Film attachment defect detection method and system based on image processing

Abstract

The application relates to the technical field of image processing, and discloses a film lamination defect detection method and system based on image processing. The method comprises the following steps: collecting and preprocessing a film-conductive layer interface by using a polarization enhancement imaging device to obtain an enhanced image; applying double-layer gradient edge detection to extract the edge of a lamination area and mark a micro-defect sensitive area; obtaining an interface smoothness index and a lamination uniformity feature through multi-scale topographic feature extraction; inputting the features into a support vector classifier to identify bubbles and wrinkles; performing three-dimensional reconstruction on the identification result to generate a bubble distribution and stress field diagram; and finally evaluating the lamination quality and outputting a completeness score. The application realizes automatic and accurate detection and quantitative evaluation of small defects on the film-conductive layer lamination interface.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a film bonding defect detection method and system based on image processing. BACKGROUND

[0002] With the rapid development of flexible electronics, OLED display screens, and wearable devices, the film and conductive layer bonding process plays an increasingly important role in electronic product manufacturing. Existing film bonding defect detection techniques mainly include manual visual inspection, ultrasonic detection, X-ray fluoroscopy, and traditional machine vision technology. Manual visual inspection relies on the experience of inspectors, is low in efficiency, and is easily affected by subjective factors; ultrasonic detection can detect internal bubbles, but has limited resolution and slow detection speed; X-ray fluoroscopy technology requires high equipment, is expensive, and has radiation risks; traditional machine vision technology mainly relies on grayscale contrast and simple threshold segmentation, usually uses a single image acquisition mode, and then uses edge detection and morphological processing to analyze the image. These methods meet the production needs to some extent, especially for obvious large-area defects.

[0003] However, existing technologies have significant shortcomings in micro-defect detection and quantitative evaluation. First, traditional optical imaging methods are difficult to capture micro-bubbles and wrinkles at the film-conductive layer interface, especially defects smaller than 100 microns are often ignored; second, conventional image processing algorithms have limited ability to distinguish different types of defects (such as bubbles and wrinkles), resulting in high misjudgment rate; third, existing detection methods mostly stay in two-dimensional plane analysis stage, and cannot effectively reconstruct and quantify the three-dimensional morphology and depth information of defects; finally, there is a lack of systematic defect evaluation system, making it difficult to standardize and quantify the bonding quality, resulting in large differences between different detection devices and different operators, affecting the continuous optimization of production process and quality control. With the development of electronic products towards high precision and high reliability, these technical limitations have become a bottleneck restricting the improvement of product yield and performance. SUMMARY

[0004] The present application provides a film bonding defect detection method and system based on image processing, which is used to realize the automatic and accurate detection and quantitative evaluation of micro-defects (especially micro-bubbles and micro-wrinkles) at the film-conductive layer bonding interface.

[0005] In a first aspect, the application provides a film lamination defect detection method based on image processing, which comprises: differentiating optical acquisition of a film-conductive layer interface by a polarization enhancement imaging device, and pre-processing an original image by interface reflection enhancement filtering to obtain an interface enhancement pre-processed image; edge extraction of a lamination area is performed according to the interface enhancement pre-processed image by a double-layer gradient edge detection algorithm to obtain a micro-defect sensitive area marker map; feature extraction of a segmented area is performed according to the micro-defect sensitive area marker map by a multi-scale topographic feature extractor to obtain an interface smoothness index and a lamination uniformity feature vector; the lamination uniformity feature vector is input into a support vector classifier for bubble-crimp recognition to obtain an interface abnormal area mapping result; three-dimensional reconstruction of micro-bubbles and micro-crimps is performed based on the interface abnormal area mapping result by a contour extraction algorithm to obtain an interlayer bubble distribution and a crimp stress field map; and the film-conductive layer lamination quality is evaluated according to the interlayer bubble distribution and the crimp stress field map by an interface integrity evaluation model to obtain a lamination interface integrity score.

[0006] In a second aspect, the application provides a film lamination defect detection system based on image processing, which comprises:

[0007] An acquisition module is configured to differentiate optical acquisition of a film-conductive layer interface by a polarization enhancement imaging device, and pre-process an original image by interface reflection enhancement filtering to obtain an interface enhancement pre-processed image;

[0008] An analysis module is configured to perform edge extraction of a lamination area according to the interface enhancement pre-processed image by a double-layer gradient edge detection algorithm to obtain a micro-defect sensitive area marker map;

[0009] An extraction module is configured to perform feature extraction of a segmented area according to the micro-defect sensitive area marker map by a multi-scale topographic feature extractor to obtain an interface smoothness index and a lamination uniformity feature vector;

[0010] An identification module is configured to input the lamination uniformity feature vector into a support vector classifier for bubble-crimp recognition to obtain an interface abnormal area mapping result;

[0011] A reconstruction module is configured to perform three-dimensional reconstruction of micro-bubbles and micro-crimps based on the interface abnormal area mapping result by a contour extraction algorithm to obtain an interlayer bubble distribution and a crimp stress field map;

[0012] An evaluation module is configured to evaluate the film-conductive layer lamination quality according to the interlayer bubble distribution and the crimp stress field map by an interface integrity evaluation model to obtain a lamination interface integrity score.

[0013] The third aspect of the present application provides a computer device, comprising a memory and at least one processor, the memory storing instructions; the at least one processor calling the instructions in the memory to enable the computer device to perform the image processing-based film lamination defect detection method.

[0014] The fourth aspect of the present application provides a computer readable storage medium, the computer readable storage medium storing instructions, when the instructions are run on a computer, enabling the computer to perform the image processing-based film lamination defect detection method.

[0015] In the technical scheme provided in the present application, the polarization-enhanced imaging device is used for differential optical acquisition, and the interface reflection-enhanced filter preprocessing is combined, thereby effectively improving the optical contrast and visibility of the interface micro-defects, solving the problem that the traditional imaging is difficult to capture the micro interface abnormalities, and especially for the 30 mu level defects, the detection rate is as high as 95%; the double-layer gradient edge detection algorithm organically fuses the Canny operator and the Sobel operator and performs adaptive threshold segmentation, thereby realizing effective extraction of edge information, greatly improving the detection sensitivity of irregular edges, enabling the micro defect area to be accurately marked, and the edge positioning accuracy is improved to the sub-pixel level; the multi-scale topographic feature extractor analyzes the features of the segmented region, comprehensively considers the geometric morphological parameters, the gray scale statistical features and the wavelet decomposition features, and constructs a comprehensive defect feature description system, the interface smoothness index and the lamination uniformity feature vector can quantitatively represent the severity of the defects, the dimension of the feature extraction reaches 32, and a rich feature space is provided for accurate classification; the support vector classifier realizes accurate identification of the bubble-crease type based on artificial intelligence technology, the classifier is obtained by a large number of labeled sample training, a nonlinear kernel function is used to map a multi-dimensional feature space, a decision boundary is automatically optimized, the classification accuracy is above 94%, and the key role of the machine learning algorithm in defect identification is embodied, and the maximum interval characteristic of the support vector machine model makes the classification result more robust, effectively dealing with noise interference and feature fluctuation in the production environment; the contour extraction algorithm is combined with the depth mapping matrix, three-dimensional reconstruction of the micro-bubble and the micro-crease is realized, two-dimensional image information is converted into a three-dimensional space structure, and key parameters such as the volume and the depth of the defects are provided, which has important guiding value for stress analysis and quality prediction; the interface integrity evaluation model comprehensively considers various defect features and indexes, establishes a standardized scoring mechanism, and converts complex defect information into an intuitive integrity score, the model combines expert knowledge and data-driven intelligent evaluation methods, can not only objectively evaluate the lamination quality of the current sample, but also track the process change trend, provides an accurate basis for parameter adjustment in the production process, realizes closed-loop control of defect detection and process optimization, and finally significantly improves the yield and consistency of the film lamination product. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort based on these drawings.

[0017] Figure 1 An embodiment of the method for detecting defects in film bonding based on image processing in the embodiments of the present application is shown in the figure.

[0018] Figure 2 An embodiment of the system for detecting defects in film bonding based on image processing in the embodiments of the present application is shown in the figure.

[0019] Figure 3 The structural schematic block diagram of the computer device in the embodiments of the present application is shown in the figure. DETAILED DESCRIPTION

[0020] The embodiments of the present application provide a method and system for detecting defects in film bonding based on image processing. The terms "first", "second", "third", "fourth" and the like (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily mean a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the term "comprising" or "having" and any variation thereof is intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0021] For the convenience of understanding, the specific process of the embodiments of the present application will be described below. Please refer to Figure 1 An embodiment of the method for detecting defects in film bonding based on image processing in the embodiments of the present application includes:

[0022] Step S101, differentiating optical acquisition of the film-conductive layer interface by a polarization enhancement imaging device, and pre-processing the original image by an interface reflection enhancement filter to obtain an interface enhancement pre-processed image;

[0023] Step S102, edge extraction of the bonding area by a double-layer gradient edge detection algorithm according to the interface enhancement pre-processed image to obtain a micro-defect sensitive area marker;

[0024] Step S103, according to the micro-defect sensitive area marking map, the interface smoothness index and the adhesion uniformity feature vector are obtained by performing feature extraction on the segmented area through a multi-scale morphology feature extractor.

[0025] Step S104, the adhesion uniformity feature vector is input into a support vector classifier to perform bubble-crease recognition, and an interface abnormal area mapping result is obtained.

[0026] Step S105, based on the interface abnormal area mapping result, three-dimensional reconstruction is performed on the micro-bubbles and micro-creases through a contour extraction algorithm, and an interlayer bubble distribution and a crease stress field map are obtained.

[0027] Step S106, according to the interlayer bubble distribution and the crease stress field map, the film-conductive layer adhesion quality is evaluated through an interface integrity evaluation model, and an adhesion interface integrity score is obtained.

[0028] It can be understood that the execution subject of the present application can be a film adhesion defect detection system based on image processing, and can also be a terminal or a server, and the specific implementation is not limited herein. The server is taken as an example for description of the embodiments of the present application.

[0029] Specifically, when the polarization-enhanced imaging device performs differential optical acquisition on the film-conductive layer interface, a polarized light source with a 45° polarization angle is used to irradiate the sample surface, and this angle can maximize the reflection contrast of the film-conductive layer interface. The line-scan camera captures the reflection intensity change at a set scan frequency, and obtains the interface reflection original image. The Gaussian filter algorithm is first applied to the original image to eliminate random noise, and the filter kernel size is usually set to 5x5 pixels. The smoothed image retains the edge information. Then, adaptive histogram equalization processing is performed, the image is divided into 8x8 sub-blocks, and histogram equalization is performed in each sub-block to extract the interface reflection contrast feature. The interface reflection enhancement filter uses a Gabor filter bank, sets 8 different direction filters, and the frequency parameter is 0.5 to strengthen the brightness change feature at the junction of the film and the conductive layer. In the geometric correction compensation stage, the bilinear interpolation method is used to eliminate lens distortion and shooting angle deviation. Finally, a non-linear contrast adjustment function is used for dynamic range expansion to obtain the interface enhancement preprocessed image.

[0030] When the double-layer gradient edge detection algorithm extracts the edge of the bonding area, the Canny operator and the Sobel operator are applied in parallel for primary edge extraction. The Canny operator includes four steps: Gaussian filtering, gradient calculation, non-maximum suppression and double-threshold detection, which is suitable for detecting fine edges; the Sobel operator calculates the horizontal and vertical direction gradients, which is suitable for detecting strong edges. The two sets of edge images are fused through logical AND operation to retain the edge points detected by both operators. The adaptive threshold segmentation algorithm calculates the optimal segmentation threshold value by the Otsu method, and divides the image into edge and non-edge regions. The double-layer gradient calculation calculates the first-order and second-order gradients at each pixel point, and records the gradient direction to form a gradient feature map. The region growing process starts from the detected edge starting point, and expands the region according to the gradient direction and gradient similarity, connecting discontinuous edges with a distance of less than 5 pixels. The watershed segmentation marker takes the complete edge contour as the boundary line of the watershed algorithm, divides the film bonding area, the conductive layer area and the interface area, and obtains the micro-defect sensitive area marker map. When the multi-scale topographic feature extractor extracts features from the segmented region, each region in the micro-defect sensitive area marker map is assigned a unique ID number, and a region search table is established. The geometric morphological parameters of each region are calculated, including area (total number of pixels in the region), perimeter (number of boundary points), equivalent radius and shape factor. The gray level histogram is extracted and the first-order statistics (mean, standard deviation, skewness and kurtosis) and second-order statistics (energy, contrast, correlation and entropy) are calculated. The interface smoothness index is calculated by fitting the interface line with a quadratic polynomial, calculating the curvature, and integrating the curvature sequence to obtain the smoothness index. The wavelet decomposition process uses Haar wavelet to perform three-level decomposition on the region image, extracts the low-frequency and high-frequency components, and constructs a multi-scale frequency spectrum feature library. The interface line is evenly divided into 20 equal-length subsegments, and the local smoothness index is calculated, and the bonding uniformity index is synthesized by weighted average. The feature dimension reduction uses principal component analysis to reduce the original feature space to 32 dimensions, and performs normalization processing to obtain the bonding uniformity feature vector.

[0031] In the bubble-crease recognition process of the support vector classifier, the uniformity feature vector is first processed by Z-score standardization. The SVM classifier with RBF kernel function is established, and the penalty parameter and kernel parameter are optimized. The distance scalar of the sample to the classification boundary is calculated, and the normal vector and bias term are obtained by classifier training. In the two-stage classification, the area is first divided into normal and abnormal by a threshold, and then the abnormal area is divided into bubble-type defects and crease-type defects by another threshold. The confidence score reflects the reliability of classification, and a threshold of 0.75 is set to filter low-confidence areas. The high-reliability recognition result is mapped back to the original image according to the original coordinates, and the bubble area is marked in red and the crease area is marked in yellow, forming the interface abnormal area mapping result. In the three-dimensional reconstruction of micro-bubbles and micro-creases by the contour extraction algorithm, the bubble defect area and the crease defect area are first separated by connected component analysis. The contour tracking algorithm is used to extract the bubble boundary line segment, and the bubble area and circularity are calculated. The depth mapping matrix is constructed based on the gray gradient information, and the correspondence between the gray value and the depth is established. The skeleton extraction algorithm obtains the crease center line through morphological thinning operation, and measures the width distribution perpendicular to the center line. The stress distribution is calculated by the thin plate model, and the mapping relationship between local deformation and stress is established. The bubble position and volume data are integrated into a three-dimensional spatial distribution model, and the crease stress value matrix data is expressed by a heat map, with red representing high stress area and blue representing low stress area.

[0032] In the evaluation of the film-conductive layer bonding quality by the interface integrity evaluation model, the bubble density value (the number of bubbles per unit area), the bubble size distribution value (the frequency distribution of bubbles of each size), the maximum bubble diameter value, and the total bubble volume value are extracted from the interlayer bubble distribution. The total crease length value, the average crease width value, the crease peak stress value, and the crease direction consistency value (the proportion of the length of the creases in the main direction to the total length) are extracted from the crease stress field map. The mapping relationship between the features and the integrity score is established, and the weighted defect score is calculated by balancing the bubble factor weight and the crease factor weight. According to the weighted defect score, the defect severity level is determined, with 90≤Q≤100 as excellent, 80≤Q<90 as good, 60≤Q<80 as medium, and Q<60 as poor. Finally, the bonding interface integrity score is obtained.

[0033] For example, in the process of film bonding of a certain OLED display screen flexible substrate, the polarization-enhanced imaging device collects the original image at a 45° polarization angle, and the image resolution is 4096x3072 pixels. After Gaussian filtering and adaptive histogram equalization, the interface contrast is significantly improved. The Gabor filter set enhances the interface line features. In the double-layer gradient detection, the low threshold of the Canny operator is set to 50, and the high threshold is set to 150; the threshold of the Sobel operator is set to 40. Three main bonding areas are detected, among which the shape factor of area #2 is 0.78 and the interface smoothness index is 2.34, indicating that there is an abnormality. After feature dimension reduction, the support vector classifier is input, and it is identified that there is a micro-bubble group in area #2, with the maximum bubble diameter of 86 microns and the depth of 12 microns; area #3 has a micro-fold with a length of 234 microns and an average width of 25 microns. After comprehensive evaluation, the bubble factor score is 15, the fold factor score is 12, and the bonding interface integrity score is 79, which belongs to the intermediate bonding quality, and the bonding process parameters need to be adjusted to improve the bonding effect.

[0034] In the embodiments of the present application, the differential optical acquisition is performed by the polarization enhancement imaging device, and the interface reflection enhancement filtering preprocessing is combined, thereby effectively improving the optical contrast and visibility of the interface micro-defects, solving the problem that the traditional imaging is difficult to capture the micro interface anomalies, and especially for the 30 pm level defects, the detection rate is as high as 95%; the double-layer gradient edge detection algorithm organically fuses the Canny operator and the Sobel operator and is segmented through the adaptive threshold, thereby realizing the effective extraction of the edge information, greatly improving the detection sensitivity of the irregular edge, and enabling the micro defect area to be accurately marked, and the edge positioning accuracy is improved to the sub-pixel level; the multi-scale topographic feature extractor analyzes the features of the segmented region, comprehensively considers the geometric morphological parameters, the gray statistical features and the wavelet decomposition features, constructs a comprehensive defect feature description system, the interface smoothness index and the uniformity feature vector can quantitatively represent the severity of the defect, the dimension of the feature extraction reaches 32, and a rich feature space is provided for accurate classification; the support vector classifier realizes the accurate identification of the bubble-crease type based on the artificial intelligence technology, the classifier is obtained through a large number of labeled samples training, a nonlinear kernel function is used to map the multi-dimensional feature space, the decision boundary is automatically optimized, the classification accuracy is above 94%, which reflects the key role of the machine learning algorithm in the defect identification, and the maximum interval characteristics of the support vector machine model make the classification result more robust, effectively dealing with the noise interference and feature fluctuation in the production environment; the contour extraction algorithm is combined with the depth mapping matrix, realizes the three-dimensional reconstruction of the micro-bubble and the micro-crease, converts the two-dimensional image information into a three-dimensional spatial structure, and provides key parameters such as the volume and the depth of the defect, which has important guiding value for stress analysis and quality prediction; the interface integrity evaluation model comprehensively considers various defect features and indexes, establishes a standardized scoring mechanism, and converts complex defect information into an intuitive integrity score, the model combines expert knowledge and data-driven intelligent evaluation methods, can not only objectively evaluate the bonding quality of the current sample, but also track the process change trend, provides accurate basis for parameter adjustment in the production process, realizes closed-loop control of defect detection and process optimization, and finally significantly improves the yield and consistency of the film bonding product.

[0035] In a specific embodiment, the process of step S101 can specifically include the following steps:

[0036] (1) A polarized light source with a polarization angle of 45° is arranged to irradiate the film-conductive layer interface, the reflected light intensity change is captured by a line scan camera, and an interface reflection original image is obtained;

[0037] (2) Based on the interface reflection original image, random noise is eliminated by a Gaussian filtering algorithm, and a noise reduction processing image is generated;

[0038] (3) Apply adaptive histogram equalization to the denoised image to extract the interface reflection contrast features and form a contrast enhanced image;

[0039] (4) Send the contrast enhanced image into the interface reflection enhancement filter to strengthen the brightness variation features at the junction of the thin film and the conductive layer and generate an interface level highlighted image;

[0040] (5) Perform geometric correction compensation on the interface level highlighted image to eliminate lens distortion and shooting angle deviation and form a geometrically accurate interface image;

[0041] (6) Perform dynamic range expansion on the geometrically accurate interface image through a nonlinear contrast adjustment function to obtain an interface enhanced preprocessed image.

[0042] Specifically, the process of illuminating the thin film-conductive layer interface with a polarized light source with a polarization angle of 45° and capturing the reflected light intensity variation with a line-scan camera to obtain an interface reflection original image is performed. The polarized light source generates light waves with a specific vibration direction, and the polarization angle of 45° is set because this angle can maximize the contrast between the reflected light and the transmitted light at the thin film-conductive layer interface. When the polarized light illuminates the thin film-conductive layer interface, due to the difference in refractive index between the two materials, specific reflection and refraction phenomena occur at the interface, and these optical characteristics are directly related to the interface bonding quality. The line-scan camera is equipped with a polarization filter that matches the light source, and the scanning frequency is usually set to 4000 lines / s, and the resolution is 4096 pixels / line, which can capture the variation of the reflected light intensity row by row to form an original image matrix containing interface reflection information. Based on the interface reflection original image, random noise is eliminated by a Gaussian filtering algorithm to generate a denoised image. The Gaussian filtering algorithm uses a two-dimensional Gaussian function as the filter kernel for convolution operation on the image, and the filter kernel size is usually set to 5x5 pixels and the standard deviation value is set to 1.2. The Gaussian function has a bell-shaped distribution in the spatial domain, with the center weight being the largest and decreasing towards the four corners, which makes it effective in smoothing image noise while preserving edge information. The specific operation is to slide the Gaussian filter kernel over the original image pixel by pixel, calculate the weighted sum of the pixel values in the kernel coverage area and the corresponding kernel weights, and replace the center pixel value. After this processing, the value of each pixel is replaced by the weighted average value of its neighborhood pixels, and the random noise is smoothed to obtain a denoised image.

[0043] The adaptive histogram equalization is applied to the denoised image to extract the interface reflection contrast features and form a contrast enhanced image. The adaptive histogram equalization divides the image into 8x8 sub-blocks, and histogram equalization is performed in each sub-block separately, and then the complete image is synthesized by bilinear interpolation to avoid the contrast over-enhancement problem caused by global histogram equalization. For each sub-block, the gray histogram is first counted to represent the gray value distribution, and then the cumulative distribution function is calculated to map the original gray value to a new gray value to achieve contrast enhancement. The bilinear interpolation is used to maintain smooth transition between adjacent sub-blocks to avoid block effect, and finally the contrast enhanced image is formed.

[0044] The contrast enhanced image is sent into the interface reflection enhancement filter to strengthen the brightness change features at the junction of the thin film and the conductive layer, and an interface level highlighting image is generated. The interface reflection enhancement filter uses a Gabor filter bank with 8 filters in different directions, and the frequency parameter is set to 0.5 and the bandwidth parameter is set to 1.0. The Gabor filter is a band-pass filter with good direction selectivity and frequency selectivity, which is suitable for detecting texture features of a specific direction and scale. The convolution results of the input image with each direction Gabor filter are calculated to obtain 8 response images, and then these response images are fused by maximum value operation to strengthen the brightness change features at the junction of the thin film and the conductive layer, and generate the interface level highlighting image. Geometric correction compensation is performed on the interface level highlighting image to eliminate lens distortion and shooting angle deviation, and a geometrically accurate interface image is formed. The distortion model is established by referring to the coordinates of the control points on the calibration plate, including radial distortion and tangential distortion. Radial distortion is caused by lens shape, which makes straight lines in the image appear curved; tangential distortion is caused by the non-parallelism of the lens and the image sensor, which causes local deformation of the image. According to the distortion model, the position of each pixel in the original image is mapped to the corrected image, and the gray value of each pixel in the corrected image is assigned by bilinear interpolation method. At the same time, the perspective transformation matrix is used to correct the shooting angle deviation, so that the interface features in the image present the true geometric relationship, and the geometrically accurate interface image is formed.

[0045] The geometrically accurate interface image is subjected to dynamic range expansion by a nonlinear contrast adjustment function to obtain an interface enhanced preprocessing image. The nonlinear contrast adjustment function uses S-shaped curve transformation to stretch the middle gray region and compress the bright and dark regions, enhancing the middle tone details while avoiding loss of details in highlight and shadow regions. The dynamic range of the processed image is expanded, and the subtle features of the thin film-conductive layer interface are more prominent, forming the final interface enhanced preprocessing image.

[0046] For example, in the process of detecting the film bonding of an OLED display panel, first, a 45° polarized light source is used to irradiate the sample, and a line-scan camera acquires an original image at a speed of 4000 lines / s, with a resolution of 4096*3072 pixels, 16-bit gray level per pixel, and records the reflected light intensity distribution of the film-conductive layer interface. Then, a 5*5 Gaussian filter kernel is applied to the original image for noise reduction processing, effectively suppressing the random noise introduced by the camera sensor and environmental light fluctuations, and significantly improving the signal-to-noise ratio. Then the image is divided into 8*8 sub-blocks for adaptive histogram equalization, with each sub-block being 512*384 pixels. By calculating the gray level distribution and cumulative distribution function of each sub-block, local contrast enhancement is achieved, making the interface reflection features more obvious. Subsequently, an 8-direction Gabor filter set is applied to process the image, and the 8 response images after filtering are fused by maximum value operation, and the brightness gradient value at the junction of the film and the conductive layer is significantly enhanced. Geometric correction is performed on the enhanced image to eliminate radial distortion and tangential distortion, and the position deviation of the interface line in the image is significantly reduced after correction. Finally, the image dynamic range is expanded through nonlinear S-shaped curve transformation, and a 16-bit gray level interface enhancement preprocessing image is output, laying the foundation for subsequent double-layer gradient edge detection. By comparing the images before and after processing, it is observed that the signal-to-noise ratio of the film-conductive layer interface in the preprocessed image is significantly improved, the edge sharpness is significantly improved, the contrast of the small bubbles and wrinkles area is enhanced, and the accuracy and sensitivity of the subsequent defect detection are greatly improved.

[0047] In a specific embodiment, the process of performing step S102 can specifically include the following steps:

[0048] (1) Apply Canny operator and Sobel operator to the interface enhancement preprocessing image respectively to perform primary edge extraction, and obtain two groups of edge images;

[0049] (2) Fuse the two groups of edge images through logical AND operation to generate a preliminary edge detection result;

[0050] (3) Apply an adaptive threshold segmentation algorithm to the preliminary edge detection result, set a dynamic threshold to control the edge detection sensitivity, and obtain an edge candidate region;

[0051] (4) Apply double-layer gradient calculation to the edge candidate region to extract gradient direction and gradient amplitude features, and form a gradient feature map;

[0052] (5) Perform region growing processing on the gradient feature map to connect discontinuous edges and construct a complete edge contour;

[0053] (6) Perform watershed segmentation labeling on the complete edge contour to divide the film bonding area, the conductive layer area and the interface area, and obtain a micro-defect sensitive area marker.

[0054] Specifically, when applying Canny and Sobel operators to the interface-enhanced preprocessed image for primary edge extraction, it is necessary to first understand the working principles and characteristics of the two operators. Canny operator includes four steps of Gaussian filtering, gradient calculation, non-maximum suppression, and double-threshold detection, which can detect fine edge structure and has strong noise resistance. In practical applications, Gaussian filtering uses a 5x5 filter kernel to smooth the image, then calculates the gradient of the image in the horizontal and vertical directions, further removes non-edge points by non-maximum suppression, and finally determines the true edge using the double-threshold method (usually the low threshold is set to 50 and the high threshold is set to 150). Sobel operator is relatively simple, which uses two 3x3 convolution kernels to calculate the gradient in the horizontal and vertical directions, then calculates the gradient amplitude and direction, and finally determines whether it is an edge pixel by a single threshold (usually set to 40). These two operators have their own advantages: Canny is suitable for detecting fine edges, while Sobel is more suitable for detecting strong edges. The combination of the two can fully capture the edge information in the image. When fusing the two sets of edge images to generate the preliminary edge detection result, the pixel-level logical AND operation is performed on the edge images generated by Canny and Sobel respectively. The logical AND operation requires that the corresponding pixels in the two images are both marked as edges (value 1), and the corresponding pixels in the result image are also marked as edges; if the corresponding pixels in any image are non-edge (value 0), then the corresponding position in the result image is also non-edge. This fusion method retains the strong edges recognized by both operators, filters out the points detected by a single operator that may be noise or false edges, thereby improving the reliability of edge detection. The preliminary edge detection result is a binary image that contains preliminary contour information of the film-conductive layer interface region.

[0055] An adaptive threshold segmentation algorithm is used for the preliminary edge detection result, and a dynamic threshold is set to control the sensitivity of edge detection. In the process of obtaining the edge candidate region, the Otsu method is used to automatically determine the optimal segmentation threshold. The Otsu method is an adaptive threshold algorithm based on the image gray histogram, which determines the optimal threshold by maximizing the inter-class variance. The processing process includes: calculating the normalized gray histogram of the image, which represents the proportion of each gray value pixel in the total pixels; for each possible threshold, calculate the inter-class variance of the foreground (gray value greater than the threshold) and the background (gray value less than or equal to the threshold); select the threshold that maximizes the inter-class variance as the optimal threshold. The adaptability is reflected in that the threshold will be dynamically adjusted according to the gray distribution of the input image, rather than using a fixed threshold. The optimal threshold is used for secondary thresholding of the preliminary edge detection result to obtain the edge candidate region. In the process of applying double-layer gradient calculation to the edge candidate region to extract gradient direction and gradient amplitude features and forming a gradient feature map, first, calculate the first-order gradient of the original interface enhancement preprocessing image, i.e. the derivative in the horizontal and vertical directions, which is realized using the Sobel operator or Prewitt operator; then calculate the gradient amplitude and gradient direction; then calculate the gradient again on the first-order gradient map to obtain the second-order gradient and second-order gradient direction. The double-layer gradient feature fusion is realized by weighted combination of the first-order and second-order gradient information, where the first-order gradient weight is usually set to 0.7 and the second-order gradient weight is set to 0.3. The gradient feature map contains the amplitude and direction information of the edge, which provides an important basis for subsequent edge connection and region segmentation.

[0056] When the gradient feature map is subjected to region growing processing to connect discontinuous edges and construct a complete edge contour, first, determine the seed point, which is usually an edge point with the locally maximum gradient amplitude; then, starting from the seed point, expand along the edges on both sides according to the gradient direction, and check whether the adjacent pixels meet the growth conditions; the growth conditions include the gradient direction similarity and the gradient amplitude threshold, and the difference between the gradient direction of the adjacent pixel and the gradient direction of the current edge point is less than a certain angle (usually 20 degrees), and the gradient amplitude is greater than a set threshold (usually 30% of the total gradient amplitude); the pixels that meet the conditions are added to the edge set and used as new expansion points; when the distance between the two edge endpoints is less than a preset threshold (usually 5 pixels), connect them by straight line or curve interpolation to form a continuous edge. After region growing processing, a complete edge contour is obtained. In the process of performing watershed segmentation labeling on the complete edge contour to divide the film bonding area, the conductive layer area and the interface area and obtain the micro-defect sensitive area label map, first, use the complete edge contour as the boundary line of the watershed algorithm; then calculate the distance from each non-edge pixel to the nearest edge pixel to form a distance map; find the local maximum points in the distance map as label points, and each label point represents a region; label the label points on the boundary line of the watershed algorithm to obtain the label map; finally, label the label map to obtain the micro-defect sensitive area label map. Figure 1The input watershed algorithm is started, and region expansion is performed from the marker point according to the gradient information of the image until adjacent regions meet or reach the boundary; finally, the segmented regions are classified into the film attachment area, the conductive layer area and the interface area according to the region characteristics (such as position, shape, gray scale statistics, etc.), and a micro-defect sensitive region marking map with different markers is generated.

[0057] For example, in the film attachment detection of an OLED display panel, after an interface enhancement preprocessed image with a resolution of 4096x3072 pixels is obtained, Canny operator and Sobel operator are applied respectively for edge extraction. The Canny operator uses a Gaussian kernel for smoothing, and the low threshold value is set to 50 and the high threshold value is set to 150 to detect fine edge structures; the Sobel operator uses a threshold value of 75 to detect main strong edges. The two sets of edge images are fused by logical AND operation, and the edges detected by both operators are retained, and the false edges generated by the independent operator are effectively filtered out. The Otsu method is applied to the preliminary edge result after fusion to calculate the best threshold value, and a dynamic threshold value of 82 is obtained, and the edge gray value is segmented again to form an edge candidate region. The double-layer gradient calculation is applied to the edge candidate region, the first-order gradient is calculated using the Sobel operator, and the second-order gradient is calculated by applying the Sobel operator again on the basis of the first-order gradient, and the gradient feature map is obtained. When the gradient feature map is subjected to region growing processing, the point with the locally maximum gradient amplitude is selected as the initial seed, and the growing condition is set to the gradient direction difference being less than 20 degrees and the gradient amplitude being greater than the set threshold value, the multiple discontinuous edges are connected, and the complete edge contour is constructed. Finally, the watershed algorithm is applied to the complete edge contour, the edge is taken as the watershed line, the internal region is taken as the catchment basin, a plurality of regions are marked out in total, and they are classified into the film attachment area, the conductive layer area and the interface area through position and shape characteristics. In the marking map, the film attachment area is marked in red, the conductive layer area is marked in green, and the interface area is marked in blue. When the characteristics of the interface area are analyzed, it is found that the edge curvature of part of the regions is abnormal, and the edge continuity is poor, which provides accurate positioning of the micro-defect sensitive region for subsequent multi-scale topographic feature extraction.

[0058] In a specific embodiment, the process of performing step S103 can specifically include the following steps:

[0059] (1) Region numbering and indexing are performed on the micro-defect sensitive region marking map, a unique identifier is assigned to each region, and a region retrieval table is established;

[0060] (2) The boundary point set of each segmented region is sequentially read from the region retrieval table, and the geometric and morphological parameters of the region are calculated, specifically including area, perimeter, equivalent radius and shape factor;

[0061] (3) For each segmented region, extract its gray level histogram, and calculate the first and second order statistics to form the gray level distribution descriptor;

[0062] (4) Apply curvature calculation algorithm to the interface region between the film and the conductive layer, calculate the curvature value of each point on the interface line through quadratic polynomial fitting, and integrate the curvature value sequence on the interface line to obtain the interface smoothness index;

[0063] (5) Apply wavelet decomposition processing to each segmented region, sequentially extract low-frequency components and high-frequency components at different scales, and construct a multi-scale spectral feature library;

[0064] (6) Divide the interface region between the film and the conductive layer into a fixed number of equal-length subsegments, calculate the local smoothness index of each subsegment, and synthesize the overall uniformity index by weighted average;

[0065] (7) Integrate all extracted morphological features, statistical features and multi-scale features, perform feature dimension reduction and normalization processing, and generate a fixed-dimension uniformity feature vector.

[0066] Specifically, the micro-defect sensitive region marker map is numbered and indexed, and each region is assigned a unique identifier. In the process of establishing the region retrieval table, the micro-defect sensitive region marker map is scanned by a connected component analysis algorithm to identify all independent regions composed of mutually connected pixels. Connected component analysis usually uses a two-pass scanning method. In the first pass, the image is scanned from left to right and from top to bottom, each non-background pixel is assigned a temporary label, and the equivalence relationship of adjacent labels is recorded. In the second pass, the image is scanned again, and the equivalent labels are merged to ensure that each connected domain has a unique identifier. In this way, the film bonding area, conductive layer area and interface area in the marker map each obtain a different region ID, which is stored in the region retrieval table, containing region ID, region type label, region center coordinates, boundary coordinate sequence and other key information, providing a query basis for subsequent feature extraction. The boundary point set of each segmented region is read in sequence from the region retrieval table, and the geometric and morphological parameters of the region are calculated, including area, perimeter, equivalent radius and shape factor. The area is calculated by counting the total number of pixels in the region, and the perimeter is obtained by the boundary tracking algorithm. The equivalent radius is the radius value when the region is assumed to be a perfect circle, which is calculated by area conversion. The shape factor is a dimensionless parameter that describes the regularity of the shape of the region, which is calculated by the relationship between area and perimeter. The closer the value is to 1, the closer the shape is to a circle, and the smaller the value, the more irregular the shape. These geometric and morphological parameters together constitute a quantitative description of the shape of the region, which can effectively identify abnormal shape regions such as irregular bubbles or wrinkles.

[0067] For each segmented region, its gray histogram is extracted, and the first-order statistics and second-order statistics are calculated to constitute the gray distribution descriptor. In the process of constructing the gray distribution descriptor, the gray values of the corresponding region in the original interface enhanced preprocessed image are first read, the frequency of the occurrence of each gray level pixel is counted, and a gray histogram is generated. The first-order statistics include the mean (the arithmetic mean of all pixel gray values in the region), the standard deviation (the dispersion degree of the gray values), the skewness (the degree of asymmetry of the gray distribution), and the kurtosis (the sharpness degree of the gray distribution). The second-order statistics are calculated based on the gray co-occurrence matrix, and include the energy (texture uniformity), the contrast (gray difference degree), the correlation (gray correlation degree), and the entropy (gray distribution randomness). These statistical characteristics can comprehensively describe the pixel gray distribution in the region, and provide an important basis for defect recognition.

[0068] In the process of applying the curvature calculation algorithm to the film-conductive layer interface region, calculating the curvature value of each point on the interface line through quadratic polynomial fitting, and integrating the curvature value sequence on the interface line to obtain the interface smoothness index, first, the boundary point set of the interface region is extracted from the region search table, and is reordered according to the spatial position to form an ordered interface line. For each point on the interface line and the points in its neighborhood, a quadratic polynomial curve is fitted using the least squares method to obtain the curve parameters. The curvature value at each point on the fitted curve is calculated, and the curvature value size reflects the bending degree of the interface at that point. The region with large curvature change often corresponds to the poorly fitted position. By integrating the curvature value sequence on the entire interface line, the interface smoothness index representing the overall interface smoothness is obtained. The larger the value, the more unsmooth the interface, and the higher the defect possibility. In the process of applying wavelet decomposition processing to each segmented region, sequentially extracting the low-frequency components and high-frequency components at different scales, and constructing a multi-scale frequency spectrum feature library, the region image is decomposed by using discrete wavelet transform. The wavelet decomposition divides the image into a low-frequency approximation component and three high-frequency detail components in the horizontal, vertical, and diagonal directions. Usually, 3-level decomposition is used to generate multi-scale representations from coarse to fine. The low-frequency component reflects the overall gray distribution and large-scale structure of the region, and the high-frequency component captures the edge, texture, and other detailed features. By extracting the feature parameters such as energy, variance, and mean of each component, a complete multi-scale frequency spectrum feature library is constructed, which can analyze the region characteristics at different scales and effectively identify defects of various sizes.

[0069] In the process of dividing the film-conductive layer interface region into a fixed number of equal-length subsegments, calculating the local smoothness index of each subsegment, and synthesizing the overall fitting uniformity index by weighted averaging, first, the interface line is equally divided into N subsegments (usually N=20) according to the arc length. For the jth subsegment, the local smoothness index LSI_j is calculated, and its calculation formula is:

[0070]

[0071] where and denotes the start and end arc length parameters of the jth subsection, denotes the curvature value of the interface line at arc length parameter s, denotes the standard deviation of curvature values within the subsection, , and are weight coefficients, respectively controlling the importance of integrated curvature, maximum curvature and curvature variance in evaluation, usually taking values of = 0.5, = 0.3, = 0.2. The overall bonding uniformity index is obtained by weighted average of local smoothness indices of each subsection, with the calculation formula as:

[0072]

[0073] where denotes the uniformity bonding index, is the weight coefficient of the jth subsection, related to the position and importance of the subsection, usually with higher weight in the central region than in the edge region.

[0074] In the process of integrating all extracted morphological features, statistical features and multi-scale features, feature dimension reduction and normalization, a fixed dimension of bonding uniformity feature vector is generated. First, all kinds of features are organized into high-dimensional feature vectors according to the pre-defined format. Since the dimensions and numerical ranges of different features are different, normalization is needed. Usually, Z-score standardization or Min-Max normalization method is used to map all feature values to similar numerical ranges. Then, principal component analysis (PCA) is used for feature dimension reduction, and the principal components accounting for more than 95% of the total variance are retained, usually compressing the original hundreds of dimensions to about 32 dimensions. Finally, according to the pre-defined sorting rule, the features are reorganized to generate a fixed dimension of bonding uniformity feature vector, which is used as the input of the subsequent support vector classifier.

[0075] For example, in the detection of film attachment defects of a certain OLED flexible display panel, 28 regions are identified in the micro-defect sensitive region marking diagram, including 10 film attachment regions, 8 conductive layer regions and 10 interface regions, each region is assigned a unique ID from 1 to 28. The geometric parameter calculation of the interface region with ID 15 shows that its area is 12468 pixels, the perimeter is 842 pixels, the equivalent radius is 63.1 pixels, and the shape factor is 0.22, which is obviously lower than the shape factor of the normal region (usually greater than 0.6), indicating that the region shape is irregular and may have defects. Extracting the gray histogram of the region, the mean value is 127.3, the standard deviation is 42.8, the skewness is 0.67 and the kurtosis is 3.24, which is significantly different from the statistical characteristics of the normal region. The interface line is fitted by a quadratic polynomial, and the maximum curvature value of 5.7 is found at the arc length parameter s = 120, which is much higher than the average curvature of 1.2, and the integral gives the interface smoothness index of 3.82, which exceeds the warning threshold of 3.0. The interface line is divided into 20 equal length subsegments, of which the 12th subsegment has the highest local smoothness index of 6.14, and the corresponding weight is set to 1.5, and the weight of the other subsegments is 1.0, and the uniformity attachment index UBI is calculated to be 2.76, which is significantly higher than the mean value of 0.95 of the normal attachment region. Through wavelet 3-level decomposition, it is found that the high-frequency energy of the region in the vertical direction is 3 times higher than that of the normal region, indicating that there is a structure abnormality in the vertical direction. Integrating all the characteristics and reducing the dimension through PCA, a 32-dimensional attachment uniformity feature vector is obtained, which is highly similar to the characteristics of the bubble type defect in the sample library.

[0076] In a specific embodiment, the process of performing step S104 can specifically include the following steps:

[0077] (1) Dimensionless and numerical normalization of the attachment uniformity feature vector are performed through standardization processing to generate a normalized feature vector;

[0078] (2) A sample training matrix is established based on the normalized feature vector, the support vector classifier parameters are optimized by gradient descent method, and a feature space decision boundary is constructed;

[0079] (3) The attachment uniformity feature vector of the sample to be detected is input into the trained support vector classifier, and the distance scalar of the sample to the classification boundary is calculated;

[0080] (4) The sample region is classified into two stages according to the distance scalar, first into normal region and abnormal region, and then the abnormal region is subdivided into bubble type defect and wrinkle type defect;

[0081] (5) The confidence score of each identified defect region is calculated, and a threshold is set to filter low confidence regions, and high reliability recognition results are retained;

[0082] (6) The high-reliability recognition result is mapped back to the original image according to the original coordinate position, to generate region markers of different color coding, forming the interface abnormal region mapping result.

[0083] Specifically, in the process of generating the normalized feature vector by standardizing the fitting uniformity feature vector, first, the values from different measurement methods in the feature vector need to be processed, which have significant differences in dimension and range. Common standardization methods include Z-score standardization and min-max standardization. Z-score standardization subtracts the mean of each feature value from the feature value and divides it by the standard deviation, so that the mean of the feature distribution is 0 and the standard deviation is 1; min-max standardization maps the feature value to a fixed interval (usually [0, 1] or [-1, 1]). Standardization ensures that features of various scales are given equal importance in the classifier, avoiding the dominance of features with large values in the classification process. In the steps of establishing a sample training matrix based on the normalized feature vector and optimizing the support vector classifier parameters by gradient descent method to construct the feature space decision boundary, first, a large number of sample feature vectors of known categories (normal, bubble-type defects, and wrinkle-type defects) are collected to form a training matrix. The support vector classifier is a supervised learning algorithm that finds the optimal hyperplane, which can separate samples of different categories in the feature space. The core task is to find the maximum interval hyperplane that maximizes the distance of sample points of different categories from the hyperplane. Gradient descent method optimizes the classifier parameters, including the penalty coefficient and the kernel function parameter, through iterative method. The penalty coefficient controls the complexity and error tolerance of the model; the kernel function maps the original features to a high-dimensional space, converting the linearly inseparable problem into a linearly separable one. Common kernel functions include linear kernel, polynomial kernel, and radial basis function kernel, among which the radial basis function kernel usually performs best for thin film fitting defect detection.

[0084] In the process of inputting the uniformity feature vector of the sample to be detected into the trained support vector classifier and calculating the distance scalar of the sample from the classification boundary, first, the feature vector of the sample to be detected is subjected to the same standardization processing as the training data, and then is sent to the trained classifier. The classifier calculates the distance of the sample from the decision boundary, i.e. the decision function value. This distance scalar not only indicates which class the sample belongs to, but also reflects the degree of certainty of the classification: the farther from the boundary, the more reliable the classification result. For a multi-class classification problem, the one-to-one or one-to-many strategy is usually adopted, and the results of various binary classifiers are finally integrated to determine the class of the sample. According to the size of the distance scalar, the sample region is classified in two stages. First, the region is divided into normal and abnormal regions, and then the abnormal region is subdivided into bubble-type defects and wrinkle-type defects. In the process of classification judgment, two thresholds are set. In the first stage, the sample region is divided into normal and abnormal regions using a threshold. If the distance scalar exceeds the threshold, it is considered to be a normal region, and if it is lower than the threshold, it is determined to be an abnormal region. In the second stage, the identified abnormal region is further classified into bubble-type defects or wrinkle-type defects by a specially trained bubble-wrinkle binary classifier or a two-level threshold based on the distance scalar. This two-stage classification strategy effectively improves the classification accuracy, especially for classification problems with obvious hierarchical structure.

[0085] In the process of calculating the confidence score for each identified defect region and setting a threshold to filter low-confidence regions and retain high-reliability identification results, the confidence score is calculated based on the distance of the sample from the decision boundary. The greater the absolute value of the distance, the higher the confidence; and a distance close to zero indicates that the sample is near the decision boundary, and the reliability of the classification result is lower. Usually, the distance value is mapped to the interval of 0 to 1 as the confidence score, and then a threshold (such as 0.75) is set for screening, retaining only the identification results with a confidence higher than the threshold. This screening mechanism effectively reduces the risk of misjudgment and improves the reliability of the detection results.

[0086] In the process of mapping high-reliability identification results back to the original image according to the original coordinate position to generate region markers with different color coding and forming the interface abnormal region mapping result, the region search table established in the previous step is used to establish a correspondence between the classification result and the original region. Normal regions are usually marked in green, bubble-type defect regions are marked in red, and wrinkle-type defect regions are marked in yellow. Each region is colored and marked according to the original coordinate position and superimposed on the original image to form an intuitive interface abnormal region mapping result, which facilitates process personnel to quickly locate and solve problems.

[0087] For example, in the OLED display film bonding quality detection of a smart watch, after obtaining the 32-dimensional bonding uniformity feature vector from the early processing, the first step is standardization. Subtract the mean value of the corresponding feature in the training sample set from each feature, and then divide it by its standard deviation to achieve the normalization of the feature. Taking the roughness feature as an example, the original value is 3.82, the corresponding mean value is 1.2, and the standard deviation is 0.8. The standardized value is 3.28. After standardizing all features, a normalized feature vector is formed. Based on a training matrix of 1000 containing 300 normal samples, 400 bubble defect samples and 300 wrinkle defect samples, a support vector machine with a radial basis kernel function is used for training, the penalty coefficient is set to 10, and the kernel parameter is set to 0.01. The parameters are adjusted through 5-fold cross-validation, and the final model reaches a classification accuracy of 94% on the validation set. The sample to be detected is input into the trained classifier, and the distance value from the decision boundary is calculated as -2.3. The negative value indicates that the sample is on the side of the abnormal class. In the two-stage classification, the first threshold is set to 0, and the distance less than 0 is judged as abnormal; the second threshold is set to -1.5, and the distance less than -1.5 is judged as a bubble defect, so the sample is classified as a bubble defect. The confidence score is 0.85, which is higher than the threshold requirement of 0.75, so the recognition result is retained. Finally, the defect area is marked in red on the display imaging map, and other defect areas detected are also marked with corresponding colors, forming a complete interface abnormal area mapping map, which intuitively shows the defect position and type in the film bonding process, providing accurate basis for production line process adjustment.

[0088] In a specific embodiment, the process of performing step S105 can specifically include the following steps:

[0089] (1) Regionally segment the defect marking area in the interface abnormal area mapping result to separate the bubble defect area and the wrinkle defect area;

[0090] (2) Extract the bubble boundary line segment using the contour tracking algorithm for the separated bubble defect area, calculate the bubble area and shape feature;

[0091] (3) Establish a depth mapping matrix based on the gray gradient information of the bubble area, estimate the depth of each bubble area, and generate a bubble height model;

[0092] (4) Apply the skeleton extraction algorithm to the wrinkle defect area to obtain the wrinkle center line and width distribution, and calculate the direction and curvature of the wrinkle;

[0093] (5) According to the wrinkle morphological feature and position information, calculate the interface stress distribution using the thin plate model to construct a stress value matrix of the wrinkle area;

[0094] (6) The position and volume data of all bubbles are integrated into a three-dimensional spatial distribution model to form the interlayer bubble distribution;

[0095] (7) The fold stress value matrix data are summarized, and the stress concentration area and propagation path are expressed in the form of a heat map to generate a fold stress field map.

[0096] Specifically, when regionally segmenting the defect marked area in the interface anomaly region mapping result to separate the bubble defect area and the fold defect area, first, the classification result marked map generated by the support vector classifier is read, and different types of defect areas are distinguished by label values. Bubble-type defect areas are usually marked in red or with the value 1, and fold-type defect areas are marked in yellow or with the value 2. The same type of marking is grouped using the connected domain analysis technique to ensure that each independent defect area obtains a unique identifier. This process establishes a bubble defect set and a fold defect set, and records the respective position, area, and boundary information, laying the foundation for subsequent separate processing. The contour tracking algorithm is used to extract bubble boundary line segments for the separated bubble defect area, and the boundary tracking algorithm is used to process each bubble area one by one when calculating the bubble area and shape features. This algorithm starts from any point on the edge of the bubble area, and then finds the adjacent boundary point in the clockwise or counterclockwise direction, until it returns to the starting point, forming a complete closed boundary line. After obtaining the boundary, the bubble area (total number of pixels inside the boundary), perimeter (number of boundary points), circularity (a measure of how close the bubble shape is to a circle), ratio of major axis to minor axis (reflects the degree of bubble flattening), and other shape feature parameters are calculated. These parameters can effectively distinguish normal small bubbles from abnormal large bubbles.

[0097] In the process of establishing the depth mapping matrix based on the gray gradient information of bubble regions, and estimating the depth of each bubble region to generate the bubble height model, the optical property changes caused by bubbles at the film interface are utilized. First, the gray distribution of the bubble region is extracted from the original enhanced image. The bubble center usually presents a bright or dark feature, forming a gradient with the surrounding area. By analyzing this gradient distribution, a corresponding relationship between the gray value and the depth is established, and the two-dimensional gray image is converted into a three-dimensional depth map. For each bubble, the depth change is estimated according to the gray change trend from the edge to the center, forming a three-dimensional profile model of the bubble. This gray-based depth estimation method fully utilizes the advantages of non-contact optical detection, and can obtain the three-dimensional structure information of the bubble without damaging the sample. When applying the skeleton extraction algorithm to the wrinkle defect area to obtain the wrinkle center line and width distribution, and calculating the strike and curvature of the wrinkle, a morphological thinning operation is used to gradually "strip" the edge pixels of the wrinkle region, and the center line structure is retained. During the thinning process, the boundary points are continuously deleted, but the region connectivity is maintained, and finally the skeleton line with a single-pixel width, i.e. the center line of the wrinkle, is obtained. After obtaining the center line, the distance from each center line point to the nearest boundary point is calculated to form a wrinkle width distribution map. Further analysis of the direction change of the center line obtains the main strike (direction angle) and curvature change of the wrinkle, which reflect the directionality and severity of the wrinkle formation, providing key inputs for stress analysis.

[0098] According to the wrinkle morphological characteristics and position information, the interfacial stress distribution is calculated using a thin plate model, and the stress value matrix of the wrinkle region is constructed by regarding the film-conductive layer interface as a thin plate structure subjected to force deformation. Based on the obtained wrinkle geometric characteristics (center line position, width distribution, curvature, etc.), the stress state of each point is estimated through the thin plate deformation theory. The strain of the wrinkle region is closely related to the center line curvature and wrinkle height, and accordingly the stress distribution of each point of the film is calculated. The stress calculation considers physical parameters such as material elastic modulus and Poisson's ratio, generating a stress value matrix covering the entire wrinkle region, which quantitatively represents the stress state of each point at the interface. When integrating the position and volume data of all bubbles into a three-dimensional spatial distribution model to form the interlayer bubble distribution, a unified three-dimensional coordinate system is first established, with the horizontal and vertical coordinates corresponding to the film plane position and the vertical coordinate representing the depth direction. The center position, maximum depth, volume and other parameters of each detected bubble are mapped into this coordinate system to form point cloud data. For a dense bubble group, clustering analysis is performed to identify the bubble aggregation area. The final generated three-dimensional model visually displays the spatial distribution characteristics of the bubbles in the film interface, including bubble density, size distribution, depth distribution and other key information.

[0099] In the process of generating the fold stress field map, the stress value matrix calculated is converted into a color-coded heat map. A red-yellow-green-blue color spectrum is usually adopted, with red representing a high stress area and blue representing a low stress area. The stress distribution condition is intuitively displayed through color gradation, and the stress concentration points and stress propagation paths are highlighted. The fold center line and the equal stress line are superimposed on the heat map, further enhancing the visualization effect, so that the process personnel can quickly locate the key problem area.

[0100] For example, in the film attachment detection of a flexible display screen, the support vector classifier identifies 3 bubble defects and 2 fold defects. First, the defect areas are separated and independent defect sets are established. The contour tracking algorithm is applied to the largest bubble area to extract 154 boundary points, calculate the bubble area of 1862 pixels, the circumference of 154 pixels, and the circularity of 0.84, indicating that the bubble shape is relatively regular. Analyzing the gray scale distribution of the bubble area, it is found that the central area has a gray scale value of 216, and the average gray scale of the edge area is 142. A gray scale-depth mapping relationship is established to estimate the maximum depth of the bubble as 15 microns, and the overall structure is semi-spherical. The skeleton extraction algorithm is applied to the most prominent fold area to obtain a center line with a length of 326 pixels, and the average width along the center line is 23 pixels, and the maximum width is 48 pixels, which appears in the middle of the fold. Analyzing the center line direction, the main direction angle is 37 degrees, which is roughly parallel to the edge of the display screen, indicating that the fold may be caused by edge stress conduction. According to the geometric characteristics of the fold, the stress distribution is calculated, and the maximum stress value appears at the widest part of the fold, reaching 3.2 MPa, which is 4 times the normal interfacial stress. Integrating all bubble data into a three-dimensional model, it is found that bubbles are mainly concentrated in the upper right corner of the display area, showing a clear clustering effect, with the maximum bubble depth of 22 microns. The fold stress heat map clearly shows a high stress zone extending from the edge inward, which provides an important reference for optimizing the attachment process. According to these analysis results, the attachment pressure distribution and temperature parameters are adjusted, significantly reducing the bubble and fold defect rate in subsequent production.

[0101] In a specific embodiment, the process of performing step S106 can specifically include the following steps:

[0102] (1) Extract bubble feature indicators from the interlayer bubble distribution, including bubble density value, bubble size distribution value, bubble maximum diameter value, and bubble volume total value;

[0103] (2) Extract fold feature indicators from the fold stress field map, including fold length total value, fold average width value, fold peak stress value, and fold direction consistency value;

[0104] ​(3) input the extracted bubble characteristic index and wrinkle characteristic index into the interface integrity evaluation model, and establish a mapping relationship between the characteristics and the integrity score;

[0105] (4) assign bubble factor weight and wrinkle factor weight in the interface integrity evaluation model, and calculate weighted defect score according to the severity of each index;

[0106] (5) compare the weighted defect score with the standard defect level threshold to determine the defect severity level of the current sample;

[0107] (6) according to the defect severity level and the weighted defect score, convert to the standardized bonding interface integrity score through the interface integrity evaluation model.

[0108] Specifically, in the process of extracting bubble characteristic indexes from the interlayer bubble distribution, including bubble density value, bubble size distribution value, bubble maximum diameter value and bubble volume total value, first read the bubble data in the three-dimensional space distribution model. The bubble density value calculation method is to count the number of bubbles in unit area, expressed as the number of bubbles per square centimeter, which reflects the concentration of bubble distribution; the bubble size distribution value is to group all detected bubbles by diameter size, and count the proportion of the number of bubbles in each size interval, which is usually divided into four levels: micro (<30 μm), small (30-100 μm), medium (100-300 μm) and large (>300 μm), forming a size distribution histogram; the bubble maximum diameter value is the value of the largest diameter among all bubbles, which directly reflects the size of the most serious bubble defect; the bubble volume total value is the cumulative volume of all bubbles, which reflects the overall size of bubble defects. The volume calculation is based on the depth mapping matrix established earlier. These four indexes together constitute a comprehensive quantitative description of bubble defects, providing a data basis for subsequent defect evaluation. In the process of extracting wrinkle characteristic indexes from the wrinkle stress field diagram, including wrinkle length total value, wrinkle average width value, wrinkle peak stress value and wrinkle direction consistency value, first analyze the wrinkle stress field heat map and wrinkle center line data. The total length of the wrinkle length total value is obtained by adding the lengths of all detected wrinkle center lines, which reflects the overall size of the wrinkle defect; the wrinkle average width value is obtained by calculating the average width of each wrinkle along the center line, which represents the average severity of the wrinkle; the wrinkle peak stress value is the maximum stress value detected in the stress field diagram, which represents the intensity of the most serious stress concentration position; the wrinkle direction consistency value is calculated by analyzing the direction angle of each wrinkle center line, which is specifically calculated by counting the proportion of the length of the main direction (the direction angle interval with the highest frequency) in the total length, which reflects the directional characteristics of the wrinkle formation. High direction consistency usually indicates that there is a systematic bonding problem. These four indexes quantify the characteristics of wrinkle defects from different angles, providing key evidence for bonding quality evaluation.

[0109] When the extracted bubble feature indicators and wrinkle feature indicators are input into the interface integrity evaluation model to establish the mapping relationship between the features and the integrity score, a multi-feature fusion evaluation strategy is adopted. The interface integrity evaluation model is a hybrid model based on expert knowledge and machine learning, which maps the standardized features to the defect score space. Standardization processing enables features with different dimensions and different numerical ranges to be compared and combined. For bubble features, the bubble density, size distribution, maximum diameter, and volume total are compared with pre-determined standard thresholds to calculate the normalized defect score; for wrinkle features, the total wrinkle length, average width, peak stress, and direction consistency are also compared with standard thresholds to obtain the normalized score. This standardization process establishes a quantitative relationship between the original feature indicators and the defect severity, laying the foundation for subsequent weighted calculation. In the interface integrity evaluation model, bubble factor weights and wrinkle factor weights are assigned, and in the process of calculating the weighted defect score according to the severity of each indicator, weight coefficients are set according to the influence of each type of defect on product performance in different application scenarios. The assignment of bubble factor weights and wrinkle factor weights follows the actual product requirements, for example, for display screens that require high optical transparency, bubble defects have a greater impact, so the bubble factor weight will be higher; for flexible screens that require high mechanical bending performance, wrinkle defects have a greater impact, so the wrinkle factor weight is higher. In specific implementation, the bubble defect score is usually multiplied by the bubble factor weight, the wrinkle defect score is multiplied by the wrinkle factor weight, and then the sum is taken to obtain the weighted defect score. This weighting mechanism ensures that the evaluation results accurately reflect the actual impact of different types of defects on product quality.

[0110] In the process of comparing the weighted defect score with the standard defect level threshold to determine the defect severity level of the current sample, multi-level defect severity classification standards are established. The defect severity is usually divided into four levels: slight (A level), general (B level), severe (C level), and extremely severe (D level), each level corresponding to a score interval. For example, weighted defect scores of 0-10 correspond to A level, 10-25 correspond to B level, 25-40 correspond to C level, and greater than 40 correspond to D level. By comparing the calculated weighted defect score with these thresholds, the defect severity level to which the sample belongs is determined. This grading mechanism makes the quality evaluation results more intuitive, facilitating process control and quality management.

[0111] In the process of converting the defect severity level and the weighted defect score into a standardized bonding interface integrity score through the interface integrity evaluation model, a composite scoring function is used for calculation. The standardized bonding interface integrity score usually uses a percentage system, with a perfect score of 100 representing perfect bonding and a score of 0 representing complete failure. The scoring formula is:

[0112]

[0113] wherein, Interface Integrity Score represents the interface integrity score; Score Max represents the maximum score value, usually 100; Chi Global is the global adjustment coefficient, controlling the score decreasing rate; Defect Score is the weighted defect score; Grade Penalty is the grade penalty factor, increasing with the defect grade; Grade Value is the numerical representation of the defect severity grade, A grade is 1, B grade is 2, C grade is 3, D grade is 4; Reference Grade is the reference grade, usually set as 2; Grade Sensitivity is the grade sensitivity parameter, controlling the score difference between different grades; Tolerance Coefficient is the tolerance coefficient of the current product type; Max Tolerance Coefficient is the maximum tolerance coefficient; Tolerance Exponent is the tolerance exponent, controlling the nonlinearity degree of the tolerance effect. Through this formula, the defect score, severity grade and product-specific factors are comprehensively considered to generate the final interface integrity score.

[0114] The above describes the thin film bonding defect detection method based on image processing in the embodiments of the present application. The thin film bonding defect detection system based on image processing in the embodiments of the present application is described below. Please refer to Figure 2 The thin film bonding defect detection system based on image processing in the embodiments of the present application includes one embodiment:

[0115] The acquisition module is configured to perform differential optical acquisition on the thin film-conductive layer interface through the polarization enhancement imaging device, and perform preprocessing on the original image through the interface reflection enhancement filter to obtain an interface enhancement preprocessing image.

[0116] The analysis module is configured to perform edge extraction on the bonding area according to the interface enhancement preprocessing image through a double-layer gradient edge detection algorithm to obtain a micro-defect sensitive area marker.

[0117] The extraction module is configured to perform feature extraction on the segmented area through a multi-scale topographic feature extractor according to the micro-defect sensitive area marker to obtain an interface smoothness index and a bonding uniformity feature vector.

[0118] The recognition module is configured to input the bonding uniformity feature vector into a support vector classifier for bubble-crinkle recognition to obtain an interface abnormal area mapping result.

[0119] The reconstruction module is configured to perform three-dimensional reconstruction on the micro-bubbles and micro-creases by a contour extraction algorithm based on the interface abnormal area mapping result, to obtain an interlayer bubble distribution and a crease stress field diagram;

[0120] The evaluation module is configured to evaluate the film-conductive layer bonding quality according to the interlayer bubble distribution and the crease stress field diagram by an interface integrity evaluation model, to obtain a bonding interface integrity score.

[0121] Through the cooperation of the above components, the differential optical collection is performed by the polarization enhancement imaging device, and the interface reflection enhancement filtering preprocessing is combined, the optical contrast and visibility of the interface micro-defects are effectively improved, the problem that the traditional imaging is difficult to capture the micro interface abnormalities is solved, especially for the 30 μm level defects, the detection rate is as high as 95%; the double-layer gradient edge detection algorithm organically fuses the Canny operator and the Sobel operator and performs adaptive threshold segmentation, the effective extraction of the edge information is realized, the detection sensitivity of the irregular edge is greatly improved, and the micro-defect area can be accurately marked, and the edge positioning accuracy is improved to the sub-pixel level; the multi-scale morphology feature extractor performs feature analysis on the segmented area, comprehensively considers the geometric morphological parameters, the gray statistical features and the wavelet decomposition features, a comprehensive defect feature description system is constructed, the interface smoothness index and the bonding uniformity feature vector can quantitatively represent the severity of the defects, the dimension of the feature extraction reaches 32, and a rich feature space is provided for accurate classification; the support vector classifier realizes accurate identification of the bubble-crease type based on the artificial intelligence technology, the classifier is obtained by a large number of labeled samples training, a nonlinear kernel function is used to map a multi-dimensional feature space, a decision boundary is automatically optimized, the classification accuracy is above 94%, the key role of the machine learning algorithm in the defect identification is embodied, and the maximum interval characteristics of the support vector machine model make the classification result more robust, and effectively cope with the noise interference and feature fluctuation in the production environment; the contour extraction algorithm is combined with the depth mapping matrix, three-dimensional reconstruction of the micro-bubbles and the micro-creases is realized, two-dimensional image information is converted into a three-dimensional space structure, and key parameters such as the volume and the depth of the defects are provided, which has important guiding value for stress analysis and quality prediction; the interface integrity evaluation model comprehensively considers various defect features and indexes, establishes a standardized scoring mechanism, and converts complex defect information into an intuitive integrity score, the model combines expert knowledge and data-driven intelligent evaluation methods, can not only objectively evaluate the bonding quality of the current sample, but also track the process change trend, provides an accurate basis for parameter adjustment in the production process, realizes closed-loop control of defect detection and process optimization, and finally significantly improves the yield and consistency of the film bonding product.

[0122] Reference Figure 3 In the embodiment of the present application, a computer device, which can be a server, is also provided. The internal structure of the computer device can be as shown inFigure 3 The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected by a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store corresponding data in the embodiment. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement the above method.

[0123] Those skilled in the art can understand that the structure shown in the embodiment is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. Figure 3

[0124] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program. The computer program is executed by the processor to implement the above method. It can be understood that the computer readable storage medium in the embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0125] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. Any reference to the memory, storage, database or other medium provided by the present application and used in the embodiments can include non-volatile and / or volatile memory. The non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. The volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM) and memory bus dynamic RAM, etc.

[0126] ​Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, system and unit can refer to the corresponding processes in the foregoing method embodiments, and will not be described here.

[0127] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0128] The above-described and above-mentioned embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An image processing-based thin film attachment defect detection method, characterized by, The film attachment defect detection method based on image processing comprises: The film-conductive layer interface is differentially optically collected by a polarization enhancement imaging device, and the original image is preprocessed by interface reflection enhancement filtering to obtain an interface enhancement preprocessing image; According to the interface enhancement preprocessing image, edge extraction is performed on the attachment area by a double-layer gradient edge detection algorithm to obtain a micro-defect sensitive area marker; According to the micro-defect sensitive area marker, feature extraction is performed on the segmented area by a multi-scale topographic feature extractor to obtain an interface smoothness index and an attachment uniformity feature vector; The attachment uniformity feature vector is input into a support vector classifier for bubble-crease recognition to obtain an interface abnormal area mapping result, including: the attachment uniformity feature vector is subjected to dimension unification and numerical normalization by standardization processing to generate a normalized feature vector; a sample training matrix is established based on the normalized feature vector, a support vector classifier parameter is optimized by a gradient descent method, and a feature space decision boundary is constructed; the attachment uniformity feature vector of the sample to be detected is input into the trained support vector classifier, and the distance scalar of the sample to the classification boundary is calculated; according to the distance scalar, the sample area is classified into two stages, first into a normal area and an abnormal area, and then the abnormal area is subdivided into a bubble-type defect and a crease-type defect; the confidence score of each recognized defect area is calculated, and a threshold is set to filter low-confidence areas and retain high-reliability recognition results; the high-reliability recognition results are mapped back to the original image according to the original coordinate position to generate region markers of different color coding, forming the interface abnormal area mapping result; Based on the interface abnormal area mapping result, micro-bubbles and micro-creases are three-dimensionally reconstructed by a contour extraction algorithm to obtain an interlayer bubble distribution and a crease stress field map, including: the defect marker area in the interface abnormal area mapping result is regionally segmented to separate bubble defect areas and crease defect areas; The bubble boundary line segment is extracted by a contour tracking algorithm for the separated bubble defect area, and the bubble area and shape feature are calculated; a depth mapping matrix is established based on the gray gradient information of the bubble area to estimate the depth of each bubble area and generate a bubble height model; the skeleton extraction algorithm is applied to the crease defect area to obtain the crease center line and width distribution, and the trend and curvature of the crease are calculated; according to the crease morphological feature and position information, the interface stress distribution is calculated by a thin plate model to construct a stress value matrix of the crease area; the position and volume data of all bubbles are integrated into a three-dimensional spatial distribution model to form the interlayer bubble distribution; the stress value matrix data of the crease is summarized, and the stress concentration area and propagation path are expressed in the form of a heat map to generate the crease stress field map; According to the interlayer bubble distribution and the crease stress field map, the film-conductive layer attachment quality is evaluated by an interface integrity evaluation model to obtain an attachment interface integrity score.

2. The image processing-based thin film attachment defect detection method according to claim 1, characterized by, The film-conductive layer interface is differentially optically collected by a polarization enhancement imaging device, and the original image is preprocessed by interface reflection enhancement filtering to obtain an interface enhancement preprocessing image, including: The interface reflection original image is obtained by setting a polarized light source with a polarization angle of 45 degrees to irradiate the film-conductive layer interface and capturing the reflected light intensity change by a line scanning camera; Based on the interface reflection original image, random noise is eliminated by a Gaussian filtering algorithm to generate a noise reduction processing image; The noise reduction processing image is subjected to adaptive histogram equalization processing to extract interface reflection contrast features and form a contrast enhancement image; The contrast enhancement image is sent into an interface reflection enhancement filter to strengthen the brightness change features at the junction of the film and the conductive layer, and an interface level highlighting image is generated; The interface level highlighting image is subjected to geometric correction compensation to eliminate lens distortion and shooting angle deviation, and a geometrically accurate interface image is formed; The geometrically accurate interface image is subjected to dynamic range expansion by a nonlinear contrast adjustment function to obtain the interface enhancement preprocessing image.

3. The image processing-based thin film attachment defect detection method according to claim 1, characterized by, The interface enhancement preprocessing image is subjected to edge extraction in the bonding area by a double-layer gradient edge detection algorithm to obtain a micro-defect sensitive area marker, including: The interface enhancement preprocessing image is subjected to primary edge extraction by applying Canny and Sobel operators respectively to obtain two groups of edge images; The two groups of edge images are fused by logical AND operation to generate a preliminary edge detection result; the preliminary edge detection result is subjected to adaptive threshold segmentation algorithm to set a dynamic threshold to control edge detection sensitivity, and an edge candidate area is obtained; The edge candidate area is subjected to double-layer gradient calculation to extract gradient direction and gradient amplitude features, and a gradient feature map is formed; the gradient feature map is subjected to region growing processing to connect discontinuous edges and construct a complete edge contour; The complete edge contour is subjected to watershed segmentation labeling to divide the film bonding area, the conductive layer area and the interface area, and the micro-defect sensitive area marker is obtained.

4. The image processing-based thin film attachment defect detection method according to claim 1, characterized by, The interface smoothness index and the bonding uniformity feature vector are obtained by feature extraction in the segmented area by a multi-scale topographic feature extractor according to the micro-defect sensitive area marker, including: The micro-defect sensitive area marker is subjected to region numbering and indexing, and each region is assigned a unique identifier to establish a region retrieval table; the boundary point set of each segmented region is sequentially read from the region retrieval table, and the geometric morphological parameters of the region are calculated, including area, perimeter, equivalent radius and shape factor; For each segmented region, its gray level histogram is extracted, and first-order statistics and second-order statistics are calculated to form a gray level distribution descriptor; The curvature calculation algorithm is applied to the film-conductive layer interface area, the curvature value of each point on the interface line is calculated by quadratic polynomial fitting, and the curvature value sequence on the interface line is integrated to obtain the interface smoothness index; Wavelet decomposition processing is applied to each segmented region to sequentially extract low-frequency components and high-frequency components at different scales to construct a multi-scale frequency spectrum feature library; The film-conductive layer interface area is divided into a fixed number of equal-length subsegments, the local smoothness index of each subsegment is calculated, and the overall bonding uniformity index is synthesized by weighted averaging. Integrate all extracted morphological features, statistical features and multi-scale features, perform feature dimension reduction and normalization processing, and generate the fixed-dimension uniformity feature vector.

5. The image processing-based thin film attachment defect detection method according to claim 1, characterized by, The film-conductive layer bonding quality is evaluated according to the interlayer bubble distribution and the wrinkle stress field map through the interface integrity evaluation model to obtain the bonding interface integrity score, which includes: Bubble feature indexes are extracted from the interlayer bubble distribution, including bubble density value, bubble size distribution value, bubble maximum diameter value and bubble volume total value; Wrinkle feature indexes are extracted from the wrinkle stress field map, including wrinkle length total value, wrinkle average width value, wrinkle peak stress value and wrinkle direction consistency value; The extracted bubble feature indexes and wrinkle feature indexes are input into the interface integrity evaluation model to establish a mapping relationship between the features and the integrity score; Bubble factor weight and wrinkle factor weight are assigned in the interface integrity evaluation model, and weighted defect scores are calculated according to the severity of each index; The weighted defect scores are compared with the standard defect level threshold to determine the defect severity level of the current sample; and the defect severity level and the weighted defect score are converted into the standardized bonding interface integrity score through the interface integrity evaluation model.

6. An image processing-based film lamination defect detection system for implementing the image processing-based film lamination defect detection method according to any one of claims 1 to 5, characterized by, The film bonding defect detection system based on image processing includes: A collection module is configured to collect the film-conductive layer interface through a polarization-enhanced imaging device and to pre-process the original image through interface reflection-enhanced filtering to obtain an interface-enhanced pre-processed image. An analysis module is configured to extract edges of the bonding area through a double-layer gradient edge detection algorithm based on the interface-enhanced pre-processed image to obtain a micro-defect sensitive area marker. An extraction module is configured to extract features of the segmented area through a multi-scale topographic feature extractor based on the micro-defect sensitive area marker to obtain an interface smoothness index and a uniformity feature vector. An identification module is configured to input the uniformity feature vector into a support vector classifier to identify bubbles and wrinkles to obtain an interface abnormal area mapping result. A reconstruction module is configured to perform three-dimensional reconstruction of micro-bubbles and micro-wrinkles based on the interface abnormal area mapping result through a contour extraction algorithm to obtain an interlayer bubble distribution and a wrinkle stress field map. An evaluation module is configured to evaluate the film-conductive layer bonding quality according to the interlayer bubble distribution and the wrinkle stress field map through an interface integrity evaluation model to obtain a bonding interface integrity score.

7. A computer device, comprising: A memory and a processor are included, and the memory stores a computer program that can run on the processor, and when the processor executes the computer program, the processor implements the film bonding defect detection method based on image processing in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, the computer program causing a processor to execute the film bonding defect detection method based on image processing in any one of claims 1 to 5 when the computer program is run on the processor.

Images