An ultrathin glass defect detection system and positioning method based on line scanning and phase deflectometry

Abstract

This invention discloses an ultrathin glass defect detection system and localization method based on line scanning and phase deflection. The defect detection system includes a line scanning camera module, a distortion-free phase deflection detection module, a high-magnification microscope module, and a central processing unit. The line scanning camera module is used to acquire 2D grayscale images of the ultrathin glass. The distortion-free phase deflection detection module is used to acquire surface height maps of the ultrathin glass. The high-magnification microscope module is used to perform depth localization of interlayer defects determined by the central processing unit. The central processing unit is used to receive the 2D grayscale images and surface height maps, execute image registration and defect classification algorithms, and send depth localization control commands to the high-magnification microscope module. This invention constructs a multi-module collaborative detection system that balances detection efficiency and accuracy, providing data support for production process improvement.

Description

Technical Field

[0001] This invention relates to the field of machine vision inspection technology, and more specifically, to an ultra-thin glass (UTG) defect detection system and positioning method based on line scanning and phase measuring deflectometry (PMD), which is particularly suitable for UTG products with protective films, and can achieve accurate differentiation between surface defects and interlayer defects and deep positioning of interlayer defects. Background Technology

[0002] Ultra-thin glass (UTG), with its excellent foldable properties, has become a core material in the field of flexible displays. During UTG production and subsequent film application, protective films (PF) are typically applied to both the top and bottom surfaces to prevent damage to the glass surface. However, various defects are inevitably introduced during production or film application, mainly including foreign matter and air bubbles between the UTG and the protective film layers, as well as dust and scratches on the protective film surface. These defects can seriously affect the appearance quality and performance of the product, thus requiring rigorous defect detection.

[0003] In existing technologies, UTG defect detection mainly employs 2D imaging methods such as line scan cameras. Line scan cameras can acquire full-surface images of the UTG at high speed, detecting all optical anomalies and offering the advantage of efficient full inspection. However, a fatal flaw of 2D imaging technology is its inability to provide depth information, making it difficult to distinguish between surface defects and interlayer defects. For example, dust particles adhering to the surface of the top protective film and foreign objects embedded between the UTG and the protective film may exhibit similar grayscale characteristics in the line scan image, making it impossible for the detection system to accurately distinguish them. This results in an extremely high overkill rate, which not only affects detection efficiency but may also lead to qualified products being mistakenly identified as defective, increasing production costs.

[0004] Phase deflection (PD) is a high-precision 3D surface topography measurement technique. By projecting a fringe pattern onto the object under test and analyzing the deformation of the reflected fringes, it can reconstruct the surface gradient and integrate to obtain a surface height map, exhibiting extremely high sensitivity to surface unevenness. However, in traditional PMD devices, to satisfy the reflection principle that "the angle of incidence equals the angle of reflection," the light source and camera usually need to be placed at an angle, making it impossible to be perpendicular to the surface being measured. This tilted acquisition method leads to severe perspective distortion in the acquired fringe images, which not only increases the complexity of subsequent algorithm processing but also introduces measurement errors, affecting the accuracy of 3D reconstruction and limiting its application in UTG defect detection.

[0005] Therefore, how to combine the high-speed 2D full inspection capability of line scanning cameras with the 3D morphology analysis capability of PMD, while solving the image distortion problem of traditional PMD systems, and achieving accurate differentiation between UTG surface defects and interlayer defects and deep positioning of interlayer defects, is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an ultrathin glass defect detection system and positioning method based on line scanning and phase deflection. Specifically, it aims to achieve the following objectives: solve the problems of existing 2D detection technologies being unable to effectively distinguish between UTG surface defects and interlayer defects, and having a high false positive rate; provide a distortion-free PMD detection module to eliminate image distortion caused by tilted shooting in traditional PMD systems and improve 3D reconstruction accuracy; achieve deep positioning of interlayer defects and clarify the specific layer where the defect is located; and construct a multi-module collaborative detection system that balances detection efficiency and accuracy, providing data support for production process improvement.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A defect detection system for ultrathin glass based on line scanning and phase deflection includes a line scanning camera module, a distortion-free phase deflection detection module, a high-magnification microscope module, and a central processing unit. The line scanning camera module acquires 2D grayscale images of the front and back surfaces of the ultrathin glass. The distortion-free phase deflection detection module performs optical optimization and calibration for the on-site environment and acquires a surface height map of the ultrathin glass, which is a distortion-free surface height map. The central processing unit receives the 2D grayscale images and the ultrathin glass surface height map, performs front-back image registration and defect classification algorithms for precise defect localization and morphological classification, and transmits the images to the high-magnification microscope. The microscope module sends a depth positioning control command; the high-magnification microscope module is used to perform depth positioning on the ultra-thin glass surface and interlayer defects determined by the central processing unit, and determine the specific location of the defects; the central processing unit is also configured to: (1) control the high-magnification microscope module to move to its coordinate position for defects classified as interlayer defects; (2) control the high-magnification microscope module to scan and autofocus in the depth direction to determine the precise depth position of the defect; (3) combine the pre-stored thickness parameters of the ultra-thin glass multilayer structure to determine the specific interlayer between the upper protective film and the glass, inside the glass, or between the glass and the lower protective film.

[0009] Furthermore, the distortion-free phase deflection detection module includes a high-resolution industrial camera, a light source screen, and a beam splitter. This module transmits the acquired stripe image to a central processing unit, which reconstructs a surface height map based on the stripe image. The optical axis of the line scan camera module, combined with the distortion-free phase deflection detection module, performs multi-dimensional imaging of defects in the ultra-thin glass. The central processing unit executes image registration and defect classification algorithms to obtain accurate defect locations, and then classifies the defects based on their location, shape, and grayscale features.

[0010] Furthermore, the high-magnification microscope module includes an objective lens, a Z-axis motorized platform, and an autofocus unit; the high-magnification microscope module is communicatively connected to the central processing unit.

[0011] This invention also provides a method for locating defects in ultrathin glass based on line scanning and phase deflection, comprising the following steps:

[0012] Step S1: The central processing unit sends an acquisition command to the line scan camera module. The line scan camera module scans the ultrathin glass sample and outputs a 2D grayscale image to the central processing unit. The central processing unit executes a defect detection algorithm on the 2D grayscale image and extracts a preliminary defect list. The preliminary defect list includes the center coordinates and defect outline of each defect.

[0013] Step S2: The central processing unit sends an acquisition command to the distortion-free phase deflection detection module. The distortion-free phase deflection detection module scans the same ultrathin glass sample and outputs a surface height map to the central processing unit.

[0014] Step S3: The central processing unit performs a spatial registration algorithm on the 2D grayscale image and the surface height map to establish a mapping relationship between the 2D grayscale image coordinate system and the surface height map coordinate system;

[0015] Step S4: Based on the mapping relationship, the central processing unit maps each defect in the preliminary defect list to the surface height map, analyzes the height characteristics of the defect mapping area, classifies the defects, and distinguishes between interlayer defects and surface defects.

[0016] Step S5: The central processing unit sends a positioning command to the high-magnification microscope module, controls the high-magnification microscope module to move to the location of the interlayer defect and perform automatic focusing to determine the depth layer where the interlayer defect is located.

[0017] Furthermore, the specific process of the distortion-free phase deflection detection module outputting the surface height map in step S2 is as follows: The central processing unit sends a control command to the light source screen, which projects a stripe pattern with N-step phase shifts; the central processing unit sends an acquisition command to the camera, which acquires the deformed stripes reflected by the ultra-thin glass surface and transmits the deformed stripe image to the central processing unit; the central processing unit performs a phase shift algorithm on the deformed stripe image to calculate the wrapped phase; then performs an unwrapping algorithm on the wrapped phase to obtain the continuous phase; the central processing unit calculates the surface horizontal gradient and the surface vertical gradient according to the mapping relationship between the continuous phase and the surface gradient; the central processing unit performs an integration algorithm on the surface horizontal gradient and the surface vertical gradient, the integration algorithm including Fourier integration or least squares integration, to reconstruct the surface height map.

[0018] Furthermore, the process of the central processing unit executing the spatial registration algorithm in step S3 includes at least one of the following methods: Method A: Gray-scale-based registration, specifically: The camera of the distortion-free phase deflection detection module synchronously acquires the intensity image I of the ultrathin glass. pmd_intensity (u,v) is transmitted to the central processing unit (CPU); the CPU constructs a transformation matrix M, which maps the coordinate system of the intensity image to the coordinate system of the 2D grayscale image; the CPU calculates the normalized cross-correlation value between the 2D grayscale image and the transformed intensity image; the CPU adjusts the parameters of the transformation matrix M through an optimization algorithm to maximize the normalized cross-correlation value, obtaining the optimal transformation matrix M1; Method B: Registration based on product boundaries, specifically: the CPU performs an edge detection algorithm on the 2D grayscale image to locate the ultrathin... The four boundaries of the glass are analyzed, and the intersection points of the four boundaries are calculated to obtain the corner point set of the line scan image. The central processing unit performs an edge detection algorithm on the surface height map to locate the four boundaries of the ultrathin glass and calculates the intersection points of the four boundaries to obtain the corner point set of the PMD image. The central processing unit uses the corner point set of the line scan image and the corner point set of the PMD image as matching features to construct an overdetermined system of equations. The central processing unit uses the least squares method to solve the overdetermined system of equations to obtain the affine transformation matrix M2, realizing the mapping from the 2D grayscale image coordinate system (x,y) to the surface height map coordinate system (u,v).

[0019] Furthermore, the specific process of the central processing unit executing the defect classification algorithm in step S4 is as follows: the central processing unit calls the transformation matrix M obtained in step S3 to map the contour Ci of each defect Di in the preliminary defect list to the surface height map HPmd(u,v), obtaining the corresponding mapped region Ri; the central processing unit calculates the local height difference of the mapped region Ri. The system performs curvature analysis and gradient continuity analysis algorithms on the mapped region Ri to extract curvature and gradient continuity features. The central processing unit presets a height threshold Th ≥ 1 μm and constructs classification criteria: if Δhi > Th and the curvature features show smooth changes and the gradient continuity features show continuous without abrupt changes, then the defect is determined to be an interlayer defect; if Δhi ≤ Th, or Δhi > Th but the curvature features show sharp abrupt changes and the gradient continuity features show discontinuities, then the defect is determined to be a surface defect.

[0020] Furthermore, the specific process of the central processing unit controlling the high-magnification microscope module to perform depth positioning in step S5 is as follows: The central processing unit extracts the center coordinates (xi, yi) of the interlayer defects from the preliminary defect list and sends a displacement command to the Z-axis motorized platform of the high-magnification microscope module, controlling the Z-axis motorized platform to move the objective lens above the (xi, yi) position; the central processing unit sends a scanning command to the Z-axis motorized platform, controlling the Z-axis motorized platform to scan uniformly within the preset Z-axis range with the objective lens, and at each scanning position zk, controls the objective lens to acquire a high-magnification image Ik and transmit it to the automatic alignment system. The autofocus unit calls the built-in sharpness evaluation function to calculate the sharpness S(zk) of each high-magnification image Ik, and transmits the sharpness S(zk) to the central processing unit. The central processing unit executes a peak detection algorithm on the sharpness S(zk) to find the Z-axis position zfocus corresponding to the maximum sharpness. The central processing unit pre-stores the structural parameters of the ultrathin glass sample, including the thickness of the upper protective film, the thickness of the ultrathin glass, and the thickness of the lower protective film. By comparing the absolute coordinates of zfocus with the numerical range of the structural parameters, the depth layer where the interlayer defects are located is determined.

[0021] Furthermore, the defect detection algorithms executed by the central processing unit include threshold segmentation algorithms, edge detection algorithms, or U-Net algorithms based on deep learning; it also includes preprocessing algorithms performed by the central processing unit on 2D grayscale images, the preprocessing algorithms including filtering algorithms and contrast enhancement algorithms, used to remove noise in 2D grayscale images and improve the recognizability of defect features.

[0022] Furthermore, the interlayer defects include foreign matter between the ultrathin glass and the upper protective film, foreign matter between the ultrathin glass and the lower protective film, and air bubbles between the ultrathin glass and the protective film; the surface defects include dust on the surface of the upper protective film, dust on the surface of the lower protective film, and scratches on the surface of the protective film.

[0023] The present invention has the following beneficial effects:

[0024] High detection accuracy: The distortion-free PMD detection module eliminates image distortion through a vertical optical path design, significantly improving the accuracy of 3D height map reconstruction; combined with a defect classification algorithm based on multi-dimensional features such as height difference and curvature, it achieves accurate differentiation between surface defects and interlayer defects, greatly reducing the misjudgment rate; the autofocus technology of the high-magnification microscope controls the depth positioning error of interlayer defects to the micrometer level.

[0025] High inspection efficiency: The line scan camera module has high-speed full inspection capability, which can quickly cover the entire surface of UTG and avoid missing defects; the defect classification algorithm automatically filters surface defects and only performs deep positioning on interlayer defects, reducing invalid inspection steps; the modules work together to realize an automated process from image acquisition to depth positioning without manual intervention, meeting the high-efficiency inspection needs of industrial production.

[0026] Highly practical: This system and method are applicable to UTG products with protective films and can detect various defect types such as foreign objects, bubbles, dust, and scratches. The defect classification results and depth positioning information provide key data support for production process improvement, which helps to trace the defect generation process and optimize the film application process or glass manufacturing process. The detection system has mature hardware configuration and highly portable algorithm process, which is convenient for industrial promotion and application.

[0027] Outstanding innovation: It proposes an optical path design for a distortion-free PMD detection module, solving the image distortion problem of traditional PMD systems from a hardware perspective; it constructs a full-process detection solution of "2D initial inspection - 3D classification - depth positioning", integrating the advantages of line scanning technology, PMD technology and autofocus technology, realizing the integration of defect detection and positioning, and filling the gap in existing technology for UTG defect depth detection.

[0028] To more clearly illustrate the structural features and effects of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the hardware architecture of the defect detection system according to an embodiment of the present invention;

[0030] Figure 2 This is a schematic diagram of the optical path structure of the distortion-free PMD detection module in an embodiment of the present invention;

[0031] Figure 3 This is a flowchart illustrating the defect location method according to an embodiment of the present invention.

[0032] Figure 4 This is a schematic diagram of the registration between the line scan image and the PMD height map in an embodiment of the present invention;

[0033] Figure 5This is a schematic diagram comparing the features of interlayer defects and surface defects on the PMD height map in an embodiment of the present invention;

[0034] Figure 6 This is a schematic diagram illustrating the principle of automatic focusing of a high-magnification microscope to determine the depth of defects in an embodiment of the present invention. Detailed Implementation

[0035] The present invention will now be further described in conjunction with the accompanying drawings and relevant knowledge, and will be described clearly and completely. Obviously, the described applications are only some embodiments of the present invention, and not all embodiments.

[0036] The purpose of this invention is to provide a method for detecting and locating UTG defects with a protective film based on line scanning and phase deflection. This solves the problem in existing technologies where the distinction between UTG surface defects and interlayer defects is ineffective, leading to a high false positive rate. It also addresses the problem in existing technologies where interlayer defects cannot be accurately located. Furthermore, it provides a distortion-free PMD measurement device to solve the image distortion problem caused by camera tilt in traditional PMD systems, thereby improving the accuracy of 3D reconstruction. Specifically, it includes the following:

[0037] Reference Figures 1-6 As shown, this invention provides an ultrathin glass defect detection system based on line scanning and phase deflection, including a line scanning camera module, a distortion-free phase deflection detection module, a high-magnification microscope module, and a central processing unit. The invention first uses a line scanning camera for high-speed full inspection to acquire the location of all potential defects; then, a distortion-free PMD system including a beam splitter is used to acquire the surface height map of the UTG; next, the line scanning image and the PMD height map are precisely registered; finally, by analyzing the morphological characteristics of the defect location on the height map, surface defects and interlayer defects are distinguished, and high-magnification microscopy with automatic focusing is performed on the interlayer defects to determine their specific depth.

[0038] Specifically, the line scan camera module is used to acquire high-resolution 2D grayscale images. line (x,y) can quickly cover the entire surface of ultra-thin glass and efficiently capture all optical anomalies;

[0039] The distortion-free phase deflection detection module (distortion-free PMD module) is used to acquire 3D height maps H. pmd (u,v) obtains accurate 3D surface topography data through distortion-free design, making up for the lack of depth information in 2D images, providing core height dimension support for distinguishing surface defects from interlayer defects, and improving the accuracy of defect classification from the data level.

[0040] More specifically, this invention employs an optimized distortion-free phase deflection detection module, specifically including: Camera 1: located directly above the product under test, with its optical axis perpendicular to the surface under test; this camera is a high-resolution industrial camera. Beam splitter 2: a semi-transparent, semi-reflective mirror placed at a 45° angle, located below the camera. Light source screen 3: typically an LCD screen, placed horizontally from the side, its light is projected onto the reflective surface of the beam splitter. Optical path: the customized stripe image emitted from the light source screen 3 is reflected by the 45° beam splitter and projected vertically downwards onto the UTG test plane. After reflection from the test plane (carrying surface topography information), the light penetrates vertically upwards through the beam splitter 2 and is directly acquired by the camera 1 located directly above. This design allows the camera 1 to acquire images perpendicular to the test plane, perfectly solving the image distortion problem that occurs when there is an angle between the camera and the normal vector of the test plane. The camera acquires a "virtual image" of the stripes in the test plane, as shown in the reference diagram. Figure 1 As shown, the distortion-free phase deflection detection module also includes a transmission light source 8 and a reference surface 9. The transmission light source 8 and the reference surface 9 constitute the reference surface-transmission light PMD system in the distortion-free phase deflection detection module. The camera 1, the light source screen 3, the beam splitter 2, and the reference surface 9 constitute the reference surface-reflected light PMD system in the distortion-free phase deflection detection module.

[0041] Furthermore, the optical axis of camera 1 is perpendicular to the surface of the ultrathin glass under test, used to acquire the stripe image reflected by the ultrathin glass surface. This perpendicular acquisition method avoids image distortion caused by tilted shooting at the hardware level, ensuring that the stripe image accurately reflects the surface morphology of the ultrathin glass. Beam splitter 2 is positioned between camera 1 and the surface of ultrathin glass 4, used to reflect the light emitted from the light source screen 3 onto the ultrathin glass surface, and to allow the light reflected from the ultrathin glass surface to pass through to the camera. The beam splitter achieves a clever reversal of the optical path, ensuring perpendicular acquisition by the camera while avoiding interference between the light source and the camera's optical path, thus compacting the system structure. To ensure the stability and perpendicularity of light transmission and improve imaging consistency, the light source screen 3 is set on the side of the beam splitter 2 to project phase-shift fringe patterns onto the beam splitter. It can accurately output the preset N-step phase-shift fringes, providing standardized optical signals for phase calculation and ensuring that the fringe intensity distribution meets the algorithm requirements. The camera is used to transmit the acquired fringe images to the central processing unit, which reconstructs the surface height map based on the fringe images. This achieves accurate acquisition and transmission of deformed fringes, ensuring that the optical signals containing surface topography information are completely transmitted to the central processing unit, and guaranteeing the continuity and accuracy of the height map reconstruction.

[0042] The high-magnification microscope module is used to locate the interlayer defects identified by the central processing unit. It features a Z-axis motorized platform 5 and autofocus (AF) function, such as 10x magnification, for the depth location of defects. With the help of high-magnification imaging and precise Z-axis control, it enables microscopic observation and depth coordinate locking of interlayer defects, providing specific depth location information for defect tracing.

[0043] Furthermore, the high-magnification microscope module includes an objective lens 6, a Z-axis motorized platform 5, and an autofocus unit. The objective lens 6 is mechanically connected to the Z-axis motorized platform 5, which receives displacement commands from the central processing unit and moves the objective lens 6 along the Z-axis, achieving high-precision, automated displacement of the objective lens in the Z-axis direction. The displacement accuracy can reach the micrometer level, providing stable mechanical support for depth scanning and ensuring the accuracy of depth positioning. The autofocus unit calculates the sharpness value based on the high-magnification image acquired by the objective lens and transmits it to the central processing unit. By calculating the image sharpness in real time, it provides a quantitative basis for judging the focal position of the defect 7, replacing manual focusing, improving focusing efficiency and consistency, and avoiding human operation errors. The sharpness evaluation function includes the Tenengrad gradient energy function or the Laplacian operator function. Both functions can efficiently reflect the sharpness of image details and are sensitive to focal changes in high-magnification images, ensuring the accuracy of sharpness peak detection and thus guaranteeing the accuracy of defect depth positioning.

[0044] The central processing unit receives 2D grayscale images and surface height maps, executes image registration and defect classification algorithms, and sends depth positioning control commands to the high-magnification microscope module. As the core control and computing hub of the system, it enables collaborative processing of data from various modules and precise scheduling of work sequences, ensuring the automation and continuous operation of the inspection process and significantly improving the consistency of inspection efficiency and accuracy.

[0045] This invention also provides a method for locating defects in ultrathin glass based on line scanning and phase deflection, comprising the following steps:

[0046] Step S1: The central processing unit sends an acquisition command to the line scan camera module. The line scan camera module scans the ultrathin glass sample and outputs 2D grayscale images of the front and back sides to the central processing unit. The central processing unit executes a defect detection algorithm on the 2D grayscale images to extract a preliminary defect list. The preliminary defect list includes the center coordinates and contours of each defect. Through preprocessing and professional defect detection algorithms, all potential defects are accurately screened from the 2D images, clarifying the planar position and contour shape of the defects, providing a clear target candidate set for subsequent 3D classification and depth localization.

[0047] Step S2: The central processing unit sends an acquisition command to the distortion-free phase deflection detection module. The distortion-free phase deflection detection module scans the same ultrathin glass sample and outputs a surface height map to the central processing unit. It acquires 3D height data that is of the same origin as the 2D image, establishes the correlation between the planar position and height characteristics of the defect, provides a core basis for distinguishing between surface defects and interlayer defects, and breaks through the depth blind zone of traditional 2D detection.

[0048] Step S3: The central processing unit performs a spatial registration algorithm on the 2D grayscale image and the surface height map to establish a mapping relationship between the coordinate system of the 2D grayscale image and the coordinate system of the surface height map; eliminates coordinate deviations caused by differences in imaging perspective and installation position of different modules, and ensures that the initial defects in the 2D image can be accurately mapped to the corresponding area of ​​the 3D height map.

[0049] Step S4: Based on the mapping relationship, the central processing unit maps each defect in the preliminary defect list to the surface height map, analyzes the height characteristics of the defect mapping area, classifies the defects, and distinguishes between interlayer defects and surface defects. Through quantified height difference, curvature and gradient continuity characteristics, it achieves accurate differentiation between surface defects and interlayer defects, significantly reduces the misjudgment rate of traditional 2D inspection, screens out interlayer defects that truly affect product quality, provides a clear target for subsequent in-depth positioning, and improves the targeting of inspection.

[0050] Step S5: The central processing unit sends a positioning command to the high-magnification microscope module, controls the high-magnification microscope module to move to the location of the interlayer defect and perform automatic focusing to determine the depth layer where the interlayer defect is located. Only key interlayer defects are located for depth positioning to avoid invalid detection and improve detection efficiency. Through automatic focusing and peak detection, the defect depth coordinates are accurately obtained, and the layer where the defect is located is determined in combination with the sample structural parameters.

[0051] Further, the specific process of the distortion-free phase deflection detection module outputting the surface height map in step S2 is as follows: The central processing unit sends a control command to the light source screen, which projects a stripe pattern with N-step phase shift, providing multiple frames of standardized stripes with phase differences to provide sufficient phase information for the phase shift algorithm to calculate the wrapping phase, ensuring the accuracy of the phase calculation; the central processing unit sends an acquisition command to the camera, which acquires the deformed stripes reflected by the ultra-thin glass surface and transmits the deformed stripe image to the central processing unit; the central processing unit performs a phase shift algorithm on the deformed stripe image to calculate the wrapping phase, accurately capturing the deformed stripes carrying surface morphology information, ensuring that the deformation features of the stripes are completely preserved, and providing reliable image data for subsequent phase calculation; then, an unwrapping algorithm is performed on the wrapping phase to obtain the continuous phase; the central processing unit calculates the surface horizontal gradient and the surface vertical gradient according to the mapping relationship between the continuous phase and the surface gradient; the central processing unit performs an integration algorithm on the surface horizontal gradient and the surface vertical gradient, the integration algorithm including Fourier integration or least squares integration, to reconstruct the surface height map. By calculating the intensity information of multiple frames of fringes, the wrapping phase is obtained, and the correlation between fringe deformation and surface phase is initially established, laying the foundation for subsequent continuous phase acquisition. The "phase entanglement" problem of the wrapping phase is eliminated, and continuous phase data that can truly reflect the surface gradient is obtained, ensuring the linear correlation between phase and surface morphology. The phase information is transformed into surface geometric feature parameters, establishing a direct correlation between phase and surface concavity and convexity, providing core geometric data for height map reconstruction. The gradient data is integrated through a professional integration algorithm to accurately reconstruct the 3D height distribution of the ultrathin glass surface. The height measurement accuracy can reach the micrometer level, providing high-quality 3D data support for defect classification.

[0052] Furthermore, the process of the central processing unit executing the spatial registration algorithm in step S3 includes at least one of the following methods:

[0053] Method A: Gray-scale-based registration, specifically: The camera of the distortion-free phase deflection detection module simultaneously acquires the intensity image I of the ultrathin glass. pmd_intensity (u,v) is generated and transmitted to the central processing unit (CPU). The CPU constructs a transformation matrix M, which maps the coordinate system of the intensity image to the coordinate system of the 2D grayscale image. The CPU calculates the normalized cross-correlation value between the 2D grayscale image and the transformed intensity image. The CPU adjusts the parameters of the transformation matrix M through an optimization algorithm to maximize the normalized cross-correlation value, thereby obtaining the optimal transformation matrix M1.

[0054] Method B: Registration based on product boundaries, specifically: The central processing unit (CPU) performs an edge detection algorithm on the 2D grayscale image to locate the four boundaries of the ultra-thin glass, calculates the intersection points of the four boundaries, and obtains the corner point set of the line scan image; the CPU performs an edge detection algorithm on the surface height map to locate the four boundaries of the ultra-thin glass, calculates the intersection points of the four boundaries, and obtains the corner point set of the PMD image; the CPU uses the corner point set of the line scan image and the corner point set of the PMD image as matching features to construct an overdetermined system of equations; the CPU uses the least squares method to solve the overdetermined system of equations to obtain the affine transformation matrix M2, realizing the mapping from the 2D grayscale image coordinate system (x,y) to the surface height map coordinate system (u,v). This method acquires a reference image with grayscale features homologous to the 2D grayscale image, providing a matching basis for grayscale correlation-based registration and ensuring the effectiveness of the registration. It achieves geometric association between the two coordinate systems through a transformation matrix, providing mathematical model support for coordinate mapping. The method quantifies the grayscale similarity between the two images, providing a clear evaluation index for the optimization of the transformation matrix and ensuring the correctness of the registration direction. It also achieves adaptive optimization of the transformation matrix, ensuring a high degree of matching of grayscale features between the two images, resulting in high registration accuracy. This method is suitable for ultrathin glass samples with rich surface textures.

[0055] Further optimization in Scheme B involves extracting the geometric feature points of the product as the registration reference. This method is unaffected by surface texture and is suitable for ultra-thin glass samples with smooth surfaces and minimal texture. Geometric reference points corresponding to the 2D image are obtained from the 3D height map to ensure the homogeneity and consistency of the registration reference. Constraint relationships are established using multiple reference points to improve the stability and accuracy of registration, avoiding registration deviations caused by a single reference point. The optimal transformation matrix is ​​solved through mathematical optimization, resulting in fast registration speed, high accuracy, and a certain degree of tolerance for boundary detection errors, ensuring the reliability of the registration results.

[0056] Furthermore, the specific process of the central processing unit executing the defect classification algorithm in step S4 is as follows: The central processing unit calls the transformation matrix M obtained in step S3 to map the contour Ci of each defect Di in the preliminary defect list to the surface height map HPmd(u,v), obtaining the corresponding mapped region Ri, realizing the accurate positioning of the defect in the 3D height map, ensuring the regional accuracy of subsequent height feature analysis, and avoiding classification errors caused by regional misalignment; the central processing unit calculates the local height difference of the mapped region Ri. The system performs curvature analysis and gradient continuity analysis algorithms on the mapped region Ri to extract curvature and gradient continuity features. Defect features are extracted from two dimensions: quantitative values ​​(height difference) and morphological features (curvature and gradient continuity), constructing a multi-dimensional classification basis to improve the accuracy and robustness of classification. The central processing unit presets a height threshold Th ≥ 1 μm and constructs classification criteria: if Δhi > Th and the curvature feature shows a smooth change, and the gradient continuity feature shows continuous without abrupt changes, then the defect is determined to be an interlayer defect; if Δhi ≤ Th, or Δhi > Th but the curvature feature shows a sharp abrupt change, and the gradient continuity feature shows discontinuity, then the defect is determined to be a surface defect. Through clear quantitative thresholds and morphological criteria, the system achieves automated and accurate differentiation of defect types, with an interlayer defect identification accuracy of over 95%, significantly reducing the false positive rate and providing accurate targets for subsequent deep localization.

[0057] Furthermore, the specific process of the central processing unit controlling the high-magnification microscope module to perform depth positioning in step S5 is as follows: The central processing unit extracts the center coordinates (xi, yi) of the interlayer defects from the preliminary defect list and sends a displacement command to the Z-axis motorized platform of the high-magnification microscope module, controlling the Z-axis motorized platform to move the objective lens above the (xi, yi) position; the central processing unit sends a scanning command to the Z-axis motorized platform, controlling the Z-axis motorized platform to scan uniformly within the preset Z-axis range with the objective lens, and at each scanning position zk, controls the objective lens to acquire a high-magnification image Ik and transmit it to the automatic... The focusing unit utilizes a built-in sharpness evaluation function to calculate the sharpness S(zk) of each high-magnification image Ik and transmits it to the central processing unit (CPU). The CPU executes a peak detection algorithm on the sharpness S(zk) to find the Z-axis position zfocus corresponding to the maximum sharpness. The CPU pre-stores the structural parameters of the ultra-thin glass sample, including the thickness of the upper protective film, the thickness of the ultra-thin glass, and the thickness of the lower protective film. By comparing the absolute coordinates of zfocus with the numerical range of the structural parameters, the depth level of the interlayer defect is determined. This ensures precise alignment of the objective lens with the plane position of the interlayer defect, guaranteeing the accuracy of the depth scan area and preventing the scan from deviating from the target defect. Uniform Z-axis scanning covers the possible depth range of the defect, acquiring multiple frames of high-magnification images at different focal lengths to provide sufficient data support for sharpness peak detection. The sharpness of each high-magnification image is quantified, converting the focusing effect into an analyzable numerical signal, providing an objective basis for determining the focus position. The system precisely locates the focal depth of defects. The peak detection algorithm is sensitive to changes in sharpness, achieving a positioning accuracy within 0.5μm, ensuring the accuracy of depth coordinates. Abstract depth coordinates are transformed into specific product-level information, clarifying whether the defect is located between the upper protective film and the ultra-thin glass, inside the ultra-thin glass, or between the ultra-thin glass and the lower protective film. In other words, the central processing unit is configured to: for defects classified as interlayer defects, control the high-magnification microscope module to move to the defect coordinate position; control the high-magnification microscope module to scan and automatically focus in the depth direction to determine the precise depth location of the defect; and, combined with pre-stored ultra-thin glass multilayer structure thickness parameters, determine the specific location of the interlayer defect—between the upper protective film and the glass, between the glass and the lower protective film, or inside the glass. The line scan camera module, combined with the distortion-free phase deflection detection module, performs multi-dimensional imaging of defects in the ultrathin glass. The central processing unit executes image registration and defect classification algorithms, combining defect location, shape, and grayscale features to obtain accurate defect location and classification results. Interlayer defects include foreign matter between the ultrathin glass and the upper protective film, foreign matter between the ultrathin glass and the lower protective film, and air bubbles between the ultrathin glass and the upper or lower protective film. Surface defects include dust on the surface of the upper protective film, dust on the surface of the lower protective film, and scratches on the surface of the upper or lower protective film.

[0058] Furthermore, the defect detection algorithms executed by the central processing unit include threshold segmentation algorithms, edge detection algorithms, or the U-Net algorithm based on deep learning. It also includes preprocessing algorithms performed by the central processing unit on the 2D grayscale image, including filtering and contrast enhancement algorithms, used to remove noise from the 2D grayscale image and improve the recognizability of defect features. Multiple algorithms can be flexibly selected according to the surface characteristics of ultra-thin glass (such as reflectivity and defect type). Threshold segmentation algorithms are suitable for scenarios with significant grayscale differences between defects and the background, edge detection algorithms are suitable for scenarios with clear defect outlines, and the U-Net algorithm is suitable for detecting small defects in complex backgrounds, improving the system's adaptability to different detection scenarios and the comprehensiveness of defect detection. Filtering algorithms effectively remove random noise and environmental interference from the image, preventing noise from being misjudged as defects; contrast enhancement algorithms amplify the grayscale differences between defects and the background, making the features of small defects more prominent, improving the recognition accuracy of subsequent defect detection algorithms, and reducing missed detections and false detections.

[0059] Furthermore, interlayer defects include foreign matter between the ultra-thin glass and the upper protective film, foreign matter between the ultra-thin glass and the lower protective film, and air bubbles between the ultra-thin glass and the protective film; surface defects include dust on the surface of the upper protective film, dust on the surface of the lower protective film, and scratches on the surface of the protective film. Clearly defining the specific types of these two categories of defects covers common defect types encountered during the production of ultra-thin glass products with protective films, making the application scenarios of the detection system more targeted. It also provides a clear classification basis for subsequent defect statistics and process optimization, thus assisting in production quality control.

[0060] In this invention, the high-speed full-inspection capability of the line scan camera module and the high-precision 3D imaging capability of the distortion-free PMD detection module work together. The line scan camera quickly filters out all potential defects, while the PMD module provides 3D height features of the defects. The combination of the two ensures detection efficiency and solves the problem that 2D detection cannot distinguish defect types. The vertical optical path design of the distortion-free PMD detection module works in conjunction with the image registration algorithm. The vertical optical path eliminates image distortion, providing a high-quality base image for the registration algorithm, while the registration algorithm ensures accurate correspondence between the 2D image and the 3D height map, providing a reliable coordinate mapping relationship for defect classification.

[0061] The defect classification algorithm works in conjunction with the high-magnification microscope module. The classification algorithm identifies the interlayer defects of true interest, providing the microscope module with clear detection targets, avoiding ineffective depth scans, and improving the efficiency of depth positioning. The high-precision positioning results from the microscope module, in turn, verify the accuracy of the classification algorithm, forming a closed-loop optimization. The central processing unit, as the core control module, collaborates with the other three functional modules to automate the image acquisition, data processing, and motion control processes. The working sequence and data transmission of each module are uniformly scheduled by the central processing unit, ensuring the continuity and stability of the detection process.

[0062] In other words, this invention uses a line scan camera to acquire a 2D grayscale image of the UTG and detects a preliminary defect list; it uses a phase deflection (PMD) system to acquire a surface height map of the UTG; it spatially registers the 2D grayscale image with the surface height map; and it classifies the defects according to the height characteristics of the area mapped by the preliminary defects on the height map to distinguish between interlayer defects and surface defects.

[0063] Gray-scale registration can be achieved by maximizing the normalized cross-correlation (NCC) between the 2D grayscale image and the intensity image obtained by the PMD system; or feature-based registration can be achieved by locating the intersection points of product boundaries on the 2D grayscale image and the height map respectively, and matching the coordinates of the intersection points.

[0064] Analyze the height difference, height continuity, or curvature characteristics of the defect mapping area; identify defects that are smooth and continuous convex or concave with a height difference greater than a preset threshold as interlayer defects; and identify other defects as surface defects.

[0065] For defects identified as interlayer defects, the high-magnification microscope module is moved to the defect location; an autofocus algorithm is executed to scan and calculate the sharpness of a series of images along the Z-axis (depth); based on the Z-axis position where the sharpness reaches its peak, the depth layer where the interlayer defect is located is determined. The sharpness evaluation functions used by the autofocus algorithm include the Tenengrad gradient energy function or the Laplacian operator function.

[0066] The PMD system is a distortion-free measurement device, which includes: a camera 1 for acquiring images; a light source screen 3; and a beam splitter 2. The camera 1 is located in the normal direction (directly above) of the UTG surface to be measured. The beam splitter 2 is positioned between the camera 1 and the UTG surface. The light source screen 3 is positioned to the side of the beam splitter 2. The light from the screen is reflected by the beam splitter and projected onto the UTG surface. The light reflected from the UTG surface passes through the beam splitter and is received by the camera.

[0067] The specific steps include:

[0068] Image and Data Acquisition: Line scan image acquisition, using a line scan camera module to scan UTG samples and acquire I... line (x,y). PMD data acquisition: Scan the same UTG sample using a distortion-free PMD module. Project a stripe pattern with N-step phase shifts (e.g., N=4). k (u,v).

[0069] ;

[0070] Among them: I k (u,v) represents the intensity of the stripe pattern at pixel (u,v), A(u,v) represents the average intensity, B(u,v) represents the intensity modulation, Φ(u,v) represents the absolute phase, and k represents the number of stripe patterns.

[0071] The camera captures deformed fringes reflected from the surface under test. The wrapping phase Φ(u,v) is calculated using a phase-shifting algorithm.

[0072] The continuous phase Φ(u,v) is obtained through an unwrapping algorithm. Based on the relationship between phase and gradient, the surface gradient (slope) G is calculated. x (u,v) and G y (u,v). The surface height map H is reconstructed by integration (such as Fourier integration or least squares integration). pmd (u,v).

[0073] Preliminary defect detection of line scan images: Line scan grayscale image I line (x,y) is preprocessed (e.g., filtering, contrast enhancement). Defect detection algorithms (e.g., threshold-based segmentation, edge detection, or deep learning-based U-Net, etc.) are used to process I. line Analyze (x, y) to extract the contours and locations of all potential defects. This yields a preliminary defect list L. defect ={D1,D2,...,D n}, where D i Contains the center coordinates (x) of defect i i ,y i ) and defect profile C i .

[0074] Line scan image registration with PMD height map:

[0075] Find a transformation matrix M that maps the line scan image coordinate system (x, y) to the PMD heightmap coordinate system (u, v). T =M·(x,y,1) T .

[0076] Registration can be performed using at least one of the following methods:

[0077] Method A: Based on grayscale registration, the PMD system can acquire an additional intensity image I while acquiring the stripe pattern. pmd_intensity (u,v). An optimization algorithm is used to find the optimal transformation matrix M1 to maximize I. line and I pmd_intensity Transformed normalized cross-correlation (NCC).

[0078] The NCC calculation formula is: ;

[0079] Where I1(u,v) is the gray value of the line scan image at pixel (u,v), and I2(u,v) is the gray value of the PMD image at pixel (u,v). The average gray value of the line scan image, This represents the average grayscale value of the PMD image.

[0080] Method B: Registration based on product boundaries (features), respectively in I line (x,y) and H pmd On (u,v), the four boundaries of the UTG product are accurately located through edge detection.

[0081] Calculate the four intersection points (i.e., product corner points) of these four boundary lines. Let P be the set of corner points in the line scan image. line ={P1,P2,P3,P4}, where P1, P2, P3, and P4 correspond to the coordinates of the four corner points of the line scan image, and the corresponding corner point set in the PMD image is P. pmd ={q1,q2,q3,q4}, where q1,q2,q3,q4 correspond to the coordinates of the four corner points of the PMD image.

[0082] The optimal affine transformation matrix M2 can be calculated by solving the overdetermined system of equations (e.g., using the least squares method).

[0083] Defect Classification and Filtering: For L defect Each defect D in i , its outline C i Mapped to PMD height map H via transformation matrix M pmd On (u,v), the corresponding mapping region R is obtained. i Analyze R i Height information within the region. Calculate the local height difference within this region. And the morphological features of the region, such as curvature and gradient continuity.

[0084] Classification criteria: Interlayer defects (foreign matter, bubbles): The presence of foreign matter or bubbles can cause localized, smooth "bulges" in flexible UTG or protective films. This is evident in H... pmdIt appears as a continuous bulge on the upper surface. Its Δhi is usually significantly greater than a certain threshold T h (e.g., T h = 1μm), and the curvature change in this area is continuous.

[0085] Surface defect (PF dust): Dust attached to the outermost protective film, if small, may not cause a significant deformation measurable by PMD (Δhi < Th). If the dust is large, although it will also cause a bulge, the continuity, reconstructed height, and morphology of the bulge are significantly different from those of the "bulge" - like interlayer defect (for example, it may be sharper, or cause noise in PMD reconstruction failure).

[0086] According to the above criteria, D with Δhi > Th and a morphology conforming to the "bulge" characteristics i is determined as an "interlayer defect"; other D i is determined as a "surface defect" and filtered.

[0087] Interlayer defect depth positioning:

[0088] For all D determined as "interlayer defects" i , record its central position (x i , y i ) in the line - scan coordinate system. Control the high - magnification microscope module (10 - fold magnification) to move above the defect position (x i , y i ). Execute the auto - focus (Auto - Focus, AF) algorithm and scan in the Z - axis (depth) direction. At each position Z k in the Z - axis scan, acquire a high - magnification image I k , and calculate the sharpness S(Z k ). The sharpness evaluation function S can select the gradient energy function (such as the Tenengrad function):

[0089] Tenengrad function: ;

[0090] where G x and G y are the horizontal and vertical gradients calculated using operators such as Sobel at (x, y) of the image I k .

[0091] Find the Z - axis position Z k that makes S(Z focus ) reach the maximum value. Due to the extremely small depth of field of the high - magnification microscope, through Z focusBy using the absolute coordinates of the UTG sample and combining them with the known structure of the UTG sample (thickness of the upper protective film, thickness of the UTG, and thickness of the lower protective film), the specific location of the interlayer defect can be determined (e.g., between the upper protective film and the UTG, or between the UTG and the lower protective film).

[0092] This invention combines the high-speed, full-inspection capability of a line-scan camera (ensuring no missed defects) with the 3D morphology analysis capability of a PMD (distinguishing defect types). Using the height information from the PMD, it effectively distinguishes between interlayer defects that truly cause deformation (such as foreign objects and bubbles) and surface defects that do not cause deformation or have different shapes (such as dust), significantly reducing the over-detection rate of traditional 2D inspection. Employing a PMD module with a beam splitter enables vertical imaging of the sample, eliminating image distortion at the hardware level and improving the accuracy and reliability of PMD 3D reconstruction. For "true defects" determined to be interlayer defects, this invention further utilizes the autofocus technology of a high-powered microscope, leveraging its shallow depth of field to precisely locate the defect along the Z-axis, clearly identifying the specific layer where the defect is located. Accurate defect classification and depth positioning information provide crucial data support for improvements in upstream bonding processes or glass manufacturing processes.

[0093] Example 1, referring to Figure 1 The defect detection system of this invention includes: a line scan camera module (for acquiring 2D grayscale image I) line A PMD detection module (for reconstructing 3D height maps H) pmd The system includes a high-magnification microscope module (e.g., a 10x objective lens with an integrated Z-axis autofocus platform) and a central processing unit. The UTG sample is placed on the motion platform, passes sequentially through the scanning areas of the line scan camera module and the PMD detection module, and is finally precisely positioned below the microscope module.

[0094] Reference Figure 2 The PMD detection module preferably employs a distortion-free optical path design. This design includes a PMD camera positioned directly above the sample, a 45° beam splitter positioned between the camera 1 and the sample, and a horizontally placed light source screen 3 (LCD screen). During operation, the striped light emitted from the light source is reflected by the beam splitter 2 and vertically illuminates the sample surface; the light reflected from the sample surface (carrying morphological information) passes through the beam splitter and is collected by the camera directly above. This orthogonal acquisition optical path avoids image distortion caused by the tilted shooting of traditional PMD cameras.

[0095] Reference Figure 3 The detection and positioning method of the present invention is as follows:

[0096] Step 1: Image Acquisition. The UTG sample moves, and the line scan camera acquires I... line (x,y), the PMD module obtains Hpmd (u, v).

[0097] Step 2: Coarse defect detection. The central processing unit processes I line Using a defect detection algorithm, extract the contours of all defects to obtain L defect .

[0098] Step 3: Image registration.

[0099] Refer to Figure 4 , in this embodiment, registration based on the product boundary is adopted. The processing unit respectively performs edge detection and Hough transform on I line and H pmd to find the four boundaries of the product, and calculate four corner points {P1, P2, P3, P4} and {q1, q2, q3, q4}. Then solve the affine transformation matrix M to establish the mapping relationship from (x, y) to (u, v).

[0100] Step 4: Defect classification. For any defect D defect in L i , map its contour C i through the M transformation to the region R pmd on H i .

[0101] Refer to Figure 5 , analyze the height profile of R i .

[0102] Step 5: If the defect is an interlayer foreign object or a bubble, it will lift the upper protective film and UTG, forming a smooth and continuous protrusion on H pmd . The processing unit calculates its height difference Δhi. If Δhi > Th (for example, T h = 1 μm), it is determined as an "interlayer defect".

[0103] Step 6: If the defect is dust on the surface of the protective film, its manifestation on H pmd may be a sharp spike or the height Δhi < Th. The processing unit determines it as a "surface defect" and filters it out.

[0104] Step 7: Depth positioning. For all D i determined as "interlayer defects", the processing unit controls the motion platform to move the center of D i to the center of the field of view of the microscope module.

[0105] Refer to Figure 6 , start autofocus. The Z-axis electric platform 5 drives the microscope objective to scan in the Z-axis (depth) direction, for example, from Z1 to Z n . At each Z k position, collect the image Ik And calculate the sharpness S(Z) k The sharpness curve S(Z) peaks at the physical level where the defect is located. (Through Z...) focus The value can be used to determine whether the defect is located between the upper protective film and the UTG (e.g., the focus is at Z). L1 ), or located between the UTG and the lower protective film (e.g., focal point at Z). L2 ).

[0106] The technical principles of the present invention have been described above with reference to specific embodiments, which are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments; all technical solutions falling within the scope of the present invention's concept are within its protection scope. Those skilled in the art can conceive of other specific embodiments of the present invention without creative effort, and these embodiments will all fall within the protection scope of the present invention.

Claims

1. A defect detection system for ultrathin glass based on line scanning and phase deflection, characterized in that, The system includes a line scan camera module, a distortion-free phase deflection detection module, a high-magnification microscope module, and a central processing unit. The line scan camera module acquires 2D grayscale images of the front and back surfaces of the ultrathin glass. The distortion-free phase deflection detection module acquires surface height maps of the ultrathin glass. The central processing unit receives the 2D grayscale images and surface height maps, performs front-back image registration and defect classification algorithms to achieve precise defect localization and morphological classification, and sends depth positioning control commands to the high-magnification microscope module. The high-magnification microscope module performs depth positioning on the ultrathin glass surface and interlayer defects determined by the central processing unit, determining the specific location of the defects. The central processing unit is configured to: control the high-magnification microscope module to move to the defect coordinate position for defects classified as interlayer defects; control the high-magnification microscope module to scan and autofocus in the depth direction to determine the precise depth position of the defects; and combine this with pre-defined parameters. Based on the stored thickness parameters of the multilayer structure of ultrathin glass, the specific location of the interlayer defect is determined between the upper protective film and the glass, between the glass and the lower protective film, or inside the glass. The specific process of the distortion-free phase deflection detection module outputting the surface height map is as follows: The central processing unit sends a control command to the light source screen, which projects an N-step phase-shifted fringe pattern; the central processing unit sends an acquisition command to the camera, which acquires the deformed fringes reflected from the ultrathin glass surface and transmits the deformed fringe image to the central processing unit; the central processing unit performs a phase-shifting algorithm on the deformed fringe image to calculate the wrapped phase; then performs an unwrapping algorithm on the wrapped phase to obtain the continuous phase; the central processing unit calculates the surface horizontal gradient and surface vertical gradient based on the mapping relationship between the continuous phase and the surface gradient; the central processing unit performs an integration algorithm on the surface horizontal gradient and surface vertical gradient, including Fourier integration or least squares integration, to reconstruct the surface height map.

2. The ultrathin glass defect detection system according to claim 1, characterized in that, The distortion-free phase deflection detection module includes a high-resolution industrial camera, a light source screen, and a beam splitter. This module transmits the acquired stripe images to a central processing unit (CPU), which reconstructs a surface height map based on the stripe images. The line scan camera module, in conjunction with the distortion-free phase deflection detection module, performs multi-dimensional imaging of defects in the ultra-thin glass. The CPU executes image registration and defect classification algorithms, combining defect location, shape, and grayscale features to obtain accurate defect location and classification results.

3. The ultrathin glass defect detection system according to claim 1, characterized in that, The high-magnification microscope module includes an objective lens, a Z-axis motorized platform, and an autofocus unit. The high-magnification microscope module is communicatively connected to the central processing unit.

4. A method for locating defects in ultrathin glass based on line scanning and phase deflection, characterized in that, The application of the ultrathin glass defect detection system according to any one of claims 1-3 includes the following steps: Step S1: The central processing unit sends an acquisition command to the line scan camera module. The line scan camera module scans the ultrathin glass sample and outputs a 2D grayscale image to the central processing unit. The central processing unit executes a defect detection algorithm on the 2D grayscale image and extracts a preliminary defect list. The preliminary defect list includes the center coordinates and defect outline of each defect. Step S2: The central processing unit sends an acquisition command to the distortion-free phase deflection detection module. The distortion-free phase deflection detection module scans the same ultrathin glass sample and outputs a surface height map to the central processing unit. Step S3: The central processing unit performs a spatial registration algorithm on the 2D grayscale image and the surface height map to establish a mapping relationship between the 2D grayscale image coordinate system and the surface height map coordinate system; Step S4: Based on the mapping relationship, the central processing unit maps each defect in the preliminary defect list to the surface height map, analyzes the height characteristics of the defect mapping area, classifies the defects, and distinguishes between interlayer defects and surface defects. Step S5: The central processing unit sends a positioning command to the high-magnification microscope module, controls the high-magnification microscope module to move to the location of the interlayer defect and perform automatic focusing to determine the depth layer where the interlayer defect is located.

5. The method for locating defects in ultrathin glass according to claim 4, characterized in that, In step S3, the central processing unit executes the spatial registration algorithm in at least one of the following ways: Method A: Gray-scale-based registration, specifically: the camera of the distortion-free phase deflection detection module synchronously acquires the intensity image of the ultrathin glass. The image is then transmitted to the central processing unit (CPU). The CPU constructs a transformation matrix M, which maps the coordinate system of the intensity image to the coordinate system of the 2D grayscale image. The CPU calculates the normalized cross-correlation value between the 2D grayscale image and the transformed intensity image. The CPU adjusts the parameters of the transformation matrix M through an optimization algorithm to maximize the normalized cross-correlation value, thus obtaining the optimal transformation matrix M1. Method B: Registration based on product boundaries, specifically: The CPU performs an edge detection algorithm on the 2D grayscale image to locate the ultra-thin glass... The four boundaries are calculated, and the intersection points of the four boundaries are obtained to obtain the corner point set of the line scan image. The central processing unit performs an edge detection algorithm on the surface height map to locate the four boundaries of the ultrathin glass and calculates the intersection points of the four boundaries to obtain the corner point set of the PMD image. The central processing unit uses the corner point set of the line scan image and the corner point set of the PMD image as matching features to construct an overdetermined system of equations. The central processing unit uses the least squares method to solve the overdetermined system of equations to obtain the affine transformation matrix M2, realizing the mapping from the 2D grayscale image coordinate system (x,y) to the surface height map coordinate system (u,v).

6. The method for locating defects in ultrathin glass according to claim 5, characterized in that, The specific process of the central processing unit executing the defect classification algorithm in step S4 is as follows: The central processing unit calls the transformation matrix M obtained in step S3 to map the contour Ci of each defect Di in the preliminary defect list to the surface height map HPmd(u,v), obtaining the corresponding mapped region Ri; the central processing unit calculates the local height difference of the mapped region Ri. The system performs curvature analysis and gradient continuity analysis algorithms on the mapped region Ri to extract curvature and gradient continuity features. The central processing unit presets a height threshold Th ≥ 1 μm and constructs classification criteria: if Δhi > Th and the curvature features show smooth changes and the gradient continuity features show continuous without abrupt changes, then the defect is determined to be an interlayer defect; if Δhi ≤ Th, or Δhi > Th but the curvature features show sharp abrupt changes and the gradient continuity features show discontinuities, then the defect is determined to be a surface defect.

7. The method for locating defects in ultrathin glass according to claim 6, characterized in that, In step S5, the specific process of the central processing unit controlling the high-magnification microscope module to perform depth positioning is as follows: The central processing unit extracts the center coordinates (xi, yi) of the interlayer defects from the preliminary defect list and sends a displacement command to the Z-axis motorized platform of the high-magnification microscope module, controlling the Z-axis motorized platform to move the objective lens above the (xi, yi) position; the central processing unit sends a scanning command to the Z-axis motorized platform, controlling the Z-axis motorized platform to scan uniformly with the objective lens within the preset Z-axis range. At each scanning position zk, the objective lens acquires a high-magnification image Ik and transmits it to the autofocus unit. The autofocus unit calls the built-in sharpness evaluation function to calculate the sharpness S(zk) of each high-magnification image Ik, and transmits the sharpness S(zk) to the central processing unit. The central processing unit executes a peak detection algorithm on the sharpness S(zk) to find the Z-axis position z_focus corresponding to the maximum sharpness. The central processing unit pre-stores the structural parameters of the ultrathin glass sample, including the thickness of the upper protective film, the thickness of the ultrathin glass, and the thickness of the lower protective film. By comparing the absolute coordinates of z_focus with the numerical range of the structural parameters, the depth level of the interlayer defect is determined.

8. The method for locating defects in ultrathin glass according to claim 7, characterized in that, The defect detection algorithms executed by the central processing unit include threshold segmentation algorithms, edge detection algorithms, or U-Net algorithms based on deep learning; it also includes preprocessing algorithms performed by the central processing unit on 2D grayscale images, the preprocessing algorithms including filtering algorithms and contrast enhancement algorithms, used to remove noise in 2D grayscale images and improve the recognizability of defect features.

9. The method for locating defects in ultrathin glass according to claim 8, characterized in that, The interlayer defects include foreign matter between the ultrathin glass and the upper protective film, foreign matter between the ultrathin glass and the lower protective film, and air bubbles between the ultrathin glass and the upper or lower protective film; the surface defects include dust on the surface of the upper protective film, dust on the surface of the lower protective film, and scratches on the surface of the upper or lower protective film.

Images