Crystal membrane screen defect detection method and system based on image recognition

Abstract

The application provides a kind of based on image recognition's diaphragm defect detection method and system, method includes collecting diaphragm image data, obtains long side edging image in image data;Sub-pixel level positioning is carried out to special-shaped Mark point;Target area and image data are subjected to first-level defect detection, and generate suspected defect candidate area graph;Suspected defect candidate area graph is classified to carry out second-level defect detection;Suspected defect candidate area graph and the classification detection result are fused by Bayes, and curvature compensation factor is introduced.The application solves the problems of deformation, positioning, adaptability, classification accuracy and insufficient fusion in the prior art, and significantly improves the detection accuracy and robustness.

Description

Technical Field

[0001] This invention belongs to the technical field of screen inspection, specifically relating to a method and system for detecting defects in crystal screens based on image recognition. Background Technology

[0002] Crystal film screen, also known as LED crystal film screen, is a new type of transparent display technology that combines extreme transparency with ultra-thin flexibility. Its transparency can reach over 95%, it is almost invisible when closed, and its thickness is only 1-2 millimeters, allowing it to be bent and rolled. It can be directly adhered to glass surfaces, supports on-site cutting and irregular shape installation, requires no complex steel structure, and features a 160° wide viewing angle and high brightness, enabling excellent dynamic display effects.

[0003] Currently, defect detection for crystal film screens mainly involves a combination of optics, electronics, and precision mechanics. Its core strategy lies in multi-dimensional screening to cover potential problems at different levels.

[0004] In the existing technology, the existing crystal screen defect detection technology still has many shortcomings in terms of image acquisition deformation correction, multi-level positioning accuracy, adaptive extraction of defect candidate regions, fine classification of small defects, and fusion output of multi-stage detection results. There is an urgent need for a crystal screen defect detection method that can comprehensively solve the above problems. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method and system for detecting defects in crystal screens based on image recognition, thereby overcoming the shortcomings of the prior art.

[0006] In a first aspect, the present invention provides a method for detecting defects in a crystal screen based on image recognition, the method comprising: Image data of the crystal screen is acquired, the long edge grinding image in the image data is obtained, and the long edge grinding image is segmented to obtain several local edge grinding images; Subpixel-level localization is performed on the irregular Mark points on several of the aforementioned local edge-grinding images to obtain the target area; The dynamic structural element morphological gray-level collaborative difference operator is used to perform first-level defect detection on the target region and the image data, and a confidence-weighted candidate region map of suspected defects is generated. A gradient-guided residual attention calibration network is used to classify the suspected defect candidate region map for second-level defect detection, so as to obtain a distinguishing detection result. A confidence backtracking and evidence fusion strategy is adopted to perform Bayesian fusion of the suspected defect candidate region map and the differentiation detection results, and a curvature compensation factor is introduced to output the final category, size and confidence of each defect region.

[0007] Compared with the prior art, the beneficial effects of the present invention are: effectively eliminating image deformation caused by motion speed fluctuations, achieving high-precision positioning of floating edge Mark points, compensating for glass elastic deformation errors, adaptively extracting candidate regions for chipped edges and missing parts, and reducing dirt interference; introducing a gradient-guided residual attention calibration network to significantly improve the classification accuracy of small chipped edges and dirt; and finally, through confidence fusion and curvature compensation, outputting reliable detection results and eliminating bending measurement errors.

[0008] Furthermore, the steps of acquiring image data from the crystal screen, obtaining a long-side edge-grinding image from the image data, and segmenting the long-side edge-grinding image to obtain several local edge-grinding images include: The image data of the crystal screen acquired by the multispectral linear array camera is acquired, the long side edge grinding image of the crystal screen acquired by the dual-sided dual-sided linear scanning method is acquired, and the line frequency of the multispectral linear array camera is dynamically adjusted by the dynamic line frequency speed adaptive synchronization algorithm. The long-side edge-grinding image is segmented using a frame-triggered segmented sampling strategy to obtain several local edge-grinding images, and these local edge-grinding images are stored.

[0009] Furthermore, the step of performing sub-pixel-level localization of the irregular Mark points on the plurality of local edge-grinding images includes: A gradient-guided pyramid frequency fusion locator is used to perform sub-pixel-level localization of irregular Mark points on the local edge-grinding image; The local edge-grinding images after positioning are corrected based on a rotation-translation affine decoupling matrix, wherein the expression of the rotation-translation affine decoupling matrix is: ; In the formula, Denotes the affine decoupling transformation matrix. This represents the angle between the line connecting Mark points and the horizontal axis. , These represent the center coordinates of the two Mark points respectively. The directional offset, the center coordinates of the two Mark points are in The offset in direction. This represents the tangential distortion compensation matrix.

[0010] Furthermore, the expression for the dynamic structuring element morphological gray-level collaborative difference operator is as follows: ; In the formula, This represents the result of the opening operation difference. , These represent the x-coordinate and y-coordinate of the pixel, respectively. Indicates the original edge-grinding image in grayscale value at that location This represents the image after opening using an adaptive structuring element. This represents an adaptive rectangular structure element. This represents the gradient magnitude at a pixel. This represents the gradient decay scale parameter.

[0011] Furthermore, before the step of classifying the suspected defect candidate region map using the gradient-guided residual attention calibration network for second-level defect detection, the method further includes: An improved pre-activated residual block is constructed, and a channel-space hybrid attention calibration module and a gradient magnitude gating mechanism are introduced to establish a gradient-guided residual attention calibration network. The expression for the gradient-guided residual attention calibration network is as follows: ; In the formula, Indicates the first Output feature maps of each residual block Indicates the first The input feature map of each residual block, Represents a nonlinear mapping function. Indicates the convolution kernel weights, This represents the convolution operation. Indicates batch normalization, This represents the linear rectification activation function.

[0012] Furthermore, the step of introducing the curvature compensation factor includes: Size evaluation of defect regions is performed based on pixel physical scale mapping; A curvature compensation factor is introduced into the size assessment to eliminate assessment errors caused by bending of the crystal screen.

[0013] Furthermore, the expression for the pixel physical scale mapping is: ; In the formula, This represents the actual physical size of the defect. This indicates the pixel size of the defect in the image. Represents pixel equivalent. This represents the curvature compensation coefficient. This represents the local curvature of the crystal film screen at the defect location. Indicates the distance to the reference plane.

[0014] Secondly, the present invention also provides a crystal screen defect detection system based on image recognition, the system comprising: The acquisition module is used to acquire image data of the crystal screen, obtain the long edge grinding image in the image data, and segment the long edge grinding image to obtain several local edge grinding images; The positioning module is used to perform sub-pixel-level positioning of irregular Mark points on several local edge-grinding images to obtain the target area; The detection generation module is used to perform first-level defect detection on the target region and the image data using a dynamic structural element morphological gray-level collaborative difference operator, and generate a confidence-weighted candidate region map of suspected defects. The classification and detection module is used to classify the suspected defect candidate region map based on the gradient-guided residual attention calibration network for second-level defect detection, so as to obtain a distinguishing detection result. The fusion output module is used to perform Bayesian fusion of the suspected defect candidate region map and the distinguishing detection results by adopting confidence backtracking and evidence fusion strategies, and introduces curvature compensation factor to output the final category, size and confidence of each defect region.

[0015] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described image recognition-based crystal screen defect detection method.

[0016] Fourthly, the present invention also provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described image recognition-based crystal screen defect detection method. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the image recognition-based crystal screen defect detection method in the first embodiment of the present invention; Figure 2 This is a structural block diagram of the image recognition-based crystal screen defect detection system according to the second embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of the electronic device in the third embodiment of the present invention.

[0019] Explanation of key component symbols: 10. Data Acquisition Module; 20. Positioning Module; 30. Detection and Generation Module; 40. Classification and Detection Module; 50. Fusion Output Module; 60. Bus; 61. Processor; 62. Memory; 63. Communication interface.

[0020] The embodiments of the present invention will be further described below with reference to the accompanying drawings. Detailed Implementation

[0021] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0022] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0024] Example 1 Please see Figure 1 The image shown illustrates a crystal screen defect detection method based on image recognition in the first embodiment of the present invention, the method comprising steps S1 to S5: S1, acquire image data of the crystal screen, obtain the long edge grinding image in the image data, and segment the long edge grinding image to obtain several local edge grinding images; Specifically, step S1 includes steps S11 to S12: S11, acquire image data of the crystal screen collected by the multispectral linear array camera, acquire the long side edge grinding image of the crystal screen collected by the dual-sided dual-sided linear scanning method, and dynamically adjust the line frequency of the multispectral linear array camera using a dynamic line frequency speed adaptive synchronization algorithm. It should be noted that four multispectral line scan cameras (model: ECO PH4KCL-200KT) were used in conjunction with a gantry direct-drive mechanism to acquire images of the upper and lower edges and the four long edges (front and back) of the crystal screen (i.e., LCD screen) using a dual-sided, dual-sided line scanning method. During the acquisition process, the line frequency of the line scan cameras was adjusted in real time using a dynamic line frequency speed adaptive synchronization algorithm to eliminate image stretching or compression deformation caused by fluctuations in motion speed. It should be explained that the dynamic line frequency speed adaptive synchronization algorithm dynamically adjusts the line frequency of the line scan cameras based on real-time speed and acceleration, and incorporates inertial compensation to solve the image stretching / compression problem during start-up and stop phases. The expression for the dynamic line frequency speed adaptive synchronization algorithm is as follows: ; In the formula, express The dynamic line frequency of the linear scan camera at any given time. Indicates time, express The instantaneous speed of the gantry mechanism. Indicates the pixel size of a line scan camera. Indicates the inertia compensation coefficient. It represents the acceleration of motion.

[0025] S12, the long edge grinding image is segmented using a frame-triggered segmented sampling strategy to obtain several local edge grinding images, and the several local edge grinding images are stored. It should be noted that a frame-triggered segmented sampling strategy is adopted: the line scan camera's line trigger mode is configured to trigger one frame every 20,000 lines, meaning the size of a single frame image is 20,000 pixels × 4096 pixels. For a 1420mm long edge grinding section, based on a pixel equivalent of 1.7μm / pixel, each edge grinding section of the crystal screen requires the acquisition of 42 local images, for a total of 168 local images for the four edges, which are stored in different cache areas of the industrial control computer's memory.

[0026] S2, perform sub-pixel-level positioning of irregular Mark points on several local edge-grinding images to obtain the target area; Specifically, step S2 includes steps S21 to S22: S21, a gradient-guided pyramid frequency fusion locator is used to perform sub-pixel-level positioning of the irregular Mark points on the local edge-grinding image; It should be noted that a gradient-guided pyramid frequency fusion locator is used to perform sub-pixel-level localization of irregularly shaped Mark points (cross-shaped Marks, approximately 2mm × 2mm in size) in local edge-grinding images. Specifically, a 5-layer image pyramid is first constructed. The top layer (layer 4) has a resolution of 1 / 16 of the original image, and an improved Canny edge detection method is used to extract the Mark point contours, achieving a coarse localization accuracy of ±0.1mm. Then, fine registration is performed at the bottom layer (original resolution) using a frequency domain phase correlation method. The cross-power spectrum of the two images is calculated using Fourier transform, and the peak position after the inverse transform is taken as the sub-pixel offset, achieving a localization accuracy of 0.05 pixels. It should be explained that the gradient-guided pyramid frequency fusion locator combines the coarse localization of the image pyramid with the fine registration of the frequency domain phase correlation, and adds a tangential distortion compensation matrix, making it suitable for locating Mark points on floating edges.

[0027] S22, Correct the local edge grinding images after positioning based on the rotation-translation-affine decoupling matrix; The expression for the rotation-translation-affine decoupling matrix is ​​as follows: ; In the formula, Denotes the affine decoupling transformation matrix. This represents the angle between the line connecting Mark points and the horizontal axis. , These represent the center coordinates of the two Mark points respectively. The directional offset, the center coordinates of the two Mark points are in The offset in direction. This represents the tangential distortion compensation matrix.

[0028] It should be noted that after correction by this matrix, the relative motion control axis of the edge grinding area is in a horizontal state, and the center point positioning accuracy reaches within 0.008mm.

[0029] S3, The dynamic structural element morphological gray-level collaborative difference operator is used to perform first-level defect detection on the target region and the image data, and a confidence-weighted candidate region map of suspected defects is generated. It should be noted that the dynamic structural element morphological grayscale collaborative differential operator is used to perform the first-level defect detection on the target area (i.e., the corrected edge grinding area). The test results show that the method achieves an accuracy of 98.2% in the preliminary separation of chipped edges and dirt.

[0030] It is worth noting that the dynamic structuring element morphological gray-scale collaborative difference operator makes the initial separation of edge collapse / dirt more accurate by dynamically changing the length of the structuring element with the local gray-scale variance or entropy and multiplying the difference result by the exponential decay of the gradient magnitude.

[0031] In this embodiment, the expression for the dynamic structuring element morphological gray-level collaborative difference operator is: ; In the formula, This represents the result of the opening operation difference. , These represent the x-coordinate and y-coordinate of the pixel, respectively. Indicates the original edge-grinding image in grayscale value at that location This represents the image after opening using an adaptive structuring element. This represents an adaptive rectangular structure element. This represents the gradient magnitude at a pixel. This represents the gradient decay scale parameter.

[0032] S4, The gradient-guided residual attention calibration network classifies the suspected defect candidate region map for second-level defect detection to obtain a distinguishing detection result; It should be noted that the gradient-guided residual attention calibration network, based on the improved pre-activated residual block, introduces channel-spatial hybrid attention and gradient magnitude gating to enhance the feature response to small edge collapse regions.

[0033] Specifically, step S4 further includes step S41: S41, improve the pre-activated residual block, and introduce a channel-space hybrid attention calibration module and a gradient magnitude gating mechanism to establish a gradient-guided residual attention calibration network, wherein the expression of the gradient-guided residual attention calibration network is: ; In the formula, Indicates the first Output feature maps of each residual block Indicates the first The input feature map of each residual block, Represents a nonlinear mapping function. Indicates the convolution kernel weights, This represents the convolution operation. Indicates batch normalization, This represents the linear rectification activation function.

[0034] It should be noted that a gradient-guided residual attention calibration network was established. This network is based on an improved ResNet50. The network was trained using a dataset containing three classes of samples: broken edges (455 images), dirty images (475 images), and normal edge smoothing (576 images). An asymmetric learning rate was used during training (0.0001 for the backbone network and 0.001 for new modules), with 100 iterations. Test results show that the network achieves a classification accuracy of 98.7% for broken edges and dirty images.

[0035] S5, using a confidence backtracking and evidence fusion strategy, the suspected defect candidate region map and the distinguishing detection results are fused using Bayesian method, and a curvature compensation factor is introduced to output the final category, size and confidence of each defect region. It should be noted that the confidence backtracking and evidence fusion strategy combines the candidate region map of suspected defects with the differentiation detection results using Bayesian fusion to obtain the final confidence score. The final output includes the category (edge ​​chipping, missing parts, dirt, cracks, or insufficient width), size (unit: μm), and fused confidence score (range 0~1) for each defect region. It is worth noting that the confidence backtracking and evidence fusion strategy combines the candidate region confidence score output by the traditional vision algorithm with the deep learning classification confidence score using Bayesian fusion to output the final judgment and uncertainty estimate.

[0036] Specifically, step S5 includes steps S51 to S52: S51, size evaluation of defect regions based on pixel physical scale mapping; S52, In the size evaluation, a curvature compensation factor is introduced to eliminate the evaluation error caused by the bending of the crystal screen; In this embodiment, the expression for the pixel physical scale mapping is: ; In the formula, This represents the actual physical size of the defect. This indicates the pixel size of the defect in the image. Represents pixel equivalent. This represents the curvature compensation coefficient. This represents the local curvature of the crystal film screen at the defect location. Indicates the distance to the reference plane.

[0037] In summary, the image recognition-based crystal screen defect detection method in the above embodiments of the present invention effectively eliminates image deformation caused by motion speed fluctuations through a dynamic line frequency-speed adaptive synchronization algorithm, ensuring the quality of acquired images; it employs pyramid-frequency fusion positioning and affine decoupling matrices to achieve high-precision positioning of floating edge Mark points, compensating for glass elastic deformation errors; it utilizes a dynamic structural element morphology-grayscale collaborative differential operator to adaptively extract candidate regions for chipped edges and defects, reducing dirt interference; it introduces a gradient-guided residual attention calibration network to significantly improve the classification accuracy of minor chipped edges and dirt; and finally, through confidence fusion and curvature compensation, it outputs reliable detection results and eliminates bending measurement errors. The present invention comprehensively solves the problems of deformation, positioning, adaptability, classification accuracy, and insufficient fusion in existing technologies, significantly improving detection accuracy and robustness.

[0038] Example 2 Please see Figure 2 The image shown is a crystal screen defect detection system based on image recognition according to a second embodiment of the present invention. The system includes: The acquisition module 10 is used to acquire image data of the crystal screen, acquire the long edge grinding image in the image data, and segment the long edge grinding image to obtain several local edge grinding images; The positioning module 20 is used to perform sub-pixel-level positioning of irregular Mark points on several local edge-grinding images to obtain the target area; The detection generation module 30 is used to perform first-level defect detection on the target region and the image data using a dynamic structuring element morphological gray-level collaborative difference operator, and to generate a confidence-weighted candidate region map of suspected defects. The expression for the dynamic structuring element morphological gray-level collaborative difference operator is as follows: ; In the formula, This represents the result of the opening operation difference. , These represent the x-coordinate and y-coordinate of the pixel, respectively. Indicates the original edge-grinding image in grayscale value at that location This represents the image after opening using an adaptive structuring element. This represents an adaptive rectangular structure element. This represents the gradient magnitude at a pixel. This represents the gradient decay scale parameter; The classification and detection module 40 is used to classify the suspected defect candidate region map based on the gradient-guided residual attention calibration network for second-level defect detection, so as to obtain a distinguishing detection result. The fusion output module 50 is used to perform Bayesian fusion of the suspected defect candidate region map and the distinguishing detection results by adopting a confidence backtracking and evidence fusion strategy, and to introduce a curvature compensation factor to output the final category, size and confidence of each defect region.

[0039] In some optional embodiments, the acquisition module 10 includes: An adjustment unit is used to acquire image data of the crystal screen collected by a multispectral linear array camera, acquire the long-side edge-grinding image of the crystal screen collected by a dual-sided dual-sided linear scanning method, and dynamically adjust the line frequency of the multispectral linear array camera using a dynamic line frequency speed adaptive synchronization algorithm. The segmented storage unit is used to segment the long-side edge-grinding image using a frame-triggered segmented sampling strategy to obtain several local edge-grinding images, and to store several of the local edge-grinding images.

[0040] In some alternative embodiments, the positioning module 20 includes: The positioning unit is used to perform sub-pixel-level positioning of irregular Mark points on the local edge-grinding image using a gradient-guided pyramid frequency fusion positioner. A correction unit is used to correct several local edge-grinding images after positioning based on a rotation-translation affine decoupling matrix, wherein the expression of the rotation-translation affine decoupling matrix is: ; In the formula, Denotes the affine decoupling transformation matrix. This represents the angle between the line connecting Mark points and the horizontal axis. , These represent the center coordinates of the two Mark points respectively. The directional offset, the center coordinates of the two Mark points are in The offset in direction. This represents the tangential distortion compensation matrix.

[0041] In some alternative embodiments, the classification detection module 40 includes: A unit is established to improve the pre-activated residual block, and a channel-space hybrid attention calibration module and a gradient magnitude gating mechanism are introduced to establish a gradient-guided residual attention calibration network. The expression for the gradient-guided residual attention calibration network is as follows: ; In the formula, Indicates the first Output feature maps of each residual block Indicates the first The input feature map of each residual block, Represents a nonlinear mapping function. Indicates the convolution kernel weights, This represents the convolution operation. Indicates batch normalization, This represents the linear rectification activation function.

[0042] In some alternative embodiments, the fusion output module 50 includes: An evaluation unit is used to evaluate the size of the defect region based on a pixel physical scale mapping, wherein the expression for the pixel physical scale mapping is: ; In the formula, This represents the actual physical size of the defect. This indicates the pixel size of the defect in the image. Represents pixel equivalent. This represents the curvature compensation coefficient. This represents the local curvature of the crystal film screen at the defect location. Indicates the distance to the reference plane; An introduction unit is used to introduce a curvature compensation factor in the size assessment to eliminate assessment errors caused by bending of the crystal screen.

[0043] The functions or operation steps implemented by the above modules and units are largely the same as those in the above method embodiments, and will not be repeated here.

[0044] The image recognition-based crystal screen defect detection system provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the system embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0045] Example 3 The third embodiment of the present invention also proposes an electronic device, please refer to [link / reference]. Figure 3 The image shows an electronic device according to a third embodiment of the present invention.

[0046] The electronic device may include a processor 61 and a memory 62 storing computer program instructions.

[0047] Specifically, the processor 61 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the present application.

[0048] The memory 62 may include a large-capacity storage device for data or instructions. For example, and not limitingly, the memory 62 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 62 may include removable or non-removable (or fixed) media. Where appropriate, the memory 62 may be internal or external to a data processing device. In a particular embodiment, the memory 62 is non-volatile memory. In a particular embodiment, the memory 62 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random-Access Memory (FPMDRAM), Extended Data Out Dynamic Random-Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0049] The memory 62 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 61.

[0050] The processor 61 reads and executes the computer program instructions stored in the memory 62 to implement the image recognition-based crystal screen defect detection method of the above embodiment 1.

[0051] In some embodiments, the electronic device may further include a communication interface 63 and a bus 60. For example, Figure 3 As shown, the processor 61, memory 62, and communication interface 63 are connected through bus 60 and complete communication with each other.

[0052] The communication interface 63 is used to enable communication between the various modules, devices, units, and / or equipment in this application. The communication interface 63 can also enable data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0053] Bus 60 includes hardware, software, or both, that couples components of a device together. Bus 60 includes, but is not limited to, at least one of the following: data bus, address bus, control bus, expansion bus, and local bus. For example, and not as a limitation, bus 60 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 60 may include one or more buses. Although this application describes and illustrates a specific bus, this application considers any suitable bus or interconnection.

[0054] The electronic device can acquire an image recognition-based crystal screen defect detection system and execute the image recognition-based crystal screen defect detection method of this embodiment.

[0055] Furthermore, in conjunction with the image recognition-based crystal screen defect detection method in Embodiment 1 above, this application can provide a storage medium for implementation. This storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement the image recognition-based crystal screen defect detection method of Embodiment 1 above.

[0056] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0057] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A method for detecting defects in a crystal film screen based on image recognition, characterized in that, The method includes: Image data of the crystal screen is acquired, the long edge grinding image in the image data is obtained, and the long edge grinding image is segmented to obtain several local edge grinding images; Subpixel-level localization is performed on the irregular Mark points on several of the aforementioned local edge-grinding images to obtain the target area; The dynamic structural element morphological gray-level collaborative difference operator is used to perform first-level defect detection on the target region and the image data, and a confidence-weighted candidate region map of suspected defects is generated. A gradient-guided residual attention calibration network is used to classify the suspected defect candidate region map for second-level defect detection, so as to obtain a distinguishing detection result. A confidence backtracking and evidence fusion strategy is adopted to perform Bayesian fusion of the suspected defect candidate region map and the differentiation detection results, and a curvature compensation factor is introduced to output the final category, size and confidence of each defect region.

2. The image recognition-based defect detection method for crystal film screens according to claim 1, characterized in that, The steps of acquiring image data from the crystal screen, obtaining the long-side edge-grinding image from the image data, and segmenting the long-side edge-grinding image to obtain several local edge-grinding images include: The image data of the crystal screen acquired by the multispectral linear array camera is acquired, the long side edge grinding image of the crystal screen acquired by the dual-sided dual-sided linear scanning method is acquired, and the line frequency of the multispectral linear array camera is dynamically adjusted by the dynamic line frequency speed adaptive synchronization algorithm. The long-side edge-grinding image is segmented using a frame-triggered segmented sampling strategy to obtain several local edge-grinding images, and these local edge-grinding images are stored.

3. The image recognition-based defect detection method for crystal film screens according to claim 1, characterized in that, The step of performing sub-pixel-level localization of irregular Mark points on several local edge-grinding images includes: A gradient-guided pyramid frequency fusion locator is used to perform sub-pixel-level localization of irregular Mark points on the local edge-grinding image; The local edge-grinding images after positioning are corrected based on a rotation-translation affine decoupling matrix, wherein the expression of the rotation-translation affine decoupling matrix is: ； In the formula, Denotes the affine decoupling transformation matrix. This represents the angle between the line connecting Mark points and the horizontal axis. , These represent the center coordinates of the two Mark points respectively. The directional offset, the center coordinates of the two Mark points are in The offset in direction. This represents the tangential distortion compensation matrix.

4. The image recognition-based defect detection method for crystal film screens according to claim 1, characterized in that, The expression for the dynamic structural element morphological gray-level collaborative difference operator is as follows: ； In the formula, This represents the result of the opening operation difference. , These represent the x-coordinate and y-coordinate of the pixel, respectively. Indicates the original edge-grinding image in grayscale value at that location This represents the image after opening using an adaptive structuring element. This represents an adaptive rectangular structure element. This represents the gradient magnitude at a pixel. This represents the gradient decay scale parameter.

5. The image recognition-based defect detection method for crystal film screens according to claim 1, characterized in that, Before the step of classifying the suspected defect candidate region map by the gradient-guided residual attention calibration network for second-level defect detection, the method further includes: An improved pre-activated residual block is constructed, and a channel-space hybrid attention calibration module and a gradient magnitude gating mechanism are introduced to establish a gradient-guided residual attention calibration network. The expression for the gradient-guided residual attention calibration network is as follows: ； In the formula, Indicates the first Output feature maps of each residual block Indicates the first The input feature map of each residual block, Represents a nonlinear mapping function. Indicates the convolution kernel weights, This represents the convolution operation. Indicates batch normalization, This represents the linear rectification activation function.

6. The image recognition-based defect detection method for crystal film screens according to claim 1, characterized in that, The step of introducing the curvature compensation factor includes: Size evaluation of defect regions is performed based on pixel physical scale mapping; A curvature compensation factor is introduced into the size assessment to eliminate assessment errors caused by bending of the crystal screen.

7. The image recognition-based defect detection method for crystal film screens according to claim 6, characterized in that, The expression for the pixel physical scale mapping is: ； In the formula, This represents the actual physical size of the defect. This indicates the pixel size of the defect in the image. Represents pixel equivalent. This represents the curvature compensation coefficient. This represents the local curvature of the crystal film screen at the defect location. Indicates the distance to the reference plane.

8. A defect detection system for a crystal film screen based on image recognition, characterized in that, The system includes: The acquisition module is used to acquire image data of the crystal screen, obtain the long edge grinding image in the image data, and segment the long edge grinding image to obtain several local edge grinding images; The positioning module is used to perform sub-pixel-level positioning of irregular Mark points on several local edge-grinding images to obtain the target area; The detection generation module is used to perform first-level defect detection on the target region and the image data using a dynamic structural element morphological gray-level collaborative difference operator, and generate a confidence-weighted candidate region map of suspected defects. The classification and detection module is used to classify the suspected defect candidate region map based on the gradient-guided residual attention calibration network for second-level defect detection, so as to obtain a distinguishing detection result. The fusion output module is used to perform Bayesian fusion of the suspected defect candidate region map and the distinguishing detection results by adopting confidence backtracking and evidence fusion strategies, and introduces curvature compensation factor to output the final category, size and confidence of each defect region.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the image recognition-based crystal screen defect detection method as described in any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the image recognition-based crystal screen defect detection method as described in any one of claims 1 to 7.

Images