Method and system for rapidly detecting defects of flexible screen after defoaming

Abstract

The invention relates to the technical field of defect detection, in particular to a method and a system for rapidly detecting defects after defoaming of a flexible screen. The method comprises the following steps: acquiring corresponding flexible screen defoaming images under different spectrum and angle irradiation, and performing image complementary registration fusion and structure phase morphology reconstruction to generate flexible screen surface three-dimensional morphology data; marking points are arranged on the edge of the flexible screen, and dynamic deformation compensation and curvature analysis of all points on the surface are carried out, so that principal curvatures and Gaussian curvatures corresponding to all the points on the surface of the flexible screen after bubble removal are obtained; performing abnormal curvature area identification and defect geometric feature analysis on the corresponding flexible screen after bubble removal to obtain defect geometric features of the flexible screen after bubble removal; and constructing a corresponding deep learning model and carrying out defect classification detection so as to detect and obtain corresponding bubble residue, indentation, scratch and degumming type defect results after the flexible screen is defoamed. According to the invention, the rapid and accurate detection process of the defects after defoaming of the flexible screen can be realized.

Description

Technical Field

[0001] The present invention relates to the field of defect detection technology, and in particular to a method and system for quickly detecting defects on a flexible screen after debubbling. Background Art

[0002] With the rapid development of flexible display technology, flexible screens are increasingly used in smartphones, wearable devices, televisions and other fields. Flexible screens have become an important part of modern electronic devices due to their thinness, bendability, and foldability. In the production process of flexible screens, the degassing process is a crucial step, which aims to remove bubbles on the surface or inside the screen to ensure the display quality of the screen. However, traditional flexible screen defect detection methods mostly rely on manual inspection and simple visual inspection tools. During the production process, manual inspection is not only time-consuming and labor-intensive, but also difficult to ensure that each screen can be accurately inspected during large-scale production. In addition, there are difficulties in discovering some minor defects and hidden defects, especially in the flexible screen after degassing. Defects such as bubble residue and fine scratches are difficult to detect, and it is difficult to cope with defect detection under complex backgrounds, thereby reducing detection efficiency and accuracy. Summary of the Invention

[0003] Based on this, it is necessary for the present invention to provide a method and system for quickly detecting defects of a flexible screen after debubbling, so as to solve at least one of the above technical problems.

[0004] To achieve the above objectives, a method for quickly detecting defects in a flexible screen after debubbling is provided, comprising the following steps:

[0005] Step S1: obtaining images of the flexible screen after debubbling corresponding to illumination at different spectra and angles, and performing image complementary registration and fusion on the images of the flexible screen after debubbling corresponding to illumination at different spectra and angles to generate a composite image of the flexible screen after debubbling;

[0006] Step S2: reconstructing the structural phase topography of the composite image of the flexible screen after degassing to generate three-dimensional topography data of the flexible screen surface; setting marking points on the edge of the flexible screen and obtaining the deformation of the edge marking position corresponding to the flexible screen according to the marking points; and dynamically compensating the three-dimensional topography data of the flexible screen surface based on the deformation of the edge marking position corresponding to the flexible screen to obtain deformation-compensated topography data of the screen surface;

[0007] Step S3: performing curvature analysis on the deformation compensation topography data of the screen surface to obtain the principal curvature and Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling; based on the principal curvature and Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling, identifying abnormal curvature areas of the corresponding flexible screen after de-bubbling to obtain abnormal defect curvature areas of the screen after de-bubbling;

[0008] Step S4: Analyze the defect geometric features of the abnormal defect curvature area after the screen is debubbled to obtain the defect geometric features of the flexible screen after debubbling; construct a deep learning model based on the U-Net architecture, and input the defect geometric features of the flexible screen after debubbling into the trained deep learning model for defect classification detection to detect the corresponding bubble residue, indentation, scratches and degumming type defect results after the flexible screen is debubbled.

[0009] Furthermore, step S1 includes the following steps:

[0010] Step S11: obtaining images of the flexible screen after debubbling under different spectral illumination, including visible light 400-700nm, near infrared 700-1100nm, and ultraviolet 200-400nm band light sources;

[0011] Step S12: obtaining images of the flexible screen after debubbling under different illumination angles, including low-angle illumination, vertical-angle illumination, and high-angle illumination;

[0012] Step S13: performing screen structure complementary information analysis on the corresponding images of the flexible screen after de-bubbling under different spectra and angles of illumination to obtain screen structure feature complementary information between the images of the flexible screen after de-bubbling;

[0013] Step S14: Based on the complementary information of screen structural features between the images of different flexible screens after de-bubbling, the images of the flexible screens after de-bubbling corresponding to the images under different spectra and angles are subjected to image complementary registration fusion optimization to ensure that the structural feature information under different spectra and angles is complementary and fused to the greatest extent, and generate a complementary registration optimization map of the flexible screen after de-bubbling;

[0014] Step S15: The complementary registration optimization image of the flexible screen after de-bubbling is composited and post-processed to enhance the details through image detail enhancement including high-pass filtering and sharpening processing, and locally correct its edges to eliminate the corresponding seams or incompletely fused image areas to generate a composite image of the flexible screen after de-bubbling.

[0015] Furthermore, step S13 includes the following steps:

[0016] Step S131: performing light source color deviation analysis on the corresponding de-bubbling images of the flexible screen under different spectral illumination, thereby converting the color space and measuring the color deviation caused by the light source change between the images with different spectra to obtain the light source change color deviation between the images with different spectra;

[0017] Step S132: obtaining corresponding flexible screen structure pixel feature points from the corresponding flexible screen after debubbling images under different irradiation angles, and performing screen structure geometric analysis on the corresponding flexible screen after debubbling images under different irradiation angles based on the flexible screen structure pixel feature points to obtain the flexible screen structure geometric relationship between the images after debubbling at different angles;

[0018] Step S133: Based on the color deviation of the light source change between the images after debubbling at different spectra and the geometric relationship of the flexible screen structure between the images after debubbling at different angles, the screen structure complementary information between the corresponding flexible screen images after debubbling is analyzed to obtain the screen structure feature complementary information between the images after debubbling at different flexible screens.

[0019] Furthermore, step S2 includes the following steps:

[0020] Step S21: reconstructing the structural phase topography of the composite image of the flexible screen after degassing to generate three-dimensional topography data of the flexible screen surface;

[0021] Step S22: setting a marking point on the edge of the flexible screen and obtaining the deformation amount of the edge marking position corresponding to the flexible screen according to the marking point;

[0022] Step S23: dynamically capturing the three-dimensional topography data of the flexible screen surface before and after deformation based on the deformation amount of the edge mark position corresponding to the flexible screen, to obtain the topography data of the flexible screen surface before and after deformation;

[0023] Step S24: Obtaining the deformation stress source distribution corresponding to each edge deformation mark point based on the deformation amount of the edge mark position corresponding to the flexible screen, and performing a shape stress compensation fitting analysis on the flexible screen surface topography data before and after deformation based on the deformation stress source distribution corresponding to each edge deformation mark point to generate a flexible screen surface topography stress dynamic compensation curve;

[0024] Step S25: performing dynamic deformation compensation on the three-dimensional surface topography data of the flexible screen based on the dynamic compensation curve of the surface topography stress of the flexible screen to obtain deformation-compensated topography data of the screen surface.

[0025] Furthermore, step S21 includes the following steps:

[0026] A sinusoidal fringe pattern is projected onto the composite image after de-bubbling the flexible screen to generate a sinusoidal fringe projection image of the flexible screen structure;

[0027] The phase deformation jump corresponding to each sinusoidal fringe in the fringe image is calculated by projecting the sinusoidal fringe image of the flexible screen structure, and phase unwrapping processing is performed based on the phase deformation jump corresponding to each sinusoidal fringe in the fringe image to obtain the two-dimensional phase information of the deformation of the flexible screen surface structure;

[0028] Based on the two-dimensional phase information of the deformation of the flexible screen surface structure, the three-dimensional morphology of each point in the corresponding flexible shielding surface structure in the composite image of the flexible screen after debubbling is reconstructed to generate three-dimensional morphology data of the flexible screen surface.

[0029] Furthermore, the step S22 of obtaining the deformation of the edge mark position corresponding to the flexible screen according to the mark point includes the following steps:

[0030] Obtain the corresponding surface deformation stress-strain distribution on the edge of the flexible screen according to the marked points;

[0031] Based on the surface deformation stress-strain distribution corresponding to each marking point on the edge of the flexible screen, the displacement increment of the corresponding marking point is calculated to obtain the displacement increment and displacement direction corresponding to each marking point on the edge of the flexible screen;

[0032] Based on the displacement direction corresponding to each mark point on the edge of the flexible screen, the position deformation of the displacement increment corresponding to each mark point on the edge of the flexible screen is predicted to obtain the position deformation of the edge mark corresponding to the flexible screen.

[0033] Furthermore, step S3 includes the following steps:

[0034] Step S31: performing high-order polynomial fitting on the screen surface deformation compensation topography data, so as to correct points that deviate from the normal curved surface due to environmental changes or screen quality fluctuations by using a local weighted regression model, thereby obtaining the screen surface abnormal shape calibration topography data;

[0035] Step S32: using a numerical differentiation method to perform a local curvature analysis on each point on the flexible screen surface in the screen surface abnormal morphology calibration topography data, to obtain the local curvature corresponding to each point on the flexible screen surface after de-bubbling;

[0036] Step S33: obtaining the corresponding normal direction of the flexible screen surface by calibrating the topography data of the abnormal morphology of the screen surface, and performing principal curvature calculation on the local curvature corresponding to each point on the surface of the flexible screen after de-bubbling based on the normal direction of the flexible screen surface, so as to obtain the principal curvature corresponding to each point on the surface of the flexible screen after de-bubbling;

[0037] Step S34: performing maximum value decomposition on the principal curvature corresponding to each point on the surface of the flexible screen after de-bubbling by combining a preset Jacobi matrix to obtain the maximum and minimum principal curvatures corresponding to each point on the surface of the flexible screen after de-bubbling, and performing Gaussian curvature calculation based on the maximum and minimum principal curvatures corresponding to each point on the surface of the flexible screen after de-bubbling to obtain the Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling;

[0038] Step S35: Based on the principal curvature and Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling, the abnormal curvature area of ​​the corresponding flexible screen after de-bubbling is identified, so as to calculate the corresponding curvature peak value according to the principal curvature and Gaussian curvature, and divide the corresponding abnormal defect area range according to the judgment result between the preset curvature division interval and the curvature peak value, so as to obtain the abnormal defect curvature area of ​​the screen after de-bubbling.

[0039] Furthermore, step S4 includes the following steps:

[0040] Step S41: obtaining the corresponding screen defect length, screen defect width, and screen defect depth after defoaming through the abnormal defect curvature area after defoaming;

[0041] Step S42: calculating the aspect ratio of the defect after defoaming the screen according to the length and width of the defect after defoaming the screen, to obtain the aspect ratio of the defect after defoaming the screen;

[0042] Step S43: estimating the corresponding area of ​​the defect after debubbling based on the length and width of the defect after debubbling and the curvature area of ​​the abnormal defect after debubbling, and taking the depth of the defect after debubbling, the aspect ratio of the defect after debubbling, and the area of ​​the defect after debubbling as defect geometric features to obtain the geometric features of the defect after debubbling on the flexible screen;

[0043] Step S44: Construct a deep learning model based on the U-Net architecture, and input the geometric features of the defects of the flexible screen after degassing into the trained deep learning model to perform defect classification detection on the abnormal defect curvature area of ​​the screen after degassing, so as to detect the corresponding bubble residue, indentation, scratch and degumming type defect results after the flexible screen is degassing.

[0044] Furthermore, step S44 includes the following steps:

[0045] By building a deep learning model based on the U-Net architecture, the U-Net architecture uses an encoder-decoder part, with a multi-scale feature extraction module in the encoder part to simultaneously capture local details and global context feature information corresponding to the defect. In addition, the attention mechanism is introduced in the decoder part to enhance the detection ability of small defects and low-contrast defect features.

[0046] Obtain a small number of labeled defect samples and a large number of unlabeled normal samples, and use a semi-supervised learning method to input a small number of labeled defect samples and a large number of unlabeled normal samples into a deep learning model based on the U-Net architecture for model training. Generate realistic defect samples through a generative adversarial network to expand the training set, and perform model enhancement based on the expanded training set to generate a trained deep learning model;

[0047] The geometric features of defects after debubbling of the flexible screen are input into the corresponding multi-scale feature extraction module in the trained deep learning model, and defect classification detection is performed on the defect areas detected in the abnormal defect curvature area after debubbling of the screen to detect the corresponding bubble residue, indentation, scratch and degumming type defect results after debubbling of the flexible screen.

[0048] Furthermore, the present invention also provides a system for quickly detecting defects of a flexible screen after degassing, which is used to perform the method for quickly detecting defects of a flexible screen after degassing as described above. The system for quickly detecting defects of a flexible screen after degassing comprises:

[0049] An image complementary registration and fusion module is used to obtain images of the flexible screen after debubbling corresponding to illumination at different spectra and angles, and to perform image complementary registration and fusion on the images of the flexible screen after debubbling corresponding to illumination at different spectra and angles to generate a composite image of the flexible screen after debubbling;

[0050] The screen topography dynamic compensation module is used to reconstruct the structural phase topography of the composite image of the flexible screen after de-bubbling to generate three-dimensional topography data of the flexible screen surface. The module sets marking points on the edge of the flexible screen and obtains the deformation amount of the edge marking position corresponding to the flexible screen based on the marking points. The module then performs dynamic deformation compensation on the three-dimensional topography data of the flexible screen surface based on the deformation amount of the edge marking position corresponding to the flexible screen, thereby obtaining deformation-compensated topography data of the screen surface.

[0051] The abnormal curvature identification module is used to analyze the curvature of each surface point on the deformation compensation topography data of the screen surface to obtain the principal curvature and Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling; based on the principal curvature and Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling, the abnormal curvature area of ​​the corresponding flexible screen after de-bubbling is identified to obtain the abnormal defect curvature area of ​​the screen after de-bubbling;

[0052] The defect classification and detection module is used to analyze the defect geometric features of the abnormal defect curvature area after the screen is debubbled, so as to obtain the defect geometric features of the flexible screen after debubbling; a deep learning model based on the U-Net architecture is constructed, and the defect geometric features of the flexible screen after debubbling are input into the trained deep learning model for defect classification detection, so as to detect the corresponding bubble residue, indentation, scratch and degumming type defect results after the flexible screen is debubbled.

[0053] Beneficial effects of the present invention:

[0054] 1. The method for rapid detection of defects on flexible screens after defoaming proposed in the present invention has the beneficial effect of significantly improving the accuracy and resolution of image data by obtaining corresponding images of the flexible screen after defoaming under different spectra and angles of illumination, and performing image complementary registration and fusion on these images. Due to the particularity of the material and structure of the flexible screen, various tiny defects or unevenness exist on the surface, and these defects are more obvious under different spectra and illumination angles. By collecting images of different spectra and angles, the surface features and defect information of the flexible screen can be captured more comprehensively, and the image complementary registration and fusion can accurately splice images from different perspectives together, eliminate the errors between the perspectives, and thus generate a more complete and high-precision composite image. This composite image can provide more reliable basic data for subsequent three-dimensional morphology reconstruction and defect detection, ensuring that the screen surface defects can be accurately analyzed and identified in subsequent steps. Secondly, the three-dimensional surface morphology data of the flexible screen is restored through precise algorithms, so as to fully understand the morphological changes of the screen surface under different conditions. This process utilizes high-precision phase measurement technology, which can effectively convert two-dimensional image data into three-dimensional data with spatial information, allowing inspection personnel to more intuitively observe the tiny ups and downs and deformations of the flexible screen surface. After obtaining the three-dimensional morphology data, by setting edge marking points and monitoring the deformation of these points, the edge deformation of the screen can be further analyzed. Based on these deformation data, dynamic deformation compensation is performed, which can not only accurately correct the morphological changes caused by external factors such as light and pressure, but also ensure the accuracy of subsequent analysis data. Then, by performing curvature analysis on the morphological data after deformation compensation of the screen surface, the curvature changes in different areas of the flexible screen surface can be effectively revealed. Curvature is a key parameter for measuring the degree of surface bending. By calculating the principal curvature and Gaussian curvature of each point, potential abnormal areas on the screen surface can be identified. High curvature areas are usually associated with defects (such as bubbles, scratches, indentations, etc.), and changes in the principal curvature help detect local deformation of the surface. Based on these curvature data, abnormal defect areas that appear on the screen surface after de-bubbling can be effectively identified. By accurately identifying and marking these abnormal curvature areas, the detection efficiency can be improved, the false detection rate can be reduced, and a key reference can be provided for the subsequent defect classification of screen defects.Finally, through geometric feature analysis, the basic morphological information of defects on the surface of flexible screens can be extracted first. These geometric features provide important input data for subsequent deep learning models. By using a deep learning model with a U-Net architecture for defect classification and detection, the advantages of big data and deep neural networks can be fully utilized. Different types of defects can be automatically identified and classified through the trained model. This method has significant advantages over traditional manual inspection methods, and can greatly improve inspection efficiency and accuracy while reducing the possibility of human error. The classification results of the deep learning model can accurately identify common defect types such as bubble residue, indentation, scratches and degumming. With the accumulation of data and iteration of the model, the accuracy and adaptability of defect detection will continue to improve, pushing the quality assurance and defect control in the flexible screen manufacturing process to a higher level.

[0055] 2. The flexible screen defect rapid detection system after bubble removal proposed in the present invention is composed of an image complementary registration and fusion module, a screen morphology dynamic compensation module, an abnormal curvature recognition module and a defect classification detection module. It can realize the method for rapid detection of defects after bubble removal on any flexible screen described in the present invention, and is used to combine the operations between computer programs running on each module to realize the method for rapid detection of defects after bubble removal on the flexible screen. The internal structures of the system cooperate with each other, which can greatly reduce repetitive work and manpower investment, and can quickly and effectively provide a more accurate and efficient flexible screen defect rapid detection process after bubble removal, thereby simplifying the operating procedures of the flexible screen defect rapid detection system after bubble removal. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0057] Figure 1 This is a schematic flow chart of the steps of the method for quickly detecting defects in a flexible screen after degassing of the present invention;

[0058] Figure 2 for Figure 1 Detailed step flow diagram of step S1;

[0059] Figure 3 for Figure 2 Detailed step flow chart of step S13 in FIG. DETAILED DESCRIPTION

[0060] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0061] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0062] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0063] To achieve this, please refer to Figures 1 to 3 The present invention provides a method for quickly detecting defects of a flexible screen after defoaming, the method comprising the following steps:

[0064] Step S1: obtaining images of the flexible screen after debubbling corresponding to illumination at different spectra and angles, and performing image complementary registration and fusion on the images of the flexible screen after debubbling corresponding to illumination at different spectra and angles to generate a composite image of the flexible screen after debubbling;

[0065] Step S2: reconstructing the structural phase topography of the composite image of the flexible screen after degassing to generate three-dimensional topography data of the flexible screen surface; setting marking points on the edge of the flexible screen and obtaining the deformation of the edge marking position corresponding to the flexible screen according to the marking points; and dynamically compensating the three-dimensional topography data of the flexible screen surface based on the deformation of the edge marking position corresponding to the flexible screen to obtain deformation-compensated topography data of the screen surface;

[0066] Step S3: performing curvature analysis on the deformation compensation topography data of the screen surface to obtain the principal curvature and Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling; based on the principal curvature and Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling, identifying abnormal curvature areas of the corresponding flexible screen after de-bubbling to obtain abnormal defect curvature areas of the screen after de-bubbling;

[0067] Step S4: Analyze the defect geometric features of the abnormal defect curvature area after the screen is debubbled to obtain the defect geometric features of the flexible screen after debubbling; construct a deep learning model based on the U-Net architecture, and input the defect geometric features of the flexible screen after debubbling into the trained deep learning model for defect classification detection to detect the corresponding bubble residue, indentation, scratches and degumming type defect results after the flexible screen is debubbled.

[0068] In the embodiment of the present invention, please refer to Figure 1 FIG. 1 is a schematic flow chart of the steps of the method for quickly detecting defects of a flexible screen after degassing according to the present invention. In this example, the method for quickly detecting defects of a flexible screen after degassing comprises the following steps:

[0069] Step S1: obtaining images of the flexible screen after debubbling corresponding to illumination at different spectra and angles, and performing image complementary registration and fusion on the images of the flexible screen after debubbling corresponding to illumination at different spectra and angles to generate a composite image of the flexible screen after debubbling;

[0070] In the embodiment of the present invention, professional imaging equipment is used to collect images of the finished product after de-bubbling the flexible screen. In terms of spectrum, a light source including visible light (400-700nm), near infrared (700-1100nm) and ultraviolet (200-400nm) bands is used, and the light source parameters are set respectively: the intensity of the visible light source is 5000lux, the power of the near infrared light source is 100mW, and the irradiance of the ultraviolet light source is 10mW / cm 2 , with a resolution of 3840×2160 pixels and an exposure time of 1 / 125 seconds, 10 images were taken each time. In terms of angle, low angle (15°), vertical angle (90°), and high angle (75°) lighting were set, using 20W Ten images were captured using an LED strip light source (low and high angles) and a 50W ring-shaped shadowless light source (vertical angle), also with a resolution of 3840×2160 pixels and an exposure time of 1 / 100-1 / 125 seconds. For complementary image registration and fusion, the SIFT algorithm was first used to extract feature points from the different images. For the visible low-angle image and the near-infrared vertical angle image, 2000 and 1800 feature points were detected, respectively, using a Difference of Gaussian pyramid. The homography matrix was then calculated using the RANSAC algorithm. The near-infrared image was geometrically transformed and aligned with the visible image. A weighted average fusion algorithm was used, with weights of 0.6 for the visible image and 0.4 for the near-infrared image in the overlapping regions. For example, if the RGB values ​​of a pixel in the two images were (200, 150, 100) and (100, 80, 60), respectively, the fused value would be (160, 122, 84). All spectral and angular combinations of images were processed to generate a composite image of the flexible screen after de-bubbling.

[0071] Step S2: reconstructing the structural phase topography of the composite image of the flexible screen after degassing to generate three-dimensional topography data of the flexible screen surface; setting marking points on the edge of the flexible screen and obtaining the deformation of the edge marking position corresponding to the flexible screen according to the marking points; and dynamically compensating the three-dimensional topography data of the flexible screen surface based on the deformation of the edge marking position corresponding to the flexible screen to obtain deformation-compensated topography data of the screen surface;

[0072] In the embodiment of the present invention, the structural phase morphology of the composite image of the flexible screen after de-bubbling is reconstructed, and the image is divided into 64×64 pixel sub-regions by using phase shift interferometry, and three sinusoidal fringe patterns with a phase difference of 2π / 3 are projected into each sub-region. Where A(x,y) is the background light intensity, B(x,y) is the modulated light intensity, f0 is the fringe frequency, is the initial phase, calculated by Get the phase distribution and then use the phase-height conversion formula (λ is the wavelength of the projected light, set to 632.8nm) and converted into height information. The sub-areas are spliced ​​to generate 3D topography data. 100 circular markers with a diameter of 0.2mm are set at equal intervals on the edge of the flexible screen. A high-precision coordinate measuring instrument (accuracy 0.001mm) is used to measure the 3D coordinates of the markers when the screen is not under stress and under stress. and Calculate the displacement, that is The deformation of the edge mark position is obtained. Based on the deformation, the thin plate spline interpolation algorithm is used to dynamically compensate the three-dimensional shape data. For any point (x, y) in the three-dimensional shape data, the thin plate spline function is used (where U(r) = r 2 lnr, (x k ,y k ) is the coordinate of the marking point), and the corresponding relationship before and after the deformation of the point is calculated in combination with the deformation amount of the marking point, and finally the deformation compensation morphology data of the screen surface is obtained.

[0073] Step S3: performing curvature analysis on the deformation compensation topography data of the screen surface to obtain the principal curvature and Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling; based on the principal curvature and Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling, identifying abnormal curvature areas of the corresponding flexible screen after de-bubbling to obtain abnormal defect curvature areas of the screen after de-bubbling;

[0074] In the embodiment of the present invention, curvature analysis is performed on the deformation compensation topography data of the screen surface to calculate the first-order partial derivative by using the five-point difference formula, such as The three-point difference formula calculates the second-order partial derivative, such as And according to Compute the first fundamental form coefficients by The local curvature is obtained, and then the eigenvalue of the shape operator matrix is ​​calculated according to the Weingarten mapping to obtain the principal curvature. The principal curvature is used to calculate the Gaussian curvature K=k1k2 (k1 and k2 are the principal curvatures). The curvature threshold is set to identify the abnormal curvature area: the maximum principal curvature k1>0.3 or the minimum principal curvature k2<-0.1, the Gaussian curvature |K|>0.02, and the curvature division interval is preset: normal area (|k1|≤0.3 and |k2|≤0.1 and |K|≤0.02), slight defect area (0.3<|k1|≤0.5 or 0.1<|k2|≤0.2 or 0.02<|K|≤0.05), and serious defect area (|k1|>0.5 or |k2|>0.2 or |K|>0.05). By adopting the region growing algorithm, the points that meet the threshold are used as seed points, and the adjacent abnormal curvature points are connected to finally determine the abnormal defect curvature area after the screen is defoamed.

[0075] Step S4: Analyze the defect geometric features of the abnormal defect curvature area after the screen is debubbled to obtain the defect geometric features of the flexible screen after debubbling; construct a deep learning model based on the U-Net architecture, and input the defect geometric features of the flexible screen after debubbling into the trained deep learning model for defect classification detection to detect the corresponding bubble residue, indentation, scratches and degumming type defect results after the flexible screen is debubbled.

[0076] In an embodiment of the present invention, the defect geometric characteristics of the abnormal defect curvature area after the screen is defoamed are analyzed, and a high-precision three-dimensional optical measuring instrument is used to measure the length, width and depth of the defect. For example, when measuring a scratch defect, the length is determined to be 25 mm by measuring the coordinates of the two end points, the width is determined to be 1 mm by measuring the coordinates of the edge points on both sides of the widest point, and the depth is determined to be 0.3 mm by the height difference between the bottom of the defect and the normal surface. The aspect ratio R=L / W=25 is calculated. For regular defects (such as rectangular bubble residues), the area is calculated using S=L×W; for irregular defects (such as debonding), the area is divided into 0.1 mm×0.1 mm grids, and the number of completely and partially included grids is counted and converted to obtain the area. The defect depth, aspect ratio and area are combined to form a feature vector, such as [0.3, 25, 25]. , the geometric features of defects in the flexible screen after de-bubbling are obtained, and a deep learning model based on the U-Net architecture is constructed. The encoder contains 4 downsampling modules (two 3×3 convolutional layers + 2×2 maximum pooling layers), and the decoder contains 4 upsampling modules (2×2 deconvolutional layers + two 3×3 convolutional layers). The channel attention mechanism is introduced and trained on a data set containing 2000 samples (800 labeled samples, 200 of each type of defect; 1200 unlabeled samples). The cross-entropy loss function and back-propagation algorithm are used. After training, the defect geometric features are input into the model. After multi-scale feature extraction and encoding and decoding processing, the fully connected layer outputs a probability vector, such as [0.1, 0.05, 0.8, 0.05]. The defect type is judged as a scratch according to the maximum value, and the defect classification detection is completed.

[0077] Further, as an embodiment of the present invention, refer to Figure 2 As shown, Figure 1 Detailed step flow diagram of step S1 in FIG. 1 , in this embodiment, step S1 includes the following steps:

[0078] Step S11: obtaining images of the flexible screen after debubbling under different spectral illumination, including visible light 400-700nm, near infrared 700-1100nm, and ultraviolet 200-400nm band light sources;

[0079] In an embodiment of the present invention, professional imaging equipment is used to photograph the finished product of the flexible screen after debubbling, and images under different spectral illumination are obtained. For the visible light band (400-700nm), a standard white light source is used, the light source intensity is set to 5000lux, the exposure time is 1 / 125 second, and the resolution is set to 3840×2160 pixels. Ten images are captured to cover the entire area of ​​the screen. For the near-infrared band (700-1100nm), the near-infrared light source is enabled and its power is adjusted to 100mW. Similarly, 10 images are captured with a resolution of 3840×2160 pixels and an exposure time of 1 / 125 seconds to ensure that all parts of the screen are clearly captured. In the ultraviolet band (200-400nm), an ultraviolet light source with a wavelength of 365nm is used, and the light source irradiance is controlled to 10mW / cm 2 , 10 images were taken while maintaining the same resolution and exposure time, and all images were stored in RAW format to retain complete spectral information for subsequent processing.

[0080] Step S12: obtaining images of the flexible screen after debubbling under different illumination angles, including low-angle illumination, vertical-angle illumination, and high-angle illumination;

[0081] In an embodiment of the present invention, when obtaining images of the flexible screen after debubbling under different angles of illumination, the flexible screen is fixed on a horizontal workbench. When the light source is illuminated at a low angle, the angle between the light source and the screen surface is set to 15°. A 20W LED strip light source is used to capture 10 images with a resolution of 3840×2160 pixels and an exposure time of 1 / 100 seconds, highlighting the slight undulations and edge details on the screen surface. For vertical angle illumination, a 50W annular shadowless light source is used to vertically illuminate the screen. 10 images are captured at the same resolution and an exposure time of 1 / 125 seconds to obtain a uniformly illuminated image of the entire screen for presenting the macro features of the screen surface. When the light source is illuminated at a high angle, the angle between the light source and the screen surface is adjusted to 75°. A 20W LED strip light source is also used to capture 10 images with an exposure time of 1 / 100 seconds and a resolution of 3840×2160 pixels to highlight the reflection and refraction of the screen surface. Images at different angles together constitute a multi-angle image dataset.

[0082] Step S13: performing screen structure complementary information analysis on the corresponding images of the flexible screen after de-bubbling under different spectra and angles of illumination to obtain screen structure feature complementary information between the images of the flexible screen after de-bubbling;

[0083] In an embodiment of the present invention, the screen structure complementary information analysis is performed on the images of the flexible screen after de-bubbling under different spectra and angles of illumination. Taking the low-angle illumination image of the visible light band and the vertical-angle illumination image of the near-infrared band as examples, the SIFT algorithm is first used to extract the feature points of the two images respectively. For the low-angle illumination image of the visible light, a Gaussian difference pyramid consisting of 5 layers and 3 scales per layer is constructed to detect 2000 feature points and generate a 128-dimensional feature descriptor. The same operation is performed on the near-infrared vertical-angle illumination image to obtain 1800 feature points. Then, the RANSAC algorithm is used to randomly select 4 pairs of matching points from the two groups of feature points to calculate the homography matrix. The near-infrared vertical angle illumination image is geometrically transformed to align it with the visible light low-angle illumination image in space. Next, the image is converted from RGB color space to CIE Lab color space, and the color deviation of the corresponding pixel under different spectra is calculated using the formula By traversing all pixels, if a pixel shows the texture details of the screen surface in the visible light image and the outline of the internal circuit in the near-infrared image, and the color deviation value is greater than the preset threshold (set to 30), then the pixel and the structural information it contains are recorded as complementary information. By performing this operation on images of all different spectral and angle combinations, we finally obtain the complementary information of the screen structural features between the images of different flexible screens after debubbling.

[0084] Step S14: Based on the complementary information of screen structural features between the images of different flexible screens after de-bubbling, the images of the flexible screens after de-bubbling corresponding to the images under different spectra and angles are subjected to image complementary registration fusion optimization to ensure that the structural feature information under different spectra and angles is complementary and fused to the greatest extent, and generate a complementary registration optimization map of the flexible screen after de-bubbling;

[0085] In an embodiment of the present invention, based on the complementary information of the screen structure features obtained previously, the images of the flexible screen after de-bubbling under different spectra and angles are complementary registered and fused for optimization. Taking the visible light high-angle illumination image and the ultraviolet low-angle illumination image as examples, the ultraviolet low-angle illumination image is geometrically transformed according to the calculated homography matrix so that it is precisely aligned with the visible light high-angle illumination image in space. The weighted average fusion algorithm is used for image fusion. For the pixels in the overlapping area, the weight of the visible light image is set to 0.6, and the weight of the ultraviolet image is set to 0.4. Assuming that the RGB value of a certain pixel in the visible light image is (200, 150, 100), in the ultraviolet image, the weight of the visible light image is 0.6, and the weight of the ultraviolet image is 0.4. The RGB values ​​in the external image are (100, 80, 60), and the fused RGB values ​​are calculated as (200×0.6+100×0.4, 150×0.6+80×0.4, 100×0.6+60×0.4)=(160, 122, 84). This registration and fusion operation is performed on images of all different spectral and angular combinations. The weights are adjusted through multiple iterations to ensure that the structural feature information under different spectral and angular illumination is complementary and fused to the greatest extent possible. Finally, a complementary registration optimization image of the flexible screen after de-bubbling is generated, which contains all the complementary information. The resolution of this image is still maintained at 3840×2160 pixels, fully presenting the comprehensive structural characteristics of the screen.

[0086] Step S15: The complementary registration optimization image of the flexible screen after de-bubbling is composited and post-processed to enhance the details through image detail enhancement including high-pass filtering and sharpening processing, and locally correct its edges to eliminate the corresponding seams or incompletely fused image areas to generate a composite image of the flexible screen after de-bubbling.

[0087] In the embodiment of the present invention, the complementary registration optimization image after de-bubbling of the flexible screen is composited and post-processed, and the image details are first enhanced, and high-pass filtering is used to design a high-pass filter matrix of size 5×5. For each pixel in the image, the pixel values ​​in the surrounding 5×5 neighborhood are multiplied by the corresponding elements of the filter matrix and then summed to obtain the filtered pixel value, highlighting the high-frequency detail information in the image, such as tiny scratches on the screen surface and component edges, and then sharpening is performed using the Laplace sharpening algorithm. (where I is the original image, α=0.5 is the sharpening coefficient, The image is processed by the Laplace operator) to enhance the image edges and details, and finally local edge correction is performed. The Poisson fusion algorithm is used to process image seams or incomplete fusion areas. Taking a seam in the image as an example, a 30×30 pixel area around the seam is selected, and the gradient information of the pixels in the area is calculated. According to the pixel values ​​and gradients of the surrounding normal areas, the Poisson equation is solved to generate pixel values ​​that naturally transition with the surrounding areas, eliminating seam traces. This operation is performed on all seams and fusion defect areas of the entire image, and finally a clear and complete composite image of the flexible screen after de-bubbling is generated, providing high-quality image data for subsequent defect detection.

[0088] Further, as an embodiment of the present invention, refer to Figure 3 As shown, Figure 2 Detailed step flow diagram of step S13 in the embodiment, step S13 includes the following steps:

[0089] Step S131: performing light source color deviation analysis on the corresponding de-bubbling images of the flexible screen under different spectral illumination, thereby converting the color space and measuring the color deviation caused by the light source change between the images with different spectra to obtain the light source change color deviation between the images with different spectra;

[0090] In the embodiment of the present invention, the images obtained under red, green, and blue spectrum illumination after the flexible screen is de-bubbled are taken as an example. The resolution of each image is 2560×1440 pixels. First, a color space conversion is performed to convert the image from the RGB color space to the CIE Lab color space. In the process of RGB to CIE Lab conversion, for each pixel point (x, y) in the image, its RGB value (R(x, y), G(x, y), B(x, y)) is first normalized to convert its value range from 0-255 to 0-1, and then the corresponding L is calculated through a series of standard conversion formulas. * (x,y)(brightness), a * (x,y) (color components from green to red), b * (x,y) (color components from blue to yellow) values, when calculating the color deviation between different spectral images, the Euclidean distance formula is used Taking the red spectrum image and the green spectrum image as an example, for the pixel point (100, 100), the L of the point in the red spectrum image is * The value is 50, a * The value is 30, b * The value is 20; the L of the point in the green spectrum image * The value is 45, a * The value is -25, b * If the value is 15, the color deviation of the pixel between the two images is Traverse all pixel points and take the average value to obtain the light source change color deviation between the red spectrum image and the green spectrum image. The same operation is performed on the red and blue, and green and blue spectrum images. Finally, the light source change color deviation between the images after debubbling with different spectra is obtained.

[0091] Step S132: obtaining corresponding flexible screen structure pixel feature points from the corresponding flexible screen after debubbling images under different irradiation angles, and performing screen structure geometric analysis on the corresponding flexible screen after debubbling images under different irradiation angles based on the flexible screen structure pixel feature points to obtain the flexible screen structure geometric relationship between the images after debubbling at different angles;

[0092] In an embodiment of the present invention, the SIFT (Scale Invariant Feature Transform) algorithm is used to obtain pixel feature points of the flexible screen structure after de-bubbling the flexible screen at three angles of 0°, 45°, and 90°. Taking the image illuminated at 0° as an example, a Gaussian difference pyramid is first constructed for the image, and extreme points are detected in different scale spaces to determine the position of the feature points; then the main direction of the feature point is calculated, and the gradient direction histogram is statistically calculated within a certain area with the feature point as the center, and the peak direction is selected as the main direction; finally, a 128-dimensional feature descriptor is generated to describe the gradient information of the area around the feature point. The same method is used to obtain the feature points of the images illuminated at 45° and 90°, and the screen structure geometry analysis is performed based on the obtained feature points. The random sampling consensus (RANSAC) algorithm is used to match the feature points. For the images illuminated at 0° and 45°, 4 pairs of matching points are randomly selected from the two groups of feature points, and the homography matrix H is calculated. The geometric transformation relationship between the two images is described by this matrix. For example, the homography matrix obtained by calculation The points in the image can be illuminated at a 45° angle. (where (x, y) is the coordinate of the 45° image point, and (x', y') is the coordinate corresponding to the 0° image after transformation) to achieve geometric alignment and relationship determination between images. This operation is performed on images of different angle combinations, and finally the geometric relationship of the flexible screen structure between images after debubbling at different angles is obtained.

[0093] Step S133: Based on the color deviation of the light source change between the images after debubbling at different spectra and the geometric relationship of the flexible screen structure between the images after debubbling at different angles, the screen structure complementary information between the corresponding flexible screen images after debubbling is analyzed to obtain the screen structure feature complementary information between the images after debubbling at different flexible screens.

[0094] In an embodiment of the present invention, screen structure complementary information analysis is performed based on the previously obtained light source change color deviation between images after debubbling at different spectra and the flexible screen structure geometric relationship between images after debubbling at different angles. Taking the red spectrum 0° angle illumination image and the blue spectrum 45° angle illumination image as an example, first, according to the geometric relationship of the flexible screen structure, the blue spectrum 45° angle illumination image is geometrically transformed using the homography matrix H calculated previously, so that it is spatially aligned with the red spectrum 0° angle illumination image. Then, the pixel information of corresponding positions in the two images is compared, and the complementary information is analyzed in combination with the color deviation of the light source change. For a certain pixel point (x, y), if the color deviation value of the point is large in the red spectrum image, and the position shows different structural details (such as edge shape differences) after geometric alignment in the blue spectrum image, the information of the position is recorded as complementary information. For example, the color deviation value of the pixel point (200, 200) in the red spectrum image is 40, and after geometric alignment in the blue spectrum image, the position shows the contour details of a component in the screen structure, while this detail is not obvious in the red spectrum image. Then, the pixel point and its related structural detail information are used as part of the screen structure feature complementary information. All pixel points are traversed, and this complementary information is collected and sorted. Finally, the screen structure feature complementary information between the images of different flexible screens after debubbling is obtained for subsequent defect detection and analysis.

[0095] Furthermore, step S2 includes the following steps:

[0096] Step S21: reconstructing the structural phase topography of the composite image of the flexible screen after degassing to generate three-dimensional topography data of the flexible screen surface;

[0097] In the embodiment of the present invention, the structural phase morphology of the composite image of the flexible screen after de-bubbling is reconstructed. The composite image is divided into multiple sub-regions of 64×64 pixels by using phase shift interferometry. For each sub-region, three sinusoidal fringe patterns with different phases are projected onto it.

[0098] Where A(x,y) is the background light intensity, B(x,y) is the modulated light intensity, f0 is the fringe frequency, is the initial phase, calculated by Get the phase distribution and then use the phase-height conversion formula (λ is the wavelength of the projected light, set to 632.8nm). The phase information is converted into height information. After performing the above operation on all sub-areas, the height information of each sub-area is spliced ​​to generate three-dimensional morphological data of the flexible screen surface with a resolution of 3840×2160. This data accurately describes the height value of each point on the screen surface and can intuitively present the convexity and concavity of the screen surface.

[0099] Step S22: setting a marking point on the edge of the flexible screen and obtaining the deformation amount of the edge marking position corresponding to the flexible screen according to the marking point;

[0100] In the embodiment of the present invention, 100 marking points are set at equal intervals on the edge of the flexible screen. The marking points are circular black stickers with a diameter of 0.2 mm. The center of the sticker is used as the marking position. A high-precision coordinate measuring instrument (measuring accuracy of 0.001 mm) is used to measure the three-dimensional coordinates of each marking point when the flexible screen is not under stress. When the flexible screen is subjected to a certain external force, the coordinate measuring machine is used again to measure the three-dimensional coordinates of the marked points Calculate the displacement of each marker point in the x, y, and z directions, that is, Take one of the marked points as an example, if Then the displacement of the point in the x direction Δx 1 = 0.020mm, the displacement of each marking point in three directions is combined into the deformation of the edge marking position In this way, the deformation of the edge mark position corresponding to the flexible screen is obtained, reflecting the deformation of the screen edge.

[0101] Step S23: dynamically capturing the three-dimensional topography data of the flexible screen surface before and after deformation based on the deformation amount of the edge mark position corresponding to the flexible screen, to obtain the topography data of the flexible screen surface before and after deformation;

[0102] In an embodiment of the present invention, the three-dimensional topography data of the flexible screen surface is dynamically captured before and after the topography deformation based on the deformation amount of the edge mark position corresponding to the flexible screen. The three-dimensional topography data before deformation is recorded as M0, which contains the height information h0 (x, y) of 3840×2160 points; the three-dimensional topography data after deformation is recorded as M1, and the corresponding height information is h1 (x, y). According to the deformation amount of the edge mark position, the deformed three-dimensional topography data is spatially transformed by the thin plate spline interpolation algorithm. For any point (x, y) in the three-dimensional topography data, the thin plate spline function is used. (where U(r) = r 2 lnr, (x k ,yk ) is the coordinate of the marking point), and the corresponding relationship of the point before and after deformation is calculated in combination with the deformation amount of the edge marking point. The deformed point is mapped to the undeformed coordinate system. After transformation, the deformed data in the same coordinate system as the undeformed three-dimensional morphology data is obtained. By comparing the two, the morphology data of the flexible screen surface before and after deformation is obtained, which clearly shows the shape change details of the screen surface during the deformation process.

[0103] Step S24: Obtaining the deformation stress source distribution corresponding to each edge deformation mark point based on the deformation amount of the edge mark position corresponding to the flexible screen, and performing a shape stress compensation fitting analysis on the flexible screen surface topography data before and after deformation based on the deformation stress source distribution corresponding to each edge deformation mark point to generate a flexible screen surface topography stress dynamic compensation curve;

[0104] In an embodiment of the present invention, the deformation of the edge mark position corresponding to the flexible screen is used to obtain the deformation stress source distribution corresponding to each edge deformation mark point. The flexible screen is modeled as a three-dimensional solid model using a finite element analysis method and divided into 100,000 tetrahedral units. The deformation of the edge mark point is applied to the model as a boundary condition, and the material properties are set to elastic modulus E = 2 GPa and Poisson's ratio v = 0.3. The finite element equation [K]{δ} = {F} (where [K] is the stiffness matrix, {δ} is the displacement vector, and {F} is the external force vector) is solved to obtain the stress distribution σ(x, y, z) inside the screen. For each edge deformation mark point, the stress data in a certain area around it (a spherical area with a radius of 1 mm) is extracted, and the deformation stress source distribution S corresponding to the mark point is calculated by a weighted average method. i The weight is determined by the distance from the marker point. The closer the distance, the greater the weight. The formula is: where Ω i is the area around the marker point i, w j is the weight, σ j is the stress value of the point in the area. Based on the deformation stress source distribution corresponding to each edge deformation mark point, the shape stress compensation fitting analysis is performed on the shape data of the flexible screen surface before and after deformation. The polynomial fitting method is used, with the stress value as the independent variable and the deformation as the dependent variable, to fit the stress-deformation data of the mark point to obtain the polynomial function f(σ)=a n σ n +a n-1 σ n-1 +…+a1σ+a0, determine the polynomial coefficient a by the least squares method i , so that the sum of square errors between the fitting curve and the data points is minimized, and finally a dynamic compensation curve of the surface morphology stress of the flexible screen is generated to describe the relationship between stress and deformation.

[0105] Step S25: performing dynamic deformation compensation on the three-dimensional surface topography data of the flexible screen based on the dynamic compensation curve of the surface topography stress of the flexible screen to obtain deformation-compensated topography data of the screen surface.

[0106] In an embodiment of the present invention, dynamic deformation compensation is performed on the three-dimensional topography data of the flexible screen surface based on the dynamic compensation curve of the topography stress of the flexible screen. For each point (x, y) in the three-dimensional topography data, the stress value σ(x, y) calculated in the previous step is substituted into the topography stress dynamic compensation curve function f(σ) to obtain the compensation deformation Δh(x, y)=f(σ(x, y)) corresponding to the point. The compensation deformation Δh(x, y) is added to the height value h(x, y) in the original three-dimensional topography data to obtain the compensated height value h compensated (x,y)=h(x,y)+Δh(x,y), perform the above operation for all points, update the entire three-dimensional morphology data, and obtain the screen surface deformation compensation morphology data. This data eliminates the deformation caused by stress and restores a more realistic surface morphology of the flexible screen, providing a reliable data basis for subsequent accurate detection of screen defects.

[0107] Furthermore, step S21 includes the following steps:

[0108] A sinusoidal fringe pattern is projected onto the composite image after de-bubbling the flexible screen to generate a sinusoidal fringe projection image of the flexible screen structure;

[0109] In the embodiment of the present invention, a high-precision projector is used to project a sinusoidal fringe pattern onto the composite image after de-bubbling on the flexible screen. The projection fringe frequency is set to 5 per millimeter and the fringe period T = 0.2 mm. The intensity distribution formula of the sinusoidal fringe pattern projected by the projector is: The background light intensity A = 128 (8-bit grayscale image median), the modulated light intensity B = 64, and the initial phase Taking the composite image of a flexible screen with a size of 3840×2160 pixels after debubbling as an example, the projector evenly covers the entire screen surface with a sinusoidal stripe pattern to generate a sinusoidal stripe projection image of the flexible screen structure. During the projection process, the distance between the projector and the screen is fixed at 500 mm, and the projection angle is perpendicular to the screen surface to ensure the uniformity and accuracy of the stripe pattern on the screen, providing a stable image basis for subsequent phase calculations.

[0110] Preferably, the phase deformation jump corresponding to each sinusoidal fringe in the fringe image is calculated by projecting the sinusoidal fringe image of the flexible screen structure, and phase unwrapping processing is performed according to the phase deformation jump corresponding to each sinusoidal fringe in the fringe image to obtain two-dimensional phase information of the deformation of the flexible screen surface structure;

[0111] In the embodiment of the present invention, a phase analysis is performed on the sinusoidal stripe projection image of the flexible screen structure. Taking one horizontal sinusoidal stripe as an example, a series of equally spaced sampling points (with a spacing of 1 pixel) are selected along the stripe direction to obtain the grayscale value I of each sampling point. i , for example, the phase can be calculated by a four-step phase shift algorithm, projecting the phase difference as The four sinusoidal fringe patterns I1(x,y), I2(x,y), I3(x,y), I4(x,y), the phase calculation formula is Since the phase calculation results have the periodicity of (-π,π], phase deformation jump (phase wrapping) phenomenon will occur. The phase unwrapping algorithm guided by the quality map is adopted. First, the phase quality map of each pixel is calculated. The calculation formula of the quality map is: in and are the phase gradients in the x and y directions, ∈ = 10 -6 To avoid a tiny constant with a denominator of zero, the phase is gradually expanded from the high-quality area to the low-quality area according to the quality map, eliminating phase jumps, and finally obtaining continuous two-dimensional phase information of the flexible screen surface structure deformation, which accurately reflects the deformation of the screen surface structure.

[0112] Preferably, three-dimensional morphology reconstruction is performed on each point in the corresponding flexible shielding surface structure in the composite image of the flexible screen after debubbling based on the two-dimensional phase information of the deformation of the flexible screen surface structure to generate three-dimensional morphology data of the flexible screen surface.

[0113] In the embodiment of the present invention, the three-dimensional morphology is reconstructed based on the two-dimensional phase information of the deformation of the flexible screen surface structure. The geometric parameters of the projection system are known: the baseline distance b between the projector optical center and the camera optical center is 50 mm, the angle θ between the projector optical axis and the camera optical axis is 15°, and the working distance L is 500 mm. For each point (x, y) on the flexible screen surface structure, the corresponding phase value is According to the principle of triangulation, the height calculation formula is: Where f0 is the fringe frequency (f0 = 5 fringe / mm). Taking the pixel (0, 0) in the upper left corner of the screen as an example, its phase value is Substituting into the formula we can get The above calculation is performed on all pixel points in the composite image of the flexible screen after degassing, and finally three-dimensional topography data containing the height information of each point is generated. The data resolution is consistent with the original image at 3840×2160, realizing accurate reconstruction of the three-dimensional topography of the flexible screen surface and providing detailed three-dimensional data support for subsequent defect detection.

[0114] Furthermore, the step S22 of obtaining the deformation of the edge mark position corresponding to the flexible screen according to the mark point includes the following steps:

[0115] Obtain the corresponding surface deformation stress-strain distribution on the edge of the flexible screen according to the marked points;

[0116] In an embodiment of the present invention, 100 circular marking points with a diameter of 0.2 mm are set at equal intervals on the edge of the flexible screen, and digital image correlation (DIC) technology is used to obtain the surface deformation stress-strain distribution corresponding to the marking points on the edge of the flexible screen. When the flexible screen is in an initial state after debubbling, a high-resolution industrial camera is used to capture a reference image to record the initial state of the marking points and the surrounding area. When the flexible screen is subjected to external stress (such as bending, stretching, etc.), the deformed image is captured again, and the images before and after deformation are imported into DIC analysis software. The software automatically identifies the features of the marking points and the surrounding areas, calculates the grayscale changes of the pixels in the image, and uses a correlation algorithm to match the points before and after deformation to obtain the displacement information of each point. Based on the displacement information, a strain calculation method is used to calculate the positive strain ε in the x and y directions for a square area with a side length of 1 mm around the marking point. xx , ε yy and shear strain γ xy , and according to Hooke's law, for isotropic elastic materials, it is known that the elastic modulus E = 2GPa and Poisson's ratio v = 0.3, through the formula Calculate the stress components to obtain the surface deformation stress-strain distribution corresponding to each marker point on the edge of the flexible screen. For example, for marker point 1, its positive strain in the x direction is ε xx =0.001, positive strain in y direction ε yy =0.0005, shear strain γ xy =0.0002, substituting into the formula we can get σ xx ≈30.3MPa,σ yy ≈20.2MPa, τ xy ≈7.7MPa.

[0117] Preferably, based on the surface deformation stress-strain distribution corresponding to each marking point on the edge of the flexible screen, displacement increments are calculated on the corresponding marking points to obtain the displacement increments and displacement directions corresponding to each marking point on the edge of the flexible screen;

[0118] In an embodiment of the present invention, based on the surface deformation stress-strain distribution corresponding to each marking point on the edge of the flexible screen obtained previously, the displacement increment of the corresponding marking point is calculated using the finite element method, and the flexible screen is modeled as a three-dimensional finite element model containing 10,000 tetrahedral units. The stress-strain distribution of each marking point is applied to the corresponding node in the model as a boundary condition, and according to the equilibrium equation and constitutive equation in elastic mechanics, the displacement solution of each node is obtained by solving the finite element equation group [K]{δ}={F} (where [K] is the overall stiffness matrix, {δ} is the node displacement vector, and {F} is the node force vector). For each marking point, its displacement increment The displacement increment Δd in the x, y, and z directions is x , Δd y , Δd z Composition, that is Taking the marked point 2 as an example, the displacement increment Δd in the x direction is obtained by solving the finite element equation. x =0.01mm, y-direction displacement increment Δd y =0.005mm, z-direction displacement increment Δd z =0.002mm, then the displacement increment of mark point 2 At the same time, the displacement direction can be determined by calculating the angle between the displacement vector and the coordinate axis. For the marker point 2, the angle between its displacement direction and the x-axis is Angle with y axis Angle with z axis Thus, the displacement increment and displacement direction corresponding to each marked point on the edge of the flexible screen are obtained.

[0119] Preferably, the position deformation prediction of the displacement increment corresponding to each mark point on the edge of the flexible screen is performed based on the displacement direction corresponding to each mark point on the edge of the flexible screen to obtain the position deformation amount of the edge mark corresponding to the flexible screen.

[0120] In the embodiment of the present invention, based on the displacement direction corresponding to each mark point on the edge of the flexible screen, the linear extrapolation method is used to predict the position deformation of the displacement increment corresponding to each mark point on the edge of the flexible screen. For each mark point, it is assumed that its displacement direction remains unchanged during the subsequent deformation process. According to the obtained displacement increment and displacement direction, combined with parameters such as time or loading steps, extrapolation is performed. Taking mark point 3 as an example, its displacement increment is known. The angle θ between the displacement direction and the x-axis x3 =30°, angle θ with the y-axis y3 =60°, angle θ with the z-axis z3 =73.9°, if the subsequent loading is expected to increase the displacement increment by a factor of 2, the predicted displacement increment For all the marking points on the edge of the flexible screen, the above prediction operation is performed, and the predicted displacement increment is combined with the initial position to obtain the predicted position coordinates of each marking point. For example, the initial coordinates of marking point 3 are (x0, y0, z0) = (10mm, 10mm, 10mm), and the predicted coordinates are (x1, y1, z1) = (10.03mm, 10.016mm, 10.006mm). By calculating the difference between the predicted coordinates and the initial coordinates, the deformation of the edge marking position corresponding to the flexible screen is finally obtained. For example, the deformation of the edge marking position of marking point 3 is This completes the position deformation prediction and deformation calculation of all marked points.

[0121] Furthermore, step S3 includes the following steps:

[0122] Step S31: performing high-order polynomial fitting on the screen surface deformation compensation topography data, so as to correct points that deviate from the normal curved surface due to environmental changes or screen quality fluctuations by using a local weighted regression model, thereby obtaining the screen surface abnormal shape calibration topography data;

[0123] In the embodiment of the present invention, a high-order polynomial fitting is performed on the deformation compensation topography data of the screen surface, and a local weighted regression model (LOESS) is used to correct the points that deviate from the normal surface morphology. The screen surface is divided into 50×50 grids, each grid contains 768×432 data points, and for each point in each grid, a cubic polynomial z=a0+a1x+a2y+a3x is used. 2 +a4xy+a5y 2 +a6x 3 +a7x 2 y+a8xy 2 +a9y 3 As the basic model, in the local weighted regression process, for the point with the center coordinate (x0, y0), its weight function is (where d is the influence radius, which is set to half the length of the grid side). Taking the point at the center of the grid (25, 25) as an example, its influence radius d = 25. For the adjacent point with coordinates (26, 26), the distance Substituting the weight function into w(26,26)≈0.999, the weighted error sum of squares is solved by the least squares method. (where z i is the actual height value, f(x i ,y i) is the coefficients a0, a1, …, a9 when the polynomial prediction value is minimized. For points that deviate from the normal surface morphology, such as when the actual height of a point is 10.5 mm, and the fitting surface prediction value is 9.8 mm, and the residual |10.5-9.8|=0.7 mm is greater than the threshold (set to 0.5 mm), the fitting value is used for correction, and finally the abnormal morphology calibration data of the screen surface is obtained, which is more consistent with the normal surface morphology of the flexible screen.

[0124] Step S32: using a numerical differentiation method to perform a local curvature analysis on each point on the flexible screen surface in the screen surface abnormal morphology calibration topography data, to obtain the local curvature corresponding to each point on the flexible screen surface after de-bubbling;

[0125] In an embodiment of the present invention, a local curvature analysis is performed on each point on the flexible screen surface in the screen surface abnormal morphology calibration data by using a numerical differentiation method. For a point (x, y) on the screen surface, its height value is z(x, y), and the first-order partial derivative is calculated using the five-point difference formula: Where Δx=Δy=0.1mm is the sampling interval. Taking point (10, 10) as an example, the height values ​​of the surrounding points are z(12, 10)=9.9mm, z(11, 10)=9.85mm, z(9, 10)=9.75mm, z(8, 10)=9.7mm, and so on. Substituting into the formula, we can get And calculate the second-order partial derivatives by using the three-point difference formula: And according to the differential geometry formula, the local curvature (in The local curvature corresponding to each point on the surface of the flexible screen after debubbling is calculated, and finally the local curvature corresponding to each point on the surface of the flexible screen after debubbling is obtained.

[0126] Step S33: obtaining the corresponding normal direction of the flexible screen surface by calibrating the topography data of the abnormal morphology of the screen surface, and performing principal curvature calculation on the local curvature corresponding to each point on the surface of the flexible screen after de-bubbling based on the normal direction of the flexible screen surface, so as to obtain the principal curvature corresponding to each point on the surface of the flexible screen after de-bubbling;

[0127] In the embodiment of the present invention, the corresponding normal direction of the flexible screen surface is obtained by calibrating the topographic data of the abnormal shape of the screen surface, and the principal curvature of the local curvature corresponding to each point on the surface of the flexible screen after de-bubbling is calculated based on the normal direction of the flexible screen surface. For a point (x, y) on the screen surface, its surface normal direction vector Taking the point (20, 20) as an example, it is known that Then the normal direction vector And according to the Weingarten map, the principal curvature is the shape operator matrix The eigenvalues ​​of ). For the point (20, 20), it is known that Substituting into the equation, we get L = 0.107, M = 0.036, N = 0.054, and the shape operator matrix Solve the characteristic equation det(S-kI)=0 (where I is the identity matrix), that is The expansion is (0.098-k)(0.183-k)-2×0.033=0, and the solution is the eigenvalues ​​k1≈0.35, k2≈-0.07, that is, the principal curvatures corresponding to the point (20, 20) are 0.35 and -0.07, thus obtaining the principal curvatures corresponding to each point on the surface of the flexible screen after debubbling.

[0128] Step S34: performing maximum value decomposition on the principal curvature corresponding to each point on the surface of the flexible screen after de-bubbling by combining a preset Jacobi matrix to obtain the maximum and minimum principal curvatures corresponding to each point on the surface of the flexible screen after de-bubbling, and performing Gaussian curvature calculation based on the maximum and minimum principal curvatures corresponding to each point on the surface of the flexible screen after de-bubbling to obtain the Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling;

[0129] In the embodiment of the present invention, the principal curvature corresponding to each point on the surface of the flexible screen after degassing is decomposed by combining a preset Jacobi matrix to obtain the maximum and minimum principal curvatures corresponding to each point on the surface of the flexible screen after degassing. The preset Jacobi matrix is (where θ is the rotation angle), for a point (x, y) on the screen surface, its shape operator matrix S is transformed by the similarity transformation J T SJ performs diagonalization, where (E, F, G are the first basic form coefficients.) Taking the point (30, 30) as an example, if E = 1.1, F = 0.15, and G = 1.05, then The corresponding Jacobi matrix Perform similarity transformation J on the shape operator matrix S T SJ, get the diagonal matrix Where k1 and k2 are the maximum and minimum principal curvatures, respectively. For point (30, 30), k1 = 0.28 and k2 = -0.05 are calculated. Based on the maximum and minimum principal curvatures corresponding to each point on the flexible screen's de-bubbled surface, the Gaussian curvature is calculated as K = k1k2. For point (30, 30), the Gaussian curvature K = 0.28 × (-0.05) = -0.014, thus obtaining the Gaussian curvature corresponding to each point on the flexible screen's de-bubbled surface.

[0130] Step S35: Based on the principal curvature and Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling, the abnormal curvature area of ​​the corresponding flexible screen after de-bubbling is identified, so as to calculate the corresponding curvature peak value according to the principal curvature and Gaussian curvature, and divide the corresponding abnormal defect area range according to the judgment result between the preset curvature division interval and the curvature peak value, so as to obtain the abnormal defect curvature area of ​​the screen after de-bubbling.

[0131] In the embodiment of the present invention, the abnormal curvature area of ​​the flexible screen after de-bubbling is identified based on the principal curvature and Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling, and the curvature peak judgment threshold is set as: the maximum principal curvature k1>0.3 or the minimum principal curvature k2<-0.1, the Gaussian curvature |K|>0.02, and the curvature division intervals are preset: normal area (|k1|≤0.3 and |k2|≤0.1 and |K|≤0.02), slight defect area domain (0.3<|k1|≤0.5 or 0.1<|k2|≤0.2 or 0.02<|K|≤0.05), serious defect area (|k1|>0.5 or |k2|>0.2 or |K|>0.05), taking the point (40, 40) on the screen surface as an example, its maximum principal curvature k1=0.4, the minimum principal curvature k2=-0.15, and the Gaussian curvature K=0.4×(-0.15)=-0.06. Since |k1|=0.4 in In the interval of 0.3<|k1|≤0.5, |k2|=0.15 in the interval of 0.1<|k2|≤0.2, |K|=0.06>0.05, which meets the judgment conditions of the severe defect area. Therefore, point (40, 40) is divided into the severe defect area. For the entire screen surface, judgment is performed point by point, and the region growing algorithm is used to connect adjacent abnormal curvature points. Point (40, 40) is used as the seed point, and the points in its 8-neighborhood are checked. If k1=0.45, k2=-0.18, K=-0.081 of a certain point (41, 40) also meets the severe defect area conditions, then the point is added to the severe defect area. This process is repeated until no adjacent points meeting the conditions can be found. Finally, a continuous severe defect area range is obtained. This operation is performed on all abnormal curvature points to divide the corresponding abnormal defect area range, and the abnormal defect curvature area of ​​the screen after bubble removal is obtained, providing clear location and range information for subsequent defect repair.

[0132] Furthermore, step S4 includes the following steps:

[0133] Step S41: obtaining the corresponding screen defect length, screen defect width, and screen defect depth after defoaming through the abnormal defect curvature area after defoaming;

[0134] In the embodiment of the present invention, a flexible screen with a size of 100 mm × 200 mm is taken as an example. A high-precision three-dimensional optical measuring instrument is used to measure the abnormal defect curvature area after debubbling. The flexible screen is placed on the measuring instrument workbench. The abnormal defect curvature area is scanned in all directions through the scanning function of the measuring instrument. For the measurement of the defect length, the measuring instrument collects data along the longest direction of the defect. The length is determined by the coordinate difference between the two end points of the defect in the measurement coordinate system. Assuming that the coordinates of one end point of a scratch defect are (10 mm, 20 mm, 0 mm) and the coordinates of the other end point are (35 mm, 20 mm, 0 mm), the length of the defect is The defect width is measured by selecting the widest part of the defect, collecting data perpendicular to the length direction, and calculating the coordinate difference of the edge points on both sides. If the scratch is widest at a certain point, and the coordinates of the edge points on both sides are (20mm, 19.5mm, 0mm) and (20mm, 20.5mm, 0mm) respectively, then the defect width is The defect depth is measured by taking the vertical height difference (z-axis) between the bottom of the defect and the surrounding normal screen surface. If the z-coordinate of the normal screen surface surrounding the defect is 0mm and the z-coordinate of a point at the bottom of the defect is -0.3mm, the defect depth at that point is 0.3mm. The average of the depth values ​​at multiple measurement points yields a depth of 0.28mm for the defect, thus determining the length, width, and depth of the defect after degassing.

[0135] Step S42: calculating the aspect ratio of the defect after defoaming the screen according to the length and width of the defect after defoaming the screen, to obtain the aspect ratio of the defect after defoaming the screen;

[0136] In the embodiment of the present invention, the aspect ratio is calculated based on the length and width of the screen after bubble removal. Taking a certain indentation defect as an example, its length is measured to be 15mm and its width is 3mm. According to the aspect ratio calculation formula (Where R is the aspect ratio, L is the defect length, and W is the defect width), substituting L=15mm and W=3mm into the formula, we can get the aspect ratio of the indentation defect R=15 / 3=5. Through this calculation, we can finally get the aspect ratio of the defect after the screen is defoamed. This ratio can be used as an important geometric feature basis for judging the defect type.

[0137] Step S43: estimating the corresponding area of ​​the defect after debubbling based on the length and width of the defect after debubbling and the curvature area of ​​the abnormal defect after debubbling, and taking the depth of the defect after debubbling, the aspect ratio of the defect after debubbling, and the area of ​​the defect after debubbling as defect geometric features to obtain the geometric features of the defect after debubbling on the flexible screen;

[0138] In the embodiment of the present invention, the defect area is estimated based on the previously obtained length and width of the screen defect after degassing, combined with the curvature area of ​​the abnormal defect after degassing. For defects with relatively regular shapes, such as rectangular bubble residual defects, the length is 8 mm and the width is 4 mm. The rectangular area formula S = L × W (where S is the area, L is the length, and W is the width) is directly used to obtain its area S = 8 × 4 = 32 mm 2 For irregularly shaped defects, such as debonding defects, a gridding method is used for estimation. The abnormal defect curvature area is divided into square grids with a side length of 0.1 mm. The number of grids m1 that are completely contained in the defect area and the number of grids m2 that are partially contained in the defect area are counted. For partially contained grids, the area is converted according to the proportion of their contained area. Assuming n1 = 200, n2 = 50, and the average conversion ratio of each partially contained grid is 0.6, the defect area S = n1 × (0.1 × 0.1) + n2 × 0.1 × 0.1 × 0.6 = 2 + 0.3 = 2.3 mm 2 The depth, aspect ratio and area of ​​the defect after screen defoaming are used as the defect geometric features. For example, a defect has a depth of 0.2mm, an aspect ratio of 4 and an area of ​​20mm. 2 , forming a feature vector [0.2, 4, 20], and finally obtaining the defect geometric features of the flexible screen after debubbling, providing data support for subsequent defect classification and detection.

[0139] Step S44: Construct a deep learning model based on the U-Net architecture, and input the geometric features of the defects of the flexible screen after degassing into the trained deep learning model to perform defect classification detection on the abnormal defect curvature area of ​​the screen after degassing, so as to detect the corresponding bubble residue, indentation, scratch and degumming type defect results after the flexible screen is degassing.

[0140] In an embodiment of the present invention, a deep learning model based on the U-Net architecture is constructed. The encoder part of the architecture includes four downsampling modules, each module consisting of two 3×3 convolutional layers and a 2×2 maximum pooling layer (with a step size of 2), which is used to extract features and downsample the input data. Taking an image with an input size of 128×128 pixels (the image is used to represent the geometric features of the defect and encode information such as the defect depth, aspect ratio and area into different channels or pixel values ​​of the image) as an example, in the first downsampling module, the number of channels of the image increases from 3 to 32 after passing through two convolutional layers, and then the size is changed to 64×64 pixels after passing through the maximum pooling layer. The decoder part also includes four upsampling modules, each module consisting of a 2×2 deconvolution layer (with a step size of 2) and two 3×3 convolutional layers, which are used to upsample and restore the features extracted by the encoder. A channel attention mechanism is introduced in the decoder to enhance the network's response to key defect features by calculating the importance of each channel, and the model is trained using a data set containing 2000 samples, where the labeled samples 800 defect samples (including four categories of defect samples: bubble residue, indentation, scratch, and debonding, 200 in each category) and 1200 unlabeled samples were trained. The cross-entropy loss function was used to calculate the error of the labeled samples, and the network parameters were adjusted through the back-propagation algorithm. After training, the previously obtained geometric features of the flexible screen defects after de-bubbling (such as images encoded in the form of feature vectors) were input into the trained deep learning model. The model's multi-scale feature extraction module processed the input data and extracted feature information at different scales. For example, for bubble residue defects, the model extracted their circular or elliptical geometric features; for scratch defects, their slender shape features were extracted. After processing by the encoder and decoder, the model output the defect classification result through the fully connected layer. Taking a certain defect geometric feature input as an example, the model outputted a probability vector of [0.1, 0.05, 0.8, 0.05], corresponding to the predicted probabilities of bubble residue, indentation, scratch, and debonding, respectively. The defect was judged to be a scratch type based on the maximum probability, thereby achieving defect classification detection in the abnormal defect curvature area after screen de-bubbling.

[0141] Furthermore, step S44 includes the following steps:

[0142] By building a deep learning model based on the U-Net architecture, the U-Net architecture uses an encoder-decoder part, with a multi-scale feature extraction module in the encoder part to simultaneously capture local details and global context feature information corresponding to the defect. In addition, the attention mechanism is introduced in the decoder part to enhance the detection ability of small defects and low-contrast defect features.

[0143] In an embodiment of the present invention, a deep learning model based on the U-Net architecture is constructed for post-debubble defect detection of flexible screens. The encoder part of the U-Net architecture includes 5 downsampling modules, each of which consists of two 3×3 convolutional layers and a 2×2 maximum pooling layer (with a step size of 2). Taking an image with an input size of 256×256 pixels as an example, in the first downsampling module, the image first passes through two convolutional layers to increase the number of channels from 3 to 64, and then passes through the maximum pooling layer to make the image size 128×128 pixels. The convolution layer of each downsampling module adopts the ReLU activation function to enhance the nonlinear expression ability of the network. A multi-scale feature extraction module is adopted in the encoder part, which is specifically implemented by three parallel convolutional layers with convolution kernels of different sizes, namely 3×3, 5×5 and 7×7. For the same input Feature map, 3×3 convolution kernel captures local detail features, 5×5 convolution kernel obtains medium-scale features, and 7×7 convolution kernel extracts global context features. The outputs of the three convolution layers are then spliced ​​in the channel dimension to obtain a feature map containing multi-scale information. The decoder part also contains 5 upsampling modules, each module consists of a 2×2 deconvolution layer (step size is 2) and two 3×3 convolution layers. The spatial attention mechanism is introduced in the decoder to enhance the detection ability of small defects and weak contrast defect features. For the feature map output by the decoder, two different feature descriptions are first obtained by average pooling and maximum pooling respectively, and then they are spliced ​​and passed through a 7×7 convolution layer to output an attention weight map of the same size as the original feature map. Finally, the attention weight map is multiplied by the corresponding element of the original feature map to highlight the defect area.

[0144] Preferably, a small number of labeled defect samples and a large number of unlabeled normal samples are obtained, and a semi-supervised learning method is used to input the small number of labeled defect samples and the large number of unlabeled normal samples into a deep learning model corresponding to the U-Net architecture for model training, and realistic defect samples are generated by a generative adversarial network to expand the training set, and the model is enhanced based on the expanded training set to generate a trained deep learning model;

[0145] In this embodiment of the present invention, 200 labeled defective samples of flexible screens after bubble removal and 2000 unlabeled normal samples are obtained to train the model using a semi-supervised learning method. The labeled samples and unlabeled samples are input into a deep learning model based on the U-Net architecture in batches. For the labeled samples, the cross entropy loss function is used to calculate the error between the predicted result and the true label. The formula is: where N labeled is the number of labeled samples, C is the number of defect categories (here C = 4, including bubble residue, indentation, scratch, and debonding), y ic is the true label, p icTo predict the probability, for unlabeled samples, the consistency regularization method is used. By adding different disturbances (such as Gaussian noise and random rotation) to the unlabeled samples, the model is required to output similar prediction results for the samples before and after the disturbance, and the loss of the unlabeled samples is calculated. where N unlabeled is the number of unlabeled samples, x i is the original unlabeled sample, is to add the perturbed samples, f is the model prediction function, and at the same time, the generative adversarial network (GAN) is used to expand the training set. The generator consists of multiple transposed convolutional layers, inputs a 100-dimensional random noise vector, and outputs a defect image with a size of 256×256 pixels. The discriminator consists of multiple convolutional layers and fully connected layers, which are used to determine whether the input image is a real defect sample or a generated sample. Through adversarial training, the generator continuously generates realistic defect samples, such as simulated bubble residue images of different sizes and shapes, scratch images of various depths and lengths, etc. The 500 generated defect samples are added to the training set to obtain the expanded training set. The model is further trained based on the expanded training set, and the network parameters are adjusted so that the model can achieve good performance on both labeled and unlabeled samples, and finally a trained deep learning model is generated.

[0146] Preferably, the geometric features of defects after degassing the flexible screen are input into the corresponding multi-scale feature extraction module in the trained deep learning model, and defect classification detection is performed on the defect area detected in the abnormal defect curvature area after degassing the screen, so as to detect the corresponding bubble residue, indentation, scratch and degumming type defect results after degassing the flexible screen.

[0147] In an embodiment of the present invention, the previously obtained geometric features of the defects of the flexible screen after debubbling (such as the shape, size, curvature change and other information within the abnormal defect curvature area, represented in the form of a 256×256 pixel image) are input into the multi-scale feature extraction module in the trained deep learning model. The multi-scale feature extraction module extracts the local details, medium-scale and global context features of the defects through 3×3, 5×5 and 7×7 convolution kernels respectively. For example, for bubble residual defects, the 3×3 convolution kernel can capture the subtle irregularities of the bubble edge, and the 7×7 convolution kernel can obtain the position and approximate shape of the entire bubble in the screen. After downsampling and feature extraction by the encoder, the feature map is passed to the decoder part. In the detector, the spatial attention mechanism highlights the characteristics of small defects and weak contrast defects, enabling the network to more accurately identify these easily overlooked defects. Finally, the model outputs the defect classification results through the fully connected layer, and divides the defect area into four types: bubble residue, indentation, scratch and degumming. Taking a certain defect area as an example, the probability vector output by the model is [0.8, 0.1, 0.05, 0.05], indicating that the probability of the defect area being bubble residue is 80%, and the probabilities of being indentation, scratch and degumming are 10%, 5% and 5% respectively. According to the maximum probability, the defect area is judged to be the bubble residue type, thereby realizing accurate defect classification detection of the defect area detected in the abnormal defect curvature area after the screen is defoamed.

[0148] Furthermore, the present invention also provides a system for quickly detecting defects of a flexible screen after degassing, which is used to perform the method for quickly detecting defects of a flexible screen after degassing as described above. The system for quickly detecting defects of a flexible screen after degassing comprises:

[0149] An image complementary registration and fusion module is used to obtain images of the flexible screen after debubbling corresponding to illumination at different spectra and angles, and to perform image complementary registration and fusion on the images of the flexible screen after debubbling corresponding to illumination at different spectra and angles to generate a composite image of the flexible screen after debubbling;

[0150] The screen topography dynamic compensation module is used to reconstruct the structural phase topography of the composite image of the flexible screen after de-bubbling to generate three-dimensional topography data of the flexible screen surface. The module sets marking points on the edge of the flexible screen and obtains the deformation amount of the edge marking position corresponding to the flexible screen based on the marking points. The module then performs dynamic deformation compensation on the three-dimensional topography data of the flexible screen surface based on the deformation amount of the edge marking position corresponding to the flexible screen, thereby obtaining deformation-compensated topography data of the screen surface.

[0151] The abnormal curvature identification module is used to analyze the curvature of each surface point on the deformation compensation topography data of the screen surface to obtain the principal curvature and Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling; based on the principal curvature and Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling, the abnormal curvature area of ​​the corresponding flexible screen after de-bubbling is identified to obtain the abnormal defect curvature area of ​​the screen after de-bubbling;

[0152] The defect classification and detection module is used to analyze the defect geometric features of the abnormal defect curvature area after the screen is debubbled, so as to obtain the defect geometric features of the flexible screen after debubbling; a deep learning model based on the U-Net architecture is constructed, and the defect geometric features of the flexible screen after debubbling are input into the trained deep learning model for defect classification detection, so as to detect the corresponding bubble residue, indentation, scratch and degumming type defect results after the flexible screen is debubbled.

[0153] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A method for quickly detecting defects in flexible screens after de-bubbling, characterized in that: The following steps are involved: Step S1: obtaining images of the flexible screen after debubbling corresponding to illumination at different spectra and angles, and performing image complementary registration and fusion on the images of the flexible screen after debubbling corresponding to illumination at different spectra and angles to generate a composite image of the flexible screen after debubbling; Step S2: reconstructing the structural phase topography of the composite image of the flexible screen after degassing to generate three-dimensional topography data of the flexible screen surface; by setting a marking point on the edge of the flexible screen and obtaining the deformation amount of the edge marking position corresponding to the flexible screen according to the marking point; Dynamically compensate the three-dimensional surface shape data of the flexible screen based on the deformation of the edge mark position corresponding to the flexible screen to obtain the deformation-compensated shape data of the screen surface; Step S3: performing curvature analysis on the deformation compensation topography data of the screen surface to obtain the principal curvature and Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling; based on the principal curvature and Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling, identifying abnormal curvature areas of the corresponding flexible screen after de-bubbling to obtain abnormal defect curvature areas of the screen after de-bubbling; Step S4: performing defect geometric feature analysis on the abnormal defect curvature area of ​​the screen after debubbling to obtain the defect geometric features of the flexible screen after debubbling; A deep learning model based on the U-Net architecture is constructed, and the geometric features of the defects after degassing the flexible screen are input into the trained deep learning model for defect classification detection to detect the corresponding bubble residue, indentation, scratch and degumming type defect results after degassing the flexible screen.

2. The method for rapid detection of defects in flexible screens after debubbling according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: obtaining images of the flexible screen after debubbling under different spectral illumination, including visible light 400-700nm, near infrared 700-1100nm, and ultraviolet 200-400nm band light sources; Step S12: obtaining images of the flexible screen after debubbling under different illumination angles, including low-angle illumination, vertical-angle illumination, and high-angle illumination; Step S13: performing screen structure complementary information analysis on the corresponding images of the flexible screen after de-bubbling under different spectra and angles of illumination to obtain screen structure feature complementary information between the images of the flexible screen after de-bubbling; Step S14: Based on the complementary information of screen structural features between the images of different flexible screens after de-bubbling, the images of the flexible screens after de-bubbling corresponding to the images under different spectra and angles are subjected to image complementary registration fusion optimization to ensure that the structural feature information under different spectra and angles is complementary and fused to the greatest extent, and generate a complementary registration optimization map of the flexible screen after de-bubbling; Step S15: The complementary registration optimization image of the flexible screen after de-bubbling is composited and post-processed to enhance the details through image detail enhancement including high-pass filtering and sharpening processing, and locally correct its edges to eliminate the corresponding seams or incompletely fused image areas to generate a composite image of the flexible screen after de-bubbling.

3. The method for rapid detection of defects in flexible screens after degassing according to claim 2, characterized in that: Step S13 includes the following steps: Step S131: performing light source color deviation analysis on the corresponding de-bubbling images of the flexible screen under different spectral illumination, thereby converting the color space and measuring the color deviation caused by the light source change between the images with different spectra to obtain the light source change color deviation between the images with different spectra; Step S132: obtaining corresponding flexible screen structure pixel feature points from the corresponding flexible screen after debubbling images under different irradiation angles, and performing screen structure geometric analysis on the corresponding flexible screen after debubbling images under different irradiation angles based on the flexible screen structure pixel feature points to obtain the flexible screen structure geometric relationship between the images after debubbling at different angles; Step S133: Based on the color deviation of the light source change between the images after debubbling at different spectra and the geometric relationship of the flexible screen structure between the images after debubbling at different angles, the screen structure complementary information between the corresponding flexible screen images after debubbling is analyzed to obtain the screen structure feature complementary information between the images after debubbling at different flexible screens.

4. The method for rapid detection of defects in flexible screens after degassing according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: reconstructing the structural phase topography of the composite image of the flexible screen after degassing to generate three-dimensional topography data of the flexible screen surface; Step S22: setting a marking point on the edge of the flexible screen and obtaining the deformation amount of the edge marking position corresponding to the flexible screen according to the marking point; Step S23: dynamically capturing the three-dimensional topography data of the flexible screen surface before and after deformation based on the deformation amount of the edge mark position corresponding to the flexible screen, to obtain the topography data of the flexible screen surface before and after deformation; Step S24: Obtaining the deformation stress source distribution corresponding to each edge deformation mark point based on the deformation amount of the edge mark position corresponding to the flexible screen, and performing a shape stress compensation fitting analysis on the flexible screen surface topography data before and after deformation based on the deformation stress source distribution corresponding to each edge deformation mark point to generate a flexible screen surface topography stress dynamic compensation curve; Step S25: performing dynamic deformation compensation on the three-dimensional surface topography data of the flexible screen based on the dynamic compensation curve of the surface topography stress of the flexible screen to obtain deformation-compensated topography data of the screen surface.

5. The method for rapid detection of defects in flexible screens after degassing according to claim 4, characterized in that: Step S21 includes the following steps: A sinusoidal fringe pattern is projected onto the composite image after de-bubbling the flexible screen to generate a sinusoidal fringe projection image of the flexible screen structure; The phase deformation jump corresponding to each sinusoidal fringe in the fringe image is calculated by projecting the sinusoidal fringe image of the flexible screen structure, and phase unwrapping processing is performed based on the phase deformation jump corresponding to each sinusoidal fringe in the fringe image to obtain the two-dimensional phase information of the deformation of the flexible screen surface structure; Based on the two-dimensional phase information of the deformation of the flexible screen surface structure, the three-dimensional morphology of each point in the corresponding flexible shielding surface structure in the composite image of the flexible screen after debubbling is reconstructed to generate three-dimensional morphology data of the flexible screen surface.

6. The method for rapid detection of defects in flexible screens after degassing according to claim 4, characterized in that: The step S22 of obtaining the deformation amount of the edge mark position corresponding to the flexible screen according to the mark point includes the following steps: Obtain the corresponding surface deformation stress-strain distribution on the edge of the flexible screen according to the marked points; Based on the surface deformation stress-strain distribution corresponding to each marking point on the edge of the flexible screen, the displacement increment of the corresponding marking point is calculated to obtain the displacement increment and displacement direction corresponding to each marking point on the edge of the flexible screen; Based on the displacement direction corresponding to each mark point on the edge of the flexible screen, the position deformation of the displacement increment corresponding to each mark point on the edge of the flexible screen is predicted to obtain the position deformation of the edge mark corresponding to the flexible screen.

7. The method for rapid detection of defects in flexible screens after degassing according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: performing high-order polynomial fitting on the screen surface deformation compensation topography data, so as to correct points that deviate from the normal curved surface due to environmental changes or screen quality fluctuations by using a local weighted regression model, thereby obtaining the screen surface abnormal shape calibration topography data; Step S32: using a numerical differentiation method to perform a local curvature analysis on each point on the flexible screen surface in the screen surface abnormal morphology calibration topography data, to obtain the local curvature corresponding to each point on the flexible screen surface after de-bubbling; Step S33: obtaining the corresponding normal direction of the flexible screen surface by calibrating the topography data of the abnormal morphology of the screen surface, and performing principal curvature calculation on the local curvature corresponding to each point on the surface of the flexible screen after de-bubbling based on the normal direction of the flexible screen surface, so as to obtain the principal curvature corresponding to each point on the surface of the flexible screen after de-bubbling; Step S34: performing maximum value decomposition on the principal curvature corresponding to each point on the surface of the flexible screen after de-bubbling by combining a preset Jacobi matrix to obtain the maximum and minimum principal curvatures corresponding to each point on the surface of the flexible screen after de-bubbling, and performing Gaussian curvature calculation based on the maximum and minimum principal curvatures corresponding to each point on the surface of the flexible screen after de-bubbling to obtain the Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling; Step S35: Based on the principal curvature and Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling, the abnormal curvature area of ​​the corresponding flexible screen after de-bubbling is identified, so as to calculate the corresponding curvature peak value according to the principal curvature and Gaussian curvature, and divide the corresponding abnormal defect area range according to the judgment result between the preset curvature division interval and the curvature peak value, so as to obtain the abnormal defect curvature area of ​​the screen after de-bubbling.

8. The method for rapid detection of defects in flexible screens after degassing according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: obtaining the corresponding screen defect length, screen defect width, and screen defect depth after defoaming through the abnormal defect curvature area after defoaming; Step S42: calculating the aspect ratio of the defect after defoaming the screen according to the length and width of the defect after defoaming the screen, to obtain the aspect ratio of the defect after defoaming the screen; Step S43: estimating the corresponding area of ​​the defect after debubbling based on the length and width of the defect after debubbling and the curvature area of ​​the abnormal defect after debubbling, and taking the depth of the defect after debubbling, the aspect ratio of the defect after debubbling, and the area of ​​the defect after debubbling as defect geometric features to obtain the geometric features of the defect after debubbling on the flexible screen; Step S44: Construct a deep learning model based on the U-Net architecture, and input the geometric features of the defects of the flexible screen after degassing into the trained deep learning model to perform defect classification detection on the abnormal defect curvature area of ​​the screen after degassing, so as to detect the corresponding bubble residue, indentation, scratch and degumming type defect results after the flexible screen is degassing.

9. The method for rapid detection of defects in flexible screens after degassing according to claim 8, characterized in that: Step S44 includes the following steps: By building a deep learning model based on the U-Net architecture, the U-Net architecture uses an encoder-decoder part, with a multi-scale feature extraction module in the encoder part to simultaneously capture local details and global context feature information corresponding to the defect. In addition, the attention mechanism is introduced in the decoder part to enhance the detection ability of small defects and low-contrast defect features. Obtain a small number of labeled defect samples and a large number of unlabeled normal samples, and use a semi-supervised learning method to input a small number of labeled defect samples and a large number of unlabeled normal samples into a deep learning model based on the U-Net architecture for model training. Generate realistic defect samples through a generative adversarial network to expand the training set, and perform model enhancement based on the expanded training set to generate a trained deep learning model; The geometric features of defects after debubbling of the flexible screen are input into the corresponding multi-scale feature extraction module in the trained deep learning model, and defect classification detection is performed on the defect areas detected in the abnormal defect curvature area after debubbling of the screen to detect the corresponding bubble residue, indentation, scratch and degumming type defect results after debubbling of the flexible screen.

10. A system for rapid detection of defects in flexible screens after de-bubbling, characterized in that: Used to perform the method for quickly detecting defects of a flexible screen after defoaming according to claim 1, the system for quickly detecting defects of a flexible screen after defoaming comprises: An image complementary registration and fusion module is used to obtain images of the flexible screen after debubbling corresponding to illumination at different spectra and angles, and to perform image complementary registration and fusion on the images of the flexible screen after debubbling corresponding to illumination at different spectra and angles to generate a composite image of the flexible screen after debubbling; The screen topography dynamic compensation module is used to reconstruct the structural phase topography of the composite image of the flexible screen after de-bubbling to generate three-dimensional topography data of the flexible screen surface. The module sets marking points on the edge of the flexible screen and obtains the deformation amount of the edge marking position corresponding to the flexible screen based on the marking points. The module then performs dynamic deformation compensation on the three-dimensional topography data of the flexible screen surface based on the deformation amount of the edge marking position corresponding to the flexible screen, thereby obtaining deformation-compensated topography data of the screen surface. The abnormal curvature identification module is used to analyze the curvature of each surface point on the deformation compensation topography data of the screen surface to obtain the principal curvature and Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling; based on the principal curvature and Gaussian curvature corresponding to each point on the surface of the flexible screen after de-bubbling, the abnormal curvature area of ​​the corresponding flexible screen after de-bubbling is identified to obtain the abnormal defect curvature area of ​​the screen after de-bubbling; The defect classification and detection module is used to analyze the defect geometric features of the abnormal defect curvature area after the screen is debubbled, so as to obtain the defect geometric features of the flexible screen after debubbling; a deep learning model based on the U-Net architecture is constructed, and the defect geometric features of the flexible screen after debubbling are input into the trained deep learning model for defect classification detection, so as to detect the corresponding bubble residue, indentation, scratch and degumming type defect results after the flexible screen is debubbled.

Images