Intelligent detection method and analysis system for defects of liquid crystal display screen based on visual perception

Abstract

The invention relates to the technical field of liquid crystal display screen detection, in particular to a liquid crystal display screen defect intelligent detection method and analysis system based on visual perception, and the method comprises the following steps: obtaining multi-angle image data of a liquid crystal display screen; generating a fused image based on the multi-angle image data; extracting defect features according to the fused image; determining a defect category based on the defect features; according to the method, multi-angle image acquisition and an innovative image fusion algorithm are adopted, so that the visibility of the defects and the comprehensiveness of detection are greatly improved. According to the method, different types of defects can be effectively captured, especially small defects which are difficult to observe at a single angle, and the omission ratio is remarkably reduced. By fusing multi-angle information, a more comprehensive and clear defect image can be generated, and high-quality input data is provided for subsequent analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of liquid crystal display screen detection, and more specifically, to an intelligent defect detection method and analysis system for liquid crystal display screens based on visual perception. Background Art

[0002] With the continuous development of liquid crystal display technology, the resolution and size of display screens have been increasing, and consumers' requirements for display quality have also been rising day by day. In the production process of liquid crystal display screens, defect detection is a key link to ensure product quality. Traditional defect detection methods mainly rely on manual visual inspection. This method not only has low efficiency but is also easily affected by human factors, making it difficult to ensure the consistency and accuracy of detection.

[0003] In recent years, with the progress of computer vision technology, automated defect detection systems have gradually been applied. Currently, the commonly used detection method in the industry is based on single-angle image acquisition and traditional image processing algorithms. This method usually uses a high-resolution camera to take pictures of the liquid crystal display screen from a fixed angle, and then uses algorithms such as edge detection and threshold segmentation to identify defects. Although it has made great progress compared with manual detection, this method still has some obvious deficiencies.

[0004] First of all, single-angle image acquisition is difficult to comprehensively capture all types of defects. Some defects, such as tiny surface scratches or bubbles, may be difficult to observe at a specific angle, resulting in a relatively high miss rate. Secondly, traditional image processing algorithms perform poorly when dealing with complex backgrounds and different types of defects, and are prone to false alarms or missed detections. In addition, with the continuous increase in the resolution of display screens, the amount of data that the detection system needs to process has increased sharply, and the processing speed of traditional algorithms is difficult to meet the requirements of real-time detection.

[0005] Some improved methods attempt to use multi-angle image acquisition to increase the visibility of defects, but still lack effective algorithms in image fusion and feature extraction. Other methods introduce machine learning techniques, such as support vector machines (SVM) or convolutional neural networks (CNN), to improve the accuracy of defect recognition. However, these methods often require a large amount of labeled data for training and perform poorly when dealing with new or rare defects.

[0006] Generally speaking, the existing liquid crystal display screen defect detection technologies still have significant deficiencies in terms of detection accuracy, efficiency, and adaptability, and are difficult to meet the quality control requirements of current high-resolution and large-size display screen production. Therefore, it is of great practical significance to develop a new method that can comprehensively, accurately, and efficiently detect defects in liquid crystal display screens. Summary of the Invention

[0007] The present invention aims to solve the problems existing in the prior art of defect detection for liquid crystal display screens, and provides an intelligent defect detection method and analysis system for liquid crystal display screens based on visual perception. Through innovative multi-angle image fusion technology, efficient feature extraction algorithms, and advanced classification methods, this method realizes comprehensive, accurate, and efficient detection of defects in liquid crystal display screens.

[0008] The present invention provides an intelligent defect detection method for liquid crystal display screens based on visual perception, including:

[0009] An acquisition step, including:

[0010] Acquiring multi-angle image data of the liquid crystal display screen;

[0011] A processing step, including:

[0012] Generating a fused image based on the multi-angle image data;

[0013] Extracting defect features according to the fused image;

[0014] Determining the defect category based on the defect features;

[0015] An output step, including:

[0016] Outputting the defect category and defect location information.

[0017] Preferably, the acquisition of the multi-angle image data of the liquid crystal display screen specifically includes:

[0018] Illuminating the liquid crystal display screen with a multi-angle lighting device;

[0019] Obtaining images of the liquid crystal display screen at different lighting angles through an image acquisition device.

[0020] Preferably, the generation of the fused image specifically includes:

[0021] Preprocessing the multi-angle image data to obtain preprocessed image data;

[0022] Fusing the preprocessed image data based on random matrix theory to obtain the fused image.

[0023] Preferably, the preprocessing of the multi-angle image data includes:

[0024] Removing noise from the multi-angle image data;

[0025] Performing geometric correction on the denoised image data.

[0026] Preferably, the extraction of the defect features specifically includes:

[0027] Apply the Laplace operator and Bessel function to the fused image to obtain an enhanced image;

[0028] Extract defect features from the enhanced image based on group theory and discrete mathematics methods.

[0029] Preferably, the extracting defect features from the enhanced image includes:

[0030] Perform a discrete Fourier transform on the enhanced image;

[0031] Multiply the transformation result by the generator matrix of the cyclic group to obtain a feature matrix.

[0032] Preferably, the determining the defect category specifically includes:

[0033] Perform a non-linear mapping on the defect features to obtain the mapped features;

[0034] Classify the mapped features based on quantum probability theory to obtain the defect category.

[0035] Preferably, the performing a non-linear mapping on the defect features includes:

[0036] Input the defect features into a chaotic system;

[0037] Transform the features through a Logistic mapping and a hyperbolic tangent function.

[0038] Preferably, it further includes:

[0039] Calculate the defect area and aspect ratio based on the defect category and defect location information;

[0040] Classify the defects according to a preset threshold.

[0041] An intelligent analysis system for liquid crystal display defects based on visual perception that executes the method includes:

[0042] An image acquisition module for acquiring multi-angle image data of a liquid crystal display;

[0043] An image processing module for:

[0044] Generate a fused image based on the multi-angle image data;

[0045] Extract defect features according to the fused image;

[0046] Determine the defect category based on the defect features;

[0047] A result output module for outputting the defect category and defect location information.

[0048] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0049] Firstly, the present invention adopts multi-angle image acquisition and an innovative image fusion algorithm, greatly improving the visibility of defects and the comprehensiveness of detection. This method can effectively capture different types of defects, especially those tiny defects that are difficult to observe from a single angle, significantly reducing the missed detection rate. By fusing multi-angle information, the present invention can generate a more comprehensive and clear defect image, providing high-quality input data for subsequent analysis.

[0050] Secondly, the present invention introduces an innovative feature extraction method based on topology and group theory, greatly improving the expression ability of defect features. This method can not only effectively extract information such as the shape and texture of defects, but also capture the topological relationship between defects and the background, making defect features more prominent and easy to identify. This innovation greatly improves the accuracy of subsequent classification, especially for defect detection in complex backgrounds.

[0051] Thirdly, the present invention adopts a classification method based on quantum probability theory, significantly improving the accuracy and robustness of defect recognition. Compared with traditional classification methods, this method can better handle the complex relationships between features and also shows good generalization ability for new or rare defects. This not only improves the detection accuracy but also reduces the false alarm rate, greatly reducing unnecessary rework and scrap.

[0052] Fourthly, while ensuring high detection accuracy, the method of the present invention can also maintain a relatively fast processing speed. By optimizing the algorithm design and utilizing parallel computing technology, the present invention can achieve real-time defect detection, meeting the high-efficiency requirements of modern liquid crystal display production lines. This high efficiency not only improves production efficiency but also reduces detection costs.

[0053] Finally, the method of the present invention has strong adaptability and scalability. It can not only adapt to different types and sizes of liquid crystal displays but also can be adjusted by simple parameters to deal with new types of defects. This flexibility enables the present invention to continuously meet the needs of evolving liquid crystal display technologies, providing long-term technical support for enterprises.

[0054] Generally speaking, through innovative technical solutions, the present invention effectively solves the problems existing in the existing liquid crystal display defect detection technology, achieving an overall improvement in detection accuracy, efficiency, and adaptability. This not only helps to improve product quality and reduce production costs but also provides a powerful quality control tool for liquid crystal display manufacturing enterprises, and is expected to promote the technological progress and quality improvement of the entire industry. Brief Description of the Drawings

[0055] Figure 1This is the flowchart of the method of the present invention.

[0056] Figure 2 This is the flowchart of the defect classification of the present invention.

[0057] Figure 3 This is the flowchart of the defect level division of the present invention. Detailed implementation manners

[0058] Please refer to Figures 1-3 The present invention provides an intelligent defect detection method and analysis system for liquid crystal display screens based on visual perception. The method includes an acquisition step, a processing step, and an output step.

[0059] In the acquisition step, the present invention realizes the comprehensive capture of defects by acquiring multi-angle image data of the liquid crystal display screen. Specifically, a multi-angle lighting device can be used to illuminate the liquid crystal display screen, and an image acquisition device is used to acquire images of the liquid crystal display screen at different lighting angles. Preferably, the present invention adopts a ring light source design, and 4-8 lighting points are set around the liquid crystal display screen to ensure that defects can be fully revealed at different angles. For example, lighting points can be selected at four angles of 0°, 45°, 90°, and 135°, so as to effectively capture different types of defects such as surface scratches and bubbles.

[0060] In the LCD manufacturing process, different types of defects such as bubbles, scratches, or color unevenness may occur. Since these defects may be more obvious at different lighting angles, taking images of the same area from multiple angles can provide more comprehensive information.

[0061] Suppose we are checking an LCD panel for tiny bubble defects. Using a multi-angle lighting device 11, we can take images of the area from four different angles (top, bottom, left, right). Each image may show different characteristics of the bubbles. For example, under side lighting, shadows may appear around the bubbles, while under top lighting, the bubbles may appear as bright spots.

[0062] In the defect detection of liquid crystal display screens (LCDs), obtaining images from multiple angles can provide more comprehensive defect information.

[0063]

[0064] Among them, I k is the image matrix at the k-th angle, R k is a random orthogonal matrix representing angle transformation, M k is the original image matrix, N k is the noise matrix, and ° represents the Hadamard product (element-wise product).

[0065] However, images taken from different angles may have noise and perspective differences, so an effective method is needed to fuse these images.

[0066] In the processing steps, the present invention first generates a fused image based on the acquired multi-angle image data. The core of this step lies in how to effectively integrate information from different angles.

[0067] The present invention adopts an innovative algorithm based on random matrix theory to achieve image fusion. Specifically, the fusion process can be expressed as:

[0068]

[0069] where F is the fused image matrix, W k is the weight matrix, erf is the error function for non-linear transformation, and I k is the image matrix at the k-th angle. This method can effectively retain the key information of the images at each angle while suppressing noise interference.

[0070] Suppose a defect suspected to be a bubble is found on an LCD panel. Multi-angle images of this area are acquired through the multi-angle illumination device 11, and these images are captured using the high-resolution industrial camera 12. After the image preprocessing unit 21 performs noise reduction on each image, the image fusion unit 22 fuses these images into a comprehensive image using the above formula. The fused image contains the key information at each angle and can more comprehensively reflect the defect. Introducing the error function erf enhances the contrast of the image and helps to highlight small defects.

[0071] These images are fused into a comprehensive image F, thus better capturing the presence and location of the bubble. After obtaining the fused image, it is necessary to further enhance the potential defect features in the image for subsequent processing.

[0072] Next, the present invention extracts defect features based on the fused image. This step uses a topology method for defect enhancement and can be expressed as:

[0073]

[0074] where E is the enhanced image, represents applying the Laplace operator to the image F (note: here actually represents the operation of processing the image through Laplace transform, not directly the Laplace operator itself); represents the Laplacian (i.e., second derivative) of the image F, used to detect rapidly changing regions in the image, such as edges or defects; Bessel 1 is the Bessel function of the first kind, is the Laplacian operator. This method can effectively highlight the edge and texture information in the image, making the defect features more obvious.

[0075] Continuing with the above-mentioned bubble defect as an example, assume that the fused image F has been obtained. Next, the feature extraction unit 23 uses the Laplacian operator to calculate the edge information of the image and uses the Bessel function to further enhance these edge features. The Laplacian operator can effectively detect the edge information in the image, while the Bessel function enhances these edge features, making the defects more obvious. Automatically enhancing the defect features through a mathematical model reduces manual intervention and improves the processing efficiency. Continuing with the above-mentioned bubble as an example, the topological method is used for defect enhancement to make it more clearly visible. In this way, even tiny bubbles can be recognized in the enhanced image E.

[0076] For the enhanced image E, the discrete Fourier transform (DFT) is used to transform it into the frequency domain, and the generator matrix G of the cyclic group is used to extract features:

[0077] Φ = DFT(E)·G,

[0078] Assume that the goal is to identify periodic defect patterns, such as regularly arranged ring pixels. DFT can effectively capture these periodic structures, while the generator matrix G of the cyclic group helps to retain these features, making subsequent classification more accurate.

[0079] To improve the discrimination between features, a non-linear mapping needs to be performed. Based on the previously extracted feature matrix Φ, the hyperbolic tangent function tanh and the Logistic mapping are used for non-linear transformation:

[0080] Ψ = tanh(Φ + λ·Logistic(Φ)),

[0081] This step can help highlight the differences between different defect types. For example, slight scratches and severe scratches may not be very obvious in the original feature matrix Φ, but after non-linear mapping, their representations in the feature matrix Ψ will become more distinguishable.

[0082] The extracted defect features need to be further transformed into data that can be used for classification. After feature extraction is completed, the present invention determines the defect category based on the extracted defect features. This step adopts a classification method based on quantum probability theory, which can be expressed as:

[0083]

[0084] where P(c|Ψ) is the probability of defect category c given the feature Ψ, H cLet \(H_c\) be the Hermitian matrix representing the category \(c\), and \(Tr\) represents the trace of the matrix. This method takes into account the quantum coherence between features and can capture the complex relationships between defect features more accurately.

[0085] Assume that the enhanced image \(E\) has been obtained. Next, the feature extraction unit 23 performs a discrete Fourier transform (DFT) on the image to convert the information in the spatial domain to the frequency domain. Then, by multiplying with the generator matrix \(G\) of the cyclic group, the feature matrix \(\varPhi\) is obtained. DFT can capture the periodic structures in the image, which is particularly effective for detecting regular defects (such as ring pixels). Through mathematical transformation, useful features are quickly extracted for subsequent processing.

[0086] In the output step, the present invention outputs the determined defect category and defect location information. These information can be directly used for quality control on the production line to help operators quickly locate and handle defects.

[0087] In one embodiment of the present invention, the step of acquiring multi - angle image data of the liquid crystal display screen further includes pre - processing the acquired images. This pre - processing process includes noise removal and geometric correction. Noise removal can be performed using methods such as median filtering or Gaussian filtering to eliminate random noise in the image. Geometric correction is to eliminate image distortion caused by the shooting angle and can be performed using methods such as affine transformation or perspective transformation. These pre - processing steps can significantly improve the accuracy of subsequent analysis.

[0088] In the step of generating the fused image, the present invention adopts an innovative algorithm based on random matrix theory. The advantage of this method is that it can effectively process the image information from different angles, taking into account the correlation and complementarity between images. For example, for scratch defects on the liquid crystal display screen, images from certain angles may show more clearly, while images from other angles may be less obvious. Through this fusion method, the information from each angle can be comprehensively utilized to obtain a more comprehensive and clearer defect image.

[0089] This method and system of the present invention have significant advantages in the field of liquid crystal display screen defect detection. First, multi - angle image acquisition can comprehensively capture various types of defects, greatly reducing the missed detection rate. Second, the image fusion algorithm based on random matrix theory and the defect enhancement method based on topology can effectively extract defect features and improve the accuracy of detection. Finally, using quantum probability theory for defect classification takes into account the complex relationships between features and further improves the accuracy of classification.

[0090] Generally speaking, the method and system provided by the present invention can significantly improve the efficiency and accuracy of liquid crystal display screen defect detection, which is of great significance for improving product quality and reducing production costs.

[0091] In a preferred embodiment of the present invention, the preprocessing of multi-angle image data includes two key steps: noise removal and geometric correction. Noise removal is a fundamental but extremely important step in image processing. The present invention adopts an adaptive median filtering algorithm to achieve this purpose. The advantage of this algorithm is that it can automatically adjust the size of the filtering window according to the local characteristics of the image, thereby removing noise effectively while retaining image details to the greatest extent.

[0092] Specifically, the adaptive median filtering algorithm can be expressed as:

[0093]

[0094] Where y(i,j) is the filtered pixel value, x(i,j) is the original pixel value, W ij is the filtering window centered at (i,j), and T is the threshold. Through a large number of experiments, the present invention determines that when T is set to 1.5 times the local standard deviation, the best denoising effect can be achieved.

[0095] Suppose there is a small scratch on an LCD panel. Due to possible noise during shooting, the scratch feature is not obvious. Through the adaptive median filtering algorithm, the noise can be removed while retaining the detailed features of the scratch.

[0096] In terms of geometric correction, the present invention adopts a perspective transformation method based on control points. First, mark points are set at the four corners of the liquid crystal display screen as control points. Then, the transformation matrix is calculated through the following formula:

[0097]

[0098] Where (x,y) is the original coordinate, (x′,y′) is the corrected coordinate, and h ij is an element of the transformation matrix. By solving this system of equations, the optimal transformation matrix can be obtained, thereby realizing the geometric correction of the image.

[0099] Suppose a defect suspected to be a bubble is found on an LCD panel. Due to the shooting angle, the bubble may appear deformed or in an inaccurate position. Through geometric correction, this distortion can be corrected to make the position of the bubble more accurate.

[0100] In another embodiment of the present invention, the steps of extracting defect features are further refined. First, the Laplace operator and Bessel function are applied to the fused image to obtain an enhanced image. The Laplace operator can effectively highlight the edge information in the image, while the Bessel function can further enhance these edge features. This process can be expressed as:

[0101]

[0102] Among them, E is the enhanced image, is the Laplacian operator, F is the fused image, J 1 is the first kind of Bessel function, α is the scaling factor, and r is the distance from the pixel to the center of the image. By adjusting the value of α, the degree of enhancement can be controlled. Through a large number of experiments, the present invention has found that when α takes values between 2 and 3, the best enhancement effect can be obtained.

[0103] Next, based on group theory and discrete mathematics methods, the present invention extracts defect features from the enhanced image. Specifically, first, the enhanced image is subjected to discrete Fourier transform:

[0104]

[0105] where F(u, v) is the frequency-domain representation, E(x, y) is the spatial-domain representation, and M and N are the width and height of the image respectively.

[0106] Then, the transformation result is multiplied by the generator matrix of the cyclic group to obtain the feature matrix:

[0107] Φ = F·G,

[0108] where Φ is the feature matrix and G is the generator matrix of the cyclic group. This method can effectively capture the periodic structure in the image, which is particularly effective for detecting regular defects (such as bad pixels) on the liquid crystal display screen.

[0109] In order to improve the discrimination between features, a non-linear mapping needs to be performed. In the step of determining the defect category, the present invention first performs a non-linear mapping on the defect features. This step uses the Logistic mapping in chaos theory and combines the hyperbolic tangent function for feature transformation:

[0110] Ψ = tanh(Φ + λ·(4x(1 - x))),

[0111] where Ψ is the mapped feature, Φ is the original feature, λ is the control parameter, and x is the initial value of the Logistic mapping. This non-linear mapping can enhance the difference between features, which is beneficial for subsequent classification.

[0112] Assume that the feature matrix Φ has been obtained. Next, the defect classification unit 24 uses the hyperbolic tangent function tanh and the Logistic mapping to perform non-linear transformation on the features to obtain the mapped feature matrix Ψ.

[0113] The non-linear mapping enhances the difference between features, which is helpful for subsequent classification. The control parameter λ can be adjusted according to actual needs to adapt to different application scenarios.

[0114] When determining the defect category, the classification method based on quantum probability theory can better capture the complex relationships between features, thereby improving the classification accuracy.

[0115] Finally, the present invention classifies the mapped features based on quantum probability theory. This method takes into account the quantum coherence between features and can more accurately capture the complex relationships between defect features. The classification process can be expressed as:

[0116]

[0117] where P(c|Ψ) is the probability of defect category c given feature Ψ, H c is the Hermitian matrix of a category c, and Tr represents the trace of the matrix.

[0118] The last step is to classify the defects according to the extracted features. Assume that the feature matrix Ψ of the defects has been obtained. Next, the defect classification unit 24 calculates the probability P(c|Ψ) of each possible defect category using the above formula. For example, assume there are four categories: minor defect, general defect, severe defect, and critical defect. By calculating the probability of each category, the system can determine the most likely defect category. Taking into account the quantum coherence between features, it can more accurately capture the complex relationships between defect features. Automatic classification through a mathematical model reduces manual intervention and improves the processing efficiency.

[0119] Assume that the feature matrix Φ of the defects has been obtained through the feature extraction unit 23 and Ψ has been obtained after non - linear mapping. Next, the defect classification unit 24 calculates the probability P(c|Ψ) of each possible defect category using the above formula. For example, assume there are four categories: minor defect, general defect, severe defect, and critical defect. By calculating the probability of each category, the system can determine the most likely defect category. The above algorithm takes into account the quantum coherence between features and can more accurately capture the complex relationships between defect features. The classification criteria can be adjusted according to the actual application scenario to adapt to different quality control requirements.

[0120] The multi - angle lighting device 11 and the high - resolution industrial camera 12 in the image acquisition module 1 work together to obtain multi - angle image data of the LCD panel. The annular light source design ensures that defects at different angles can be fully revealed, while the high - resolution camera ensures that the details of tiny defects can also be captured.

[0121] The image preprocessing unit 21 performs noise reduction and geometric correction on the original image to eliminate image distortion caused by the shooting angle. The image fusion unit 22 uses an innovative algorithm of random matrix theory to fuse multi-angle image data into a comprehensive image, retaining the key information of each angle. The feature extraction unit 23 extracts defect features using topology methods and group theory algorithms, highlighting the edge and texture information in the image. The defect classification unit 24 realizes high-precision defect classification based on quantum probability theory, providing reliable defect category and location information. These data and processing steps are closely linked, constituting a complete defect detection process. High-quality data acquisition ensures the basis for subsequent processing, while precise processing algorithms improve the reliability and accuracy of the detection results, ultimately providing strong support for the quality control of the production line.

[0122] Through this series of innovative algorithms, the present invention realizes high-precision detection and classification of defects in liquid crystal display screens. Compared with traditional methods, the present invention has significant improvements in both detection accuracy and classification precision, especially for some difficult-to-detect tiny defects or defects with complex shapes, where the present invention shows obvious advantages.

[0123] In a preferred embodiment of the present invention, after determining the defect category and location information, it further includes further quantitative analysis of the defects. Specifically, based on the defect category and location information, this method calculates the defect area and aspect ratio, and classifies the defects according to a preset threshold. This step is crucial for evaluating the severity of the defects and formulating corresponding treatment strategies.

[0124] When calculating the defect area, the present invention adopts an improved region growing algorithm. This algorithm starts from the center point of the detected defect and gradually expands to the surrounding until it encounters the defect boundary. The defect area S can be expressed as:

[0125] S = Σ i,j∈R p(i,j),

[0126] where R is the defect region, and p(i,j) is an indicator function indicating whether the pixel (i,j) belongs to the defect region. To improve the calculation efficiency, this method adopts an eight-connected neighborhood search strategy.

[0127] In the defect detection of liquid crystal displays (LCDs), accurately calculating the defect area is an important step in evaluating its severity. The improved region growing algorithm starts from the detected defect center point and gradually expands outward until it encounters the defect boundary. Suppose an image of an LCD panel is captured by a high-resolution industrial camera 12 with a 4K resolution. After denoising and geometric correction of the image by the image preprocessing unit 21, the image fusion unit 22 fuses the multi-angle image data into a comprehensive image. Next, the feature extraction unit 23 uses topological methods to identify potential defect regions and marks the defect center points.

[0128] For example, suppose a defect region suspected of being a scratch is found. Through the improved region growing algorithm, starting from this center point, it can gradually expand outward until the boundary of the entire defect region is determined. During this process, all pixels (i, j) belonging to the defect region are marked as 1, and pixels not belonging to the defect region are marked as 0. Finally, by summing all p(i, j), the total area S of the defect can be obtained. The actual area of the defect can be accurately calculated. Adopting an eight-connected neighborhood search strategy improves the calculation efficiency and is suitable for real-time detection requirements.

[0129] The aspect ratio is another important defect quantification index used to evaluate the shape characteristics of defects. Principal component analysis (PCA) determines the main direction and the secondary main direction of the defect by calculating the eigenvalues and eigenvectors of the covariance matrix.

[0130] For the calculation of the aspect ratio, the present invention uses the principal component analysis (PCA) method. First, calculate the covariance matrix of the defect region:

[0131]

[0132] where x i is the pixel coordinate within the defect region, is the average coordinate. Then, solve the eigenvalues and eigenvectors of the covariance matrix. The direction of the eigenvector corresponding to the largest eigenvalue is the main direction of the defect, and its length is the length of the defect; the direction of the eigenvector corresponding to the second largest eigenvalue is perpendicular to the main direction, and its length is the width of the defect. The aspect ratio is the ratio of these two lengths.

[0133] Continuing with the above-mentioned scratch defect as an example, assume that all pixel coordinates of the defect area have been determined through the region growing algorithm. Next, calculate the covariance matrix C of these coordinates and find two main directions through eigenvalue decomposition. Assume that the length in the direction of the eigenvector corresponding to the largest eigenvalue is 5 mm, and the length in the direction of the eigenvector corresponding to the second largest eigenvalue is 1 mm, then the aspect ratio of this defect is 5. This not only considers the area of the defect but also combines its shape characteristics, providing a more comprehensive evaluation. Automatically calculating the aspect ratio through a mathematical model reduces manual intervention and improves processing efficiency.

[0134] In terms of defect level classification, the present invention sets a set of scientific classification criteria according to the actual application scenarios of liquid crystal display screens. Preferably, the defects are divided into four levels:

[0135] 1. Minor defect: Area S < 0.1 mm 2 and aspect ratio < 3;

[0136] 2. General defect: 0.1 mm 2 ≤ S < 0.5 mm 2 or 3 ≤ aspect ratio < 5;

[0137] 3. Severe defect: 0.5 mm 2 ≤ S < 1 mm 2 or 5 ≤ aspect ratio < 10;

[0138] 4. Fatal defect: S ≥ 1 mm 2 or aspect ratio ≥ 10;

[0139] This classification method considers the size and shape of the defects and can more comprehensively evaluate the impact of the defects on the display quality.

[0140] The present invention also provides an intelligent analysis system for liquid crystal display screen defects based on visual perception. This system includes an image acquisition module 1, an image processing module 2, and a result output module 3.

[0141] The image acquisition module 1 is used to acquire multi-angle image data of the liquid crystal display screen. This module includes a multi-angle lighting device 11 and a high-resolution industrial camera 12. The multi-angle lighting device 11 adopts a ring design and includes multiple independently controllable LED light sources, which can achieve lighting at different angles. The high-resolution industrial camera 12 adopts the latest CMOS sensor technology with a resolution of 4K and can capture tiny defect details.

[0142] The image processing module 2 is the core of this system and is used to process and analyze the acquired image data. This module includes an image preprocessing unit 21, an image fusion unit 22, a feature extraction unit 23, and a defect classification unit 24. The image preprocessing unit 21 is responsible for denoising and geometric correction of the original image. The image fusion unit 22 uses the innovative algorithm of the present invention to fuse multi-angle image data into a comprehensive image. The feature extraction unit 23 extracts defect features using topology methods and group theory algorithms. The defect classification unit 24 realizes high-precision defect classification based on quantum probability theory.

[0143] The result output module 3 is used to output defect category and location information. This module includes a visual display unit 31 and a data storage unit 32. The visual display unit 31 can intuitively display the detection results, including the location, type, and severity of the defects. The data storage unit 32 is responsible for saving the detection results to the database for subsequent statistical analysis and quality tracking.

[0144] This system is also equipped with a high-performance GPU acceleration computing unit, which can realize real-time defect detection and analysis. Through parallel computing technology, this system can simultaneously process the detection tasks of multiple liquid crystal display screens, greatly improving the detection efficiency of the production line.

[0145] Generally speaking, the method and system provided by the present invention have significant innovation and practicality in the field of liquid crystal display screen defect detection. Through multi-angle image acquisition, innovative image processing algorithms, and quantum probability classification methods, the present invention greatly improves the accuracy and efficiency of defect detection. At the same time, through the precise quantification and grading of defects, reliable data support is provided for subsequent quality control and production optimization. This not only helps to improve product quality but also significantly reduces production costs, bringing substantial economic benefits to liquid crystal display screen manufacturing enterprises. To verify the superiority of the intelligent defect detection method and analysis system for liquid crystal display screens based on visual perception of the present invention, a series of comparative experiments were carried out. This article will introduce in detail an embodiment and two comparative examples and comprehensively evaluate their performance through objective indicators.

[0146] Example 1 uses the method of the present invention, including innovative technologies such as multi-angle image acquisition, image fusion based on random matrix theory, topology defect enhancement, and quantum probability classification. Comparative Example 1 uses traditional single-angle image acquisition and convolutional neural network (CNN) for defect detection. Comparative Example 2 uses multi-angle image acquisition but uses conventional image processing methods and support vector machine (SVM) for classification.

[0147] The following five key indicators were selected to evaluate the performance of each method:

[0148] 1. Detection accuracy rate: The proportion of correctly identified defects.

[0149] 2. False positive rate: The proportion of misjudging normal areas as defects.

[0150] 3. Miss rate: The proportion of failing to detect actual existing defects.

[0151] 4. Processing speed: The number of images that can be processed per second.

[0152] 5. Minimum detectable defect size: The minimum defect size that can be reliably detected.

[0153] These metrics comprehensively reflect the performance of the defect detection system, covering multiple aspects such as accuracy, reliability, and efficiency.

[0154] The detection method uses standard ten-fold cross-validation. A dataset containing 10,000 liquid crystal display images with various types and sizes of defects is used. Each method is tested in the same hardware environment to ensure the comparability of the results.

[0155] The following are the detailed test results:

[0156] Index Example 1 Comparative Example 1 Comparative Example 2 Detection accuracy 98.7％ 92.3％ 95.1％ False alarm rate 0.5％ 2.8％ 1.7％ Missed detection rate 0.8％ 4.9％ 3.2％ Processing speed (frames per second) 25 40 18

[0157] It can be clearly seen from the test results that the method of the present invention performs excellently in most key metrics. Especially in terms of detection accuracy, false positive rate, and miss rate, the performance of the present invention is significantly better than the comparative method. This is mainly due to the innovative multi-angle image fusion technology and quantum probability classification method of the present invention, which can capture defect features more comprehensively and accurately identify defects in complex backgrounds.

[0158] It is worth noting that the present invention also performs well in terms of the minimum detectable defect size and can reliably detect tiny defects of 10 μm. This is crucial for the quality control of high-resolution display screens because even extremely small defects may affect the user experience.

[0159] In terms of processing speed, the present invention is slightly inferior to the single-angle CNN method. However, considering that the present invention processes multi-angle images, its processing speed of 25 frames per second is sufficient to meet the real-time detection requirements of most production lines. Moreover, compared with the multi-angle SVM method in Comparative Example 2, the present invention still has an obvious advantage in processing speed.

[0160] These test results fully prove the superiority of the present invention in the field of liquid crystal display defect detection. High detection accuracy and low false positive rate mean that product defects can be more reliably identified, reducing unnecessary rework and scrap. The low miss rate ensures that almost all defects can be detected in a timely manner, greatly improving the effectiveness of quality control. And the smaller detectable defect size enables the present invention to meet the continuously improving display quality standards.

[0161] Generally speaking, while ensuring high detection accuracy, the present invention can also maintain a relatively fast processing speed, which is of great significance for improving production efficiency and product quality. Although the hardware requirements and algorithm complexity may be slightly higher than those of traditional methods, considering the high added value of liquid crystal display production, such an investment is very worthwhile.

[0162] It should be noted that the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An intelligent detection method for liquid crystal display screen defects based on visual perception, characterized in that: include: The acquisition steps include: Acquire multi-angle image data of the LCD display; Processing steps include: Based on the multi-angle image data, generating a fused image; Extracting defect features according to the fused image; Determining a defect category based on the defect characteristics; Output steps include: The defect category and defect location information are output.

2. The method according to claim 1, characterized in that The acquiring of multi-angle image data of the liquid crystal display screen specifically includes: Use multi-angle lighting equipment to illuminate the LCD screen; The images of the liquid crystal display screen under different lighting angles are acquired through the image acquisition device.

3. The method according to claim 1, characterized in that The generating of the fused image specifically comprises: Preprocessing the multi-angle image data to obtain preprocessed image data; Based on random matrix theory, the preprocessed image data is fused to obtain the fused image.

4. The method according to claim 3, characterized in that The preprocessing of the multi-angle image data comprises: removing noise from the multi-angle image data; Perform geometric correction on the denoised image data.

5. The method according to claim 1, characterized in that The extraction of defect features specifically includes: Applying the Laplace operator and the Bessel function to the fused image to obtain an enhanced image; Based on group theory and discrete mathematics methods, defect features are extracted from the enhanced image.

6. The method according to claim 5, characterized in that Extracting defect features from the enhanced image includes: Performing discrete Fourier transform on the enhanced image; Multiply the transformation result with the generator matrix of the cyclic group to obtain the characteristic matrix.

7. The method according to claim 1, characterized in that Determining the defect category specifically includes: Performing nonlinear mapping on the defect features to obtain mapped features; Based on quantum probability theory, the mapped features are classified to obtain defect categories.

8. The method according to claim 7, characterized in that The nonlinear mapping of defect characteristics comprises: inputting the defect characteristics into a chaotic system; The features are transformed by Logistic mapping and hyperbolic tangent function.

9. The method according to claim 1, characterized in that: Also includes: Calculating the defect area and aspect ratio based on the defect category and defect location information; The defects are classified into different levels according to a preset threshold.

10. A liquid crystal display screen defect intelligent analysis system based on visual perception that implements the method according to any one of claims 1 to 9, characterized in that: include: An image acquisition module, used to acquire multi-angle image data of a liquid crystal display screen; Image processing module for: Based on the multi-angle image data, generating a fused image; Extracting defect features according to the fused image; Determining a defect category based on the defect characteristics; The result output module is used to output the defect category and defect location information.

Images