tip converging towards a point.

[0123] Specifically, in implementation, when obtaining edge continuity, first extract the sequence of contour points of the scratch edge, then calculate the edge standard deviation of the Euclidean distance between adjacent points, and determine that the edge continuity = 1 / (1 + edge standard deviation). When obtaining edge smoothness, first calculate the curvature by performing the second derivative on the edge contour, then calculate the mean curvature of all curvatures, and determine that the edge smoothness = 1 / (1 + mean curvature). When obtaining tip sharpness, first locate the three core points of the scratch tip, then calculate the included angle θ formed by the three points, and determine that the tip sharpness = 1 - θ / 18. 0. When obtaining the tip convergence, first locate the maximum curvature point of the scratch tip vertex, take ±5 contour points around the vertex to form the tip region point set S, establish a local coordinate system with the maximum curvature point as the origin, calculate the angle between each point in S and the X-axis, calculate the angle standard deviation of the angle sequence based on the angle, and determine the tip convergence as = 1 / (1 + angle standard deviation). The normalized edge continuity, edge smoothness, tip sharpness and tip convergence are concatenated into a one-dimensional vector and input into the MLP model for feature encoding to output the scratch feature vector.

[0124] In a specific embodiment, the first-level historical feature vector is set to [0.90, 0.85, 0.70, 0.80], and the second-level historical feature vector is set to [0.60, 0.55, 0.40, 0.45]. The calculated edge continuity is 0.92, edge smoothness is 0.88, tip sharpness is 0.75, and tip convergence is 0.81. Therefore, the scratch feature vector is [0.92, 0.88, 0.75, 0.81]. The first similarity is 0.9996, and the second similarity is 0.997. In response to the first similarity being greater than the second similarity, the die-cutting material is determined to be the first-level die-cutting material.

[0125] Specifically, this invention extracts the continuity and smoothness of the scratch edges, as well as the sharpness and convergence of the scratch tips from scratch images, and fuses them into a scratch feature vector to reflect the severity of the scratch. Sharp, discontinuous scratches are potential sources of crack initiation and are prone to propagation under stress or environmental erosion. The grade is determined by calculating the similarity between the current scratch feature vector and several historical feature vectors. This allows for the direct screening of die-cut materials with good light transmittance, haze, and visual uniformity. The grading standard is based on the statistical accumulation of a large amount of historical data. Die-cut materials with strong deviation trends and die-cut materials with directly detected longitudinal scratches are collected side by side for scratch image acquisition and grading. Dividing the die-cut materials into first and second grades enables the production line to achieve fine sorting of die-cut materials, thereby further improving the accuracy of the AI ​​vision sorting-based die-cut material AOI detection method.

[0126] Please see Figure 4As shown, this is a logic diagram for determining whether to energize the die-cutting material according to an embodiment of the present invention. The process of determining whether to energize the second-grade die-cutting material based on the sorting ratio of the second-grade die-cutting material within a preset period includes:

[0127] The sorting percentage is determined as the ratio of the number of second-grade die-cut materials identified within a preset period to the total number of sorting times of the die-cut materials.

[0128] If the sorting ratio is greater than the sorting ratio threshold, it is determined that the second-grade die-cut material is energized.

[0129] If the sorting ratio is less than or equal to the sorting ratio threshold, it is determined that the second-level die-cut material will not be energized.

[0130] The total number of sorting operations is the total number of first-grade die-cut materials and second-grade die-cut materials.

[0131] Specifically, the purpose of setting the sorting ratio threshold is to characterize the consistency between the sorting of the second-grade die-cutting material and the first-grade die-cutting material in each detection cycle. The value range of the sorting ratio threshold can be [0.35, 0.45], preferably 0.4.

[0132] Specifically, the preset cycle can be set to 2 days. In practice, the preset cycle can be adjusted according to the specific batches of test materials each day. If the number of test batches increases, the preset cycle can be extended.

[0133] Specifically, the process of calculating the simulated scratch density coefficient to correct historical eigenvectors by combining the current, voltage, and scratch images of the second-grade die-cut material includes:

[0134] The mean of the ratio of the maximum scratch length to the texture length of the die-cut material in each scratch image is determined as the scratch length factor;

[0135] The ratio of the scratch area to the total area of ​​each scratch image is determined as the scratch surface factor.

[0136] The reciprocal of the ratio of current to voltage is determined as the conductivity factor;

[0137] The product of the scratch length factor, scratch surface factor, and electrical conductivity factor is determined to be the simulated scratch density coefficient;

[0138] The variance of several simulated scratch density coefficients is determined as correction coefficients, and the historical feature vectors are corrected based on the correction coefficients.

[0139] In one specific embodiment, if the correction coefficient is less than 0.03, the historical feature vector of the first-grade die-cutting material is reduced by 5% and the historical feature vector of the second-grade die-cutting material is increased by 10%. If the correction coefficient is in the range [0.03, 0.07], the historical feature vector of the first-grade die-cutting material is reduced by 10% and the historical feature vector of the second-grade die-cutting material is increased by 15%. If the correction coefficient is greater than 0.07, the historical feature vector of the first-grade die-cutting material is reduced by 15% and the historical feature vector of the second-grade die-cutting material is increased by 20%.

[0140] Specifically, this invention combines visual inspection of die-cut materials with electrical performance. When the proportion of second-grade materials reaches a certain threshold, an electrical test is triggered on the second-grade die-cut materials. The physical electrical performance is used to verify the visual grading results. If the electrical test results and grading results are inconsistent, it indicates that the sorting proportion is unreasonable and the grading of the die-cut materials is too strict, requiring correction of the historical feature vector. If the electrical test results and grading results are consistent, it indicates that the processing technology of the die-cut materials has become abnormal, requiring an alarm signal to be issued to inspect the processing workshop. The historical feature vector is corrected based on the simulated scratch density coefficient, avoiding misjudgment of grading caused by the use of inaccurate historical feature vectors. At the same time, it realizes the automatic optimization of the inspection standards for the next batch of die-cut materials, thereby further improving the accuracy of the AI ​​visual sorting-based die-cut material AOI inspection method.

[0141] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. An AOI detection method for connected strip materials based on AI visual sorting, characterized in that, include: Based on the texture image of the collected die-cut material surface, the texture spacing and texture gray level are obtained to generate texture characterization coefficients, so as to preliminarily determine whether there are scratches along the texture of the die-cut material; AI visual inspection is performed on the die-cutting materials that are initially determined to have scratches along the grain. The texture images of the die-cutting materials that are initially determined not to have scratches along the grain are compared with the preset standard texture images to determine whether there is a tendency for distortion in AOI detection. In response to the distortion tendency of AOI detection, a texture deviation coefficient is calculated based on the texture characterization coefficient and texture characterization threshold of each die slice material that is initially determined to have no scratches along the texture, so as to determine the deviation trend of each die slice material and obtain the pixel difference coefficient of the die slice material with a weak deviation trend. Based on the pixel difference coefficient, determine whether to change the deviation trend of the die-cut material with a weak deviation trend to a strong deviation trend, and collect scratch images of the die-cut material with preliminarily determined to have scratches along the grain and the die-cut material with a strong deviation trend. Scratch edge features and scratch sharpness features are extracted from the scratch image to calculate scratch feature vectors. The similarity between the scratch feature vectors and several historical feature vectors is calculated to determine whether the die-cutting material is a first-grade die-cutting material or a second-grade die-cutting material. Based on the sorting ratio of the second-grade die-cutting material within a preset period, it is determined whether to energize the second-grade die-cutting material. The simulated scratch density coefficient is calculated by combining the current and voltage of the second-grade die-cutting material with the scratch image to correct the historical feature vector and sort the next batch of die-cutting material. The process of obtaining texture spacing and texture grayscale to generate texture representation coefficients includes: The texture image is divided into several acquisition regions, and the texture spacing and texture grayscale of each acquisition region are extracted. The standard deviation of several texture spacings is determined as a spacing factor, and the standard deviation of several texture grayscales is determined as a grayscale factor; The weighted sum of the spacing factor and the grayscale factor is determined to be the texture characterization coefficient; The process of determining whether to change the deviation trend of the weak deviation trend of the die-cut material to a strong deviation trend based on the pixel difference coefficient includes, Given that at least one feature factor is less than its corresponding standard feature factor, the ratio of the difference between each feature factor and its corresponding standard feature factor is determined as the pixel difference coefficient. If all the pixel difference coefficients are greater than or equal to the pixel difference threshold, it is determined that the deviation trend of the die-cut material will be changed to a strong deviation trend. The process of calculating the simulated scratch density coefficient by combining the current and voltage of the second-level die-cutting material with the scratch image to correct the historical feature vector includes: The mean of the ratios of the maximum scratch length to the texture length of the die-cut material in each scratch image is determined as the scratch length factor; The ratio of the scratch area to the total area of ​​each scratch image is determined as the scratch surface factor; The reciprocal of the ratio of the current to the voltage is determined as the conductivity factor; The product of the scratch length factor, the scratch surface factor, and the conductivity factor is determined to be the simulated scratch density coefficient; The variance of several simulated scratch density coefficients is determined as correction coefficients, and the historical feature vector is corrected based on the correction coefficients.

2. The AOI detection method for die-cutting material interconnection based on AI visual sorting according to claim 1, characterized in that, The textural spacing and grayscale values ​​are obtained from the acquired texture image of the die-cut material surface to generate texture characterization coefficients, in order to preliminarily determine whether there are scratches along the grain of the die-cut material. If the texture characterization coefficient is less than or equal to the texture characterization threshold, it is determined that the die-cut material does not have scratches along the grain. If the texture characterization coefficient is greater than the texture characterization threshold, it is determined that the die-cut material has scratches along the grain.

3. The AOI detection method for die-cutting material interconnection based on AI visual sorting according to claim 2, characterized in that, The process of comparing the pixel features of the texture image of the preliminarily determined non-scratched material with a preset standard texture image to determine whether AOI detection has a distortion tendency includes: Alignment processing is performed on the texture image and the preset standard texture image respectively. The texture image is converted into a grayscale image to determine the grayscale value of each pixel as the texture grayscale. The specific values ​​of L, a, and b of each pixel in the texture image are obtained. The grayscale image is converted into a black and white binary image, and pixels with abrupt grayscale changes are selected to generate continuous and closed contour regions. The average value of the texture grayscale in the texture image is determined to be the brightness feature factor; The average value of the Lab color in the texture image is determined as the color feature factor; The area of ​​the contour region in the texture image is determined as the region feature factor; Based on the comparison results of the brightness feature factor, the color feature factor, and the region feature factor with the brightness standard feature factor, color standard feature factor, and region standard feature factor of the preset standard texture image, it is determined whether there is a tendency for AOI detection to be distorted. The pixel features include brightness features, color features, and display area features. The preset standard texture image is a texture image of a die-cutting material with the same texture characterization coefficient as the die-cutting material in historical detection data, provided that there is no distortion tendency in AOI detection.

4. The AOI detection method for inter-line cutting materials based on AI visual sorting according to claim 3, characterized in that, In response to the existence of any feature factor less than the corresponding standard feature factor, it is determined that AOI detection has a tendency to be distorted. The feature factors include the luminance feature factor, the color feature factor, and the region feature factor, and the standard feature factors include the luminance standard feature factor, the color standard feature factor, and the region standard feature factor.

5. The AOI detection method for die-cutting material interconnection based on AI visual sorting according to claim 4, characterized in that, The process of calculating the texture deviation coefficient based on the texture characterization coefficient and texture characterization threshold of each die-cut material that is initially determined to have no scratches along the grain, to determine the deviation trend of each die-cut material includes, The difference between the texture representation threshold and the texture representation coefficient is determined as the texture deviation coefficient; In response to the texture deviation coefficient being greater than or equal to the texture deviation threshold, the deviation trend of the die-cut material is determined to be a weak deviation trend; In response to the texture deviation coefficient being less than the texture deviation threshold, the deviation trend of the die-cut material is determined to be a strong deviation trend.

6. The AOI detection method for die-cutting material interconnection based on AI visual sorting according to claim 5, characterized in that, The process of calculating the similarity between the scratch feature vector and several historical feature vectors to determine whether the die-cutting material is a first-grade die-cutting material or a second-grade die-cutting material includes the following steps: The scratch edge features and scratch sharpness features are input into the learning model for feature encoding to generate scratch feature vectors; The first similarity is calculated as the similarity between the scratch feature vector and the historical feature vector of the first-grade die-cut material; The similarity between the scratch feature vector and the historical feature vector of the second-grade die-cut material is calculated as the second similarity. If the first similarity is greater than or equal to the second similarity, the die-cutting material is determined to be first-grade die-cutting material. If the first similarity is less than the second similarity, the die-cutting material is determined to be second-grade die-cutting material. The scratch edge features include edge continuity and edge smoothness, and the scratch sharpness features include tip sharpness and tip convergence. The scratch severity of the first-grade die-cutting material is lower than that of the second-grade die-cutting material.

7. The AOI detection method for inter-line cutting materials based on AI visual sorting according to claim 6, characterized in that, The process of determining whether to energize the second-grade die-cutting material based on the sorting ratio of the second-grade die-cutting material within a preset period includes: The sorting percentage is determined as the ratio of the number of second-grade die-cut materials identified within a preset period to the total number of sorting times of the die-cut materials. If the sorting ratio is greater than the sorting ratio threshold, it is determined that the second-level die-cut material is energized. The total number of sorting times is the total number of first-grade die-cut materials and second-grade die-cut materials.

Images