LCD system fault detection system and method based on image analysis

Abstract

The invention relates to the field of intelligent detection, and provides an LCD system fault detection system and method based on image analysis, and the method comprises the steps: firstly obtaining an LCD panel image collected by a high-resolution industrial camera, then carrying out the homomorphic filtering and CLAHE contrast enhancement processing of the image, and obtaining an enhanced LCD panel image; the method comprises the steps that firstly, an LCD panel image is obtained, then multi-scale feature enhancement fusion is carried out on the enhanced image to obtain an LCD panel image multi-scale fusion feature map, then statistical analysis is carried out on the feature map based on a Mura segmentation area to obtain a Mura area statistical feature coding vector, and finally a Mura defect recognition result is obtained according to the coding vector. Therefore, the characteristics of the Mura defect can be captured more accurately, misjudgment and missing detection caused by interference factors such as a light source and noise are reduced, the detection precision of the Mura defect is improved, and meanwhile, the detection system can adapt to different production conditions and ensure the stability of a detection result.

Description

Technical Field

[0001] The present application relates to the field of intelligent detection, and more specifically, to an LCD system fault detection system and method based on image analysis. Background Art

[0002] With the widespread application of liquid crystal display technology (LCD) in consumer electronics, industrial control and other fields, quality inspection in its manufacturing process has become a key link to ensure product performance. Among the many types of LCD defects, Mura defects have become a difficult problem for industry detection due to the particularity of their visual presentation. This type of defect is manifested as cloud-like patterns with uneven brightness in the display area. Its causes involve multiple manufacturing links such as abnormal backlight components and disordered arrangement of liquid crystal molecules.

[0003] Traditional manual visual inspection methods rely on engineers' subjective judgment based on their experience, and have inherent defects such as low inspection efficiency and inconsistent standards. Especially when facing low-contrast gradient Mura, the human eye's perception threshold of brightness difference is easily affected by external factors such as ambient light and observation angle, resulting in high defect missed detection rate and misjudgment rate. Although the automated inspection technology based on machine vision has developed in recent years, it still faces multiple technical bottlenecks in practical applications: the abnormal brightness distribution caused by insufficient light source uniformity in the image acquisition stage is easily confused with real defects, and the interference of camera noise and surface reflection significantly affects the image signal-to-noise ratio; in the image processing stage, the traditional algorithm is not sensitive enough to the gradient characteristics of Mura, and the threshold segmentation is difficult to accurately capture the defect area with low brightness difference, and the simple edge detection operator cannot effectively distinguish between the artifacts caused by uneven backlight and real defects. More importantly, the existing methods lack the ability to integrate and analyze the multi-scale features of defects, and cannot establish an objective quantitative evaluation system consistent with subjective perception, which makes it difficult to ensure the robustness of the inspection system under different production batches and environmental parameters.

[0004] Therefore, an optimized LCD system fault detection scheme based on image analysis is desired. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present application provides an LCD system fault detection system and method based on image analysis.

[0006] According to one aspect of the present application, a method for detecting faults in an LCD system based on image analysis is provided, which includes: acquiring an LCD panel image acquired by a high-resolution industrial camera; performing homomorphic filtering and CLAHE contrast enhancement on the LCD panel image to obtain an enhanced LCD panel image; performing multi-scale feature enhancement fusion on the enhanced LCD panel image to obtain a multi-scale fusion feature map of the LCD panel image, including: performing wavelet decomposition and feature extraction on the enhanced LCD panel image to obtain a low-frequency approximate subband feature map, a horizontal detail subband feature map, a vertical detail subband feature map, and a diagonal detail subband feature map; performing feature enhancement fusion processing based on spatial-semantic dual dimensions on the low-frequency approximate subband feature map, the horizontal detail subband feature map, the vertical detail subband feature map, and the diagonal detail subband feature map to obtain a multi-scale fusion feature map of the LCD panel image; performing statistical analysis based on Mura segmentation areas on the multi-scale fusion feature map of the LCD panel image to obtain a Mura area statistical feature coding vector; and obtaining a Mura defect recognition result based on the Mura area statistical feature coding vector.

[0007] According to another aspect of the present application, there is provided an LCD system fault detection system based on image analysis, which includes: an image acquisition module, used to acquire an LCD panel image acquired by a high-resolution industrial camera; an image preprocessing module, used to perform homomorphic filtering and CLAHE contrast enhancement on the LCD panel image to obtain an enhanced LCD panel image; an image encoding module, used to perform multi-scale feature enhancement fusion on the enhanced LCD panel image to obtain a multi-scale fusion feature map of the LCD panel image, wherein the image encoding module is used to: perform wavelet decomposition and feature extraction on the enhanced LCD panel image to obtain a low-frequency approximate sub-band feature map, a horizontal detail sub-band feature map, a vertical detail subband feature map and a diagonal detail subband feature map; the low-frequency approximate subband feature map, the horizontal detail subband feature map, the vertical detail subband feature map and the diagonal detail subband feature map are subjected to feature enhancement fusion processing based on the spatial-semantic dual dimensions to obtain the multi-scale fusion feature map of the LCD panel image; a Mura area analysis module is used to perform statistical analysis based on the Mura segmentation area on the multi-scale fusion feature map of the LCD panel image to obtain a Mura area statistical feature coding vector; a defect recognition module is used to obtain a Mura defect recognition result based on the Mura area statistical feature coding vector.

[0008] The present application has significant technical effects due to the adoption of the above technical solutions: the LCD system fault detection system and method based on image analysis provided by the present application first obtains the LCD panel image collected by a high-resolution industrial camera, then performs homomorphic filtering and CLAHE contrast enhancement processing on the image to obtain an enhanced LCD panel image, and then performs multi-scale feature enhancement fusion on the enhanced image to obtain a multi-scale fusion feature map of the LCD panel image, and then performs statistical analysis on the feature map based on the Mura segmentation area to obtain the Mura area statistical feature coding vector, and finally obtains the Mura defect recognition result based on this coding vector. In this way, the characteristics of Mura defects can be captured more accurately, and the misjudgment and missed detection caused by interference factors such as light source and noise can be reduced, thereby improving the detection accuracy of Mura defects. At the same time, the detection system can adapt to different production conditions to ensure the stability of the detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 Flow chart of an LCD system fault detection method based on image analysis according to an embodiment of the present application.

[0011] Figure 2 Flow chart of step S3 in the LCD system fault detection method based on image analysis according to an embodiment of the present application.

[0012] Figure 3 Flow chart of step S31 in the LCD system fault detection method based on image analysis according to an embodiment of the present application.

[0013] Figure 4 Flow chart of step S32 in the LCD system fault detection method based on image analysis according to an embodiment of the present application.

[0014] Figure 5 Flow chart of step S321 in the LCD system fault detection method based on image analysis according to an embodiment of the present application.

[0015] Figure 6 Flow chart of step S321 - 3 in the LCD system fault detection method based on image analysis according to an embodiment of the present application.

[0016] Figure 7Flow chart of step S4 in the LCD system fault detection method based on image analysis according to an embodiment of the present application.

[0017] Figure 8 4 is a system block diagram of an LCD system fault detection system based on image analysis according to an embodiment of the present application. DETAILED DESCRIPTION

[0018] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0019] Based on this, the technical concept of this application is to build a full-process detection system for LCD panel Mura defects through a multi-level image enhancement and multi-scale feature fusion framework. Specifically, after first acquiring the original image with a high-resolution camera, the illumination component and the reflection component are separated by homomorphic filtering, and the local contrast is enhanced by combining the CLAHE algorithm to effectively overcome the uneven background illumination and surface reflection interference; then the image is decomposed into different frequency bands and directional features, and defect features with a larger receptive field are extracted to enhance the sensitivity to low-contrast gradient defects; then, multi-scale feature fusion and semantic segmentation are used to accurately locate the boundary distribution of Mura defects of different forms; finally, statistical features such as texture and brightness distribution of the defect area are extracted to establish a defect quantification model to achieve an objective evaluation consistent with the subjective perception of the human eye. This solution suppresses background noise interference through frequency-spatial joint enhancement technology, uses multi-resolution analysis to improve the detection capability of tiny defects, and solves the problem of the lack of quantitative standards for Mura defects in traditional algorithms through statistical feature modeling, significantly improving the system's adaptability to complex industrial environments while ensuring detection sensitivity.

[0020] Figure 1 FIG. 1 is a flow chart of an LCD system fault detection method based on image analysis according to an embodiment of the present application. Figure 1 As shown, according to the LCD system fault detection method based on image analysis according to the embodiment of the present application, the method includes: S1, acquiring an LCD panel image collected by a high-resolution industrial camera; S2, performing homomorphic filtering and CLAHE contrast enhancement on the LCD panel image to obtain an enhanced LCD panel image; S3, performing multi-scale feature enhancement fusion on the enhanced LCD panel image to obtain a multi-scale fusion feature map of the LCD panel image; S4, performing statistical analysis based on the Mura segmentation area on the multi-scale fusion feature map of the LCD panel image to obtain a Mura area statistical feature coding vector; S5, obtaining a Mura defect recognition result based on the Mura area statistical feature coding vector.

[0021] In step S1, an LCD panel image captured by a high-resolution industrial camera is obtained. It should be understood that the information in the LCD panel image is essentially a digital mapping of the optical performance of the panel. In the spatial dimension, the grayscale value of each pixel accurately reflects the brightness intensity of the corresponding area. This brightness distribution characteristic is directly related to the uniformity of the arrangement of the liquid crystal molecules and the optical diffusion efficiency of the backlight module. For example, when the grayscale value of a local area fluctuates abnormally, it may imply a change in transmittance caused by uneven thickness of the liquid crystal layer or abnormal driving voltage, which is a typical manifestation of Mura defects. At the texture feature level, high-resolution images can clearly present the periodic arrangement structure of the TFT array. When this regularity is destroyed, it may mean pixel dislocation caused by packaging process defects or external force damage. Such texture anomalies are often accompanied by specific forms of Mura defects. In general, by deeply analyzing the information in the acquired LCD panel image, Mura defects can be effectively identified.

[0022] In step S2, the LCD panel image is subjected to homomorphic filtering and CLAHE contrast enhancement to obtain an enhanced LCD panel image. Accordingly, it is considered that the illumination non-uniformity and surface reflection interference in the image acquisition stage will significantly affect the recognition of defects. Since the light source distribution in the industrial detection environment is difficult to achieve absolute uniformity, the image captured by the camera often contains low-frequency brightness fluctuations caused by the brightness difference of the backlight module or the reflection of the external environment. This type of interference is manifested as a large-area gradual brightness offset in the airspace, and its frequency characteristics overlap with the low-frequency gradual characteristics of the real Mura defect. In addition, the relative position change between the high-reflectivity material on the LCD surface and the detection equipment is prone to specular reflection, resulting in abnormal saturation of local pixel values ​​or the formation of highlight artifacts. These interference factors will mask the slight contrast changes of the real defects, making it difficult for the traditional threshold segmentation algorithm to effectively distinguish between defective areas and optical artifacts. Therefore, in order to enhance the local contrast of the image while eliminating the influence of unfavorable factors such as uneven illumination, noise and reflection, the present application obtains an enhanced LCD panel image by performing homomorphic filtering and CLAHE contrast enhancement on the LCD panel image. That is, homomorphic filtering decomposes the image into illumination and reflection components through frequency domain filtering operations, compresses the dynamic range of the low-frequency global brightness distribution, and enhances the high-frequency detail features, effectively suppressing the low-frequency brightness drift caused by uneven backlight. The CLAHE algorithm breaks through the limitations of traditional histogram equalization on global contrast adjustment by performing contrast-limited histogram equalization in local sub-regions, while enhancing the local contrast of Mura defects and avoiding information distortion caused by excessive enhancement of surface reflective areas. The combination of the two forms a dual enhancement mechanism of "global brightness correction-local contrast optimization": homomorphic filtering eliminates large-area background interference, and CLAHE enhances the gradient changes at the edges of defects, jointly improving the signal-to-noise ratio of low-contrast defects and background.

[0023] In step S3, multi-scale feature enhancement fusion is performed on the enhanced LCD panel image to obtain a multi-scale fusion feature map of the LCD panel image. Specifically, Figure 2 FIG. 1 is a flow chart of step S3 in the LCD system fault detection method based on image analysis according to an embodiment of the present application. Figure 2 As shown, the step S3 includes: S31, performing wavelet decomposition and feature extraction on the enhanced LCD panel image to obtain a low-frequency approximate subband feature map, a horizontal detail subband feature map, a vertical detail subband feature map and a diagonal detail subband feature map; S32, performing feature enhancement fusion processing based on spatial-semantic dual dimensions on the low-frequency approximate subband feature map, the horizontal detail subband feature map, the vertical detail subband feature map and the diagonal detail subband feature map to obtain a multi-scale fusion feature map of the LCD panel image.

[0024] In step S31, the enhanced LCD panel image is subjected to wavelet decomposition and feature extraction to obtain a low-frequency approximate sub-band feature map, a horizontal detail sub-band feature map, a vertical detail sub-band feature map, and a diagonal detail sub-band feature map. Specifically, Figure 3 FIG. 1 is a flow chart of step S31 in the LCD system fault detection method based on image analysis according to an embodiment of the present application. Figure 3 As shown, the step S31 includes: S311, performing wavelet decomposition on the enhanced LCD panel image to obtain a low-frequency approximate subband matrix, a horizontal detail subband matrix, a vertical detail subband matrix and a diagonal detail subband matrix; S312, performing feature extraction based on hole convolution coding on the low-frequency approximate subband matrix, the horizontal detail subband matrix, the vertical detail subband matrix and the diagonal detail subband matrix to obtain the low-frequency approximate subband feature map, the horizontal detail subband feature map, the vertical detail subband feature map and the diagonal detail subband feature map.

[0025] In step S311, the enhanced LCD panel image is subjected to wavelet decomposition to obtain a low-frequency approximate subband matrix, a horizontal detail subband matrix, a vertical detail subband matrix, and a diagonal detail subband matrix. Accordingly, considering that the morphological characteristics of defects often present multi-scale distribution characteristics. When the traditional image processing method directly processes the full-frequency domain information, the coupling of high-frequency noise and low-frequency background interference will lead to feature confusion, especially for the gradual Mura with a contrast lower than the human eye recognition threshold, its weak brightness gradient change is easily submerged in the background texture. More importantly, the diffuse edge of the Mura defect has a differential response in different spatial directions. For example, a horizontally extending band defect may present a stronger feature response in the vertical detail subband, and the traditional single-scale feature extraction method cannot effectively separate this directional sensitivity feature. Therefore, in order to be able to analyze and process detail information in different directions more carefully, the present application performs wavelet decomposition on the enhanced LCD panel image to obtain a low-frequency approximate subband matrix, a horizontal detail subband matrix, a vertical detail subband matrix, and a diagonal detail subband matrix. In this way, the image is decomposed into low-frequency approximate subbands and three high-frequency detail subbands: horizontal, vertical, and diagonal. The low-frequency approximate subband focuses on the overall brightness distribution of the image, and can effectively characterize the low-frequency brightness gradient characteristics of the Mura defect; the horizontal detail subband strengthens the sudden response of the vertical edge in the image, and is suitable for detecting linear defects with horizontal extension characteristics; the vertical detail subband enhances the horizontal edge characteristics and has a higher sensitivity to the vertically distributed Mura area; the diagonal detail subband simultaneously captures the oblique texture changes, and is used to identify irregular cloud-like defects. This multi-directional and multi-scale decomposition method breaks through the limitation of the traditional spatial domain processing on the singleness of feature dimensions, and provides a physically meaningful feature expression basis for subsequent feature encoding.

[0026] In step S312, the low-frequency approximate subband matrix, the horizontal detail subband matrix, the vertical detail subband matrix and the diagonal detail subband matrix are subjected to feature extraction based on hole convolution coding to obtain the low-frequency approximate subband feature map, the horizontal detail subband feature map, the vertical detail subband feature map and the diagonal detail subband feature map. Accordingly, considering that the low-frequency approximate subband matrix, the horizontal detail subband matrix, the vertical detail subband matrix and the diagonal detail subband matrix respectively contain information of different scales and directions of the image. When processing low-frequency approximate subbands, the fixed receptive field of the conventional convolution kernel may not be able to capture the Mura area formed by a large range of brightness gradients due to insufficient coverage; in the high-frequency detail subbands, a convolution window that is too small is prone to excessively focus on local noise and ignore the continuity characteristics of the defect edge. In addition, the morphological differences of Mura defects in the horizontal, vertical, and diagonal directions require the feature extractor to have directional perception capabilities, but the traditional convolution operation lacks pertinence in responding to directional features, resulting in the key information in different directional sub-bands being difficult to be effectively mined. Based on this, in the technical solution of the present application, the low-frequency approximate sub-band matrix, the horizontal detail sub-band matrix, the vertical detail sub-band matrix, and the diagonal detail sub-band matrix are subjected to feature extraction based on hole convolution coding to obtain the low-frequency approximate sub-band feature map, the horizontal detail sub-band feature map, the vertical detail sub-band feature map, and the diagonal detail sub-band feature map. Specifically, for low-frequency approximate subbands, a dilated convolution layer with a larger dilation rate is used to expand the receptive field without increasing the number of parameters, so that it can cover a larger range of brightness gradient areas and effectively capture the overall distribution of low-contrast Mura; in the horizontal, vertical, and diagonal high-frequency detail subbands, a directional dilated convolution kernel group is used to adjust the dilation direction of the convolution kernel to align with the spatial characteristics of each subband, such as matching the horizontally extended dilated convolution mode for the horizontal detail subband, thereby enhancing the response sensitivity to edge features in a specific direction. This encoding method breaks through the isotropic limitation of traditional convolution operations and enables the feature extraction process to adapt to the spatial-frequency characteristics of different frequency bands.

[0027] In step S32, the low-frequency approximate subband feature map, the horizontal detail subband feature map, the vertical detail subband feature map and the diagonal detail subband feature map are subjected to feature enhancement fusion processing based on the spatial-semantic dual dimension to obtain the LCD panel image multi-scale fusion feature map. Specifically, Figure 4 FIG. 1 is a flow chart of step S32 in the LCD system fault detection method based on image analysis according to an embodiment of the present application. Figure 4As shown, the step S32 includes: S321, performing feature enhancement based on spatial-semantic dual dimensions on the low-frequency approximate subband feature map, the horizontal detail subband feature map, the vertical detail subband feature map and the diagonal detail subband feature map to obtain an enhanced low-frequency approximate subband feature map, an enhanced horizontal detail subband feature map, an enhanced vertical detail subband feature map and an enhanced diagonal detail subband feature map; S322, fusing the enhanced low-frequency approximate subband feature map, the enhanced horizontal detail subband feature map, the enhanced vertical detail subband feature map and the enhanced diagonal detail subband feature map to obtain the multi-scale fusion feature map of the LCD panel image.

[0028] In step S321, feature enhancement based on space-semantic dual dimensions is performed on the low-frequency approximate subband feature map, the horizontal detail subband feature map, the vertical detail subband feature map, and the diagonal detail subband feature map to obtain an enhanced low-frequency approximate subband feature map, an enhanced horizontal detail subband feature map, an enhanced vertical detail subband feature map, and an enhanced diagonal detail subband feature map. Specifically, Figure 5 FIG. 4 is a flow chart of step S321 in the LCD system fault detection method based on image analysis according to an embodiment of the present application. Figure 5 As shown, the step S321 includes: S321-1, extracting the channel feature vector of the (i, j)th pixel position from the low-frequency approximate subband feature map as the low-frequency approximate subband feature channel feature vector to be enhanced; S321-2, performing n random scans on the low-frequency approximate subband feature map to obtain n low-frequency approximate subband feature channel feature vectors as a sparse set of low-frequency approximate subband feature reference vectors; S321-3, performing implicit compensation enhancement based on the mask control component on the sparse set of low-frequency approximate subband feature reference vectors and the low-frequency approximate subband feature channel feature vector to be enhanced to obtain an implicit coding vector of a low-frequency approximate subband feature enhancement component; S321-4, fusing the low-frequency approximate subband feature enhancement component implicit coding vector and the low-frequency approximate subband feature channel feature vector to be enhanced to obtain an enhanced low-frequency approximate subband feature channel feature vector, wherein the enhanced low-frequency approximate subband feature channel feature vector is the channel feature vector of the pixel position (i, j) of the enhanced low-frequency approximate subband feature map.

[0029] It should be understood that although the sub-band feature maps of different frequency bands and directions realize multi-scale expression of defect features through wavelet decomposition, each sub-band still has inherent interference and feature expression limitations. In the low-frequency approximation sub-band, the global brightness gradient caused by uneven backlight overlaps with the low-frequency characteristics of the real defect. Traditional feature enhancement methods are difficult to suppress background interference while retaining the overall morphology of the defect; in the high-frequency detail sub-band, the high-frequency artifacts formed by camera noise and surface reflection are easily confused with the edge features of the defect, and the differentiated responses of sub-bands in different directions to the defect morphology (such as horizontal sub-bands are more sensitive to vertical extension defects) require the enhancement process to take into account both spatial distribution and semantic association. More importantly, the local gradient characteristics of mura defects may only appear as weak signals in a single sub-band, and effective representation can only be formed through cross-regional contextual information mining, but conventional enhancement methods lack the ability to explicitly model the implicit semantic associations in the feature space, resulting in the inability of the enhancement process to adapt to the physical characteristics of different sub-bands. To this end, the present application performs feature enhancement based on spatial-semantic dual dimensions on the low-frequency approximate subband feature map, the horizontal detail subband feature map, the vertical detail subband feature map and the diagonal detail subband feature map to obtain an enhanced low-frequency approximate subband feature map, an enhanced horizontal detail subband feature map, an enhanced vertical detail subband feature map and an enhanced diagonal detail subband feature map.

[0030] In particular, the processing process of the low-frequency approximate subband feature map is taken as an example to explain in detail. In detail, the low-frequency approximate subband feature map is first randomly scanned to obtain a set of reference vectors covering different brightness distribution areas. The hierarchical association between the feature vector to be enhanced and each reference vector in the hyperbolic space is measured by the Poincare distance, and the spatial modulation parameter matrix is ​​constructed to quantify the spatial correlation strength; at the same time, the implicit semantic association between the feature to be enhanced and the reference vector is calculated, and the abstract similarity between the two in the brightness gradient pattern is mined to generate a semantic association encoding matrix. Subsequently, the reference vector is screened at the semantic level using the semantic association encoding matrix as the basic mask control component, and then the spatial modulation matrix is ​​used as the secondary mask control component to implement spatial correlation modulation, forming an information compensation mechanism for double-layer mask control. This process dynamically injects the reference vector information that conforms to the defect brightness distribution pattern into the feature to be enhanced. For example, when there is a weak brightness change suspected of a defect in a certain area, the reference area with a similar gradient pattern is screened out through semantic association, and then its contribution weight is strengthened according to the spatial correlation, and finally the enhanced low-frequency approximate subband feature map is generated through feature fusion.

[0031] Specifically, in the embodiment of the present application, step S321-1 includes: extracting a channel feature vector at the (i, j)th pixel position from the low-frequency approximate sub-band feature map as a channel feature vector of the low-frequency approximate sub-band feature to be enhanced, which can be expressed as follows: ;in, is the low-frequency approximate subband feature map, is the set of real numbers, and They are The height and width of each feature matrix along the channel dimension, yes The number of channels, express The real number set consisting of the number of channels, yes The channel feature vector of the (i, j)th pixel position in , It is the feature vector of the low-frequency approximate sub-band feature channel to be enhanced.

[0032] It should be understood that extracting the channel feature vector of a specific position (i, j) from the low-frequency approximate sub-band feature map as the object to be enhanced is intended to focus on the brightness distribution characteristics of the area. Specifically, due to the spectral overlap between the global backlight unevenness and the low-frequency characteristics of the real defect in the low-frequency sub-band, directly processing the entire feature map may cause background interference to be mistakenly enhanced. Through local feature vector extraction, the local brightness gradient pattern and background noise can be separated in a targeted manner while retaining the overall morphology of the defect (such as the gradient characteristics of the Mura defect). In other words, through localized operations, precise targets to be enhanced are provided for subsequent spatial-semantic modulation, avoiding excessive smoothing or information loss of non-defective areas by global operations, while maintaining the spatial continuity of the feature map, laying the foundation for cross-regional context mining.

[0033] Specifically, in the embodiment of the present application, the step S321-2 includes: performing n random scans on the low-frequency approximate sub-band feature map to obtain n low-frequency approximate sub-band feature channel feature vectors as a sparse set of low-frequency approximate sub-band feature reference vectors, which can be expressed as follows: ;in, is a sparse set of low-frequency approximate subband feature reference vectors, and are the first, second, and third sparse sets of low-frequency approximate subband feature reference vectors. and low-frequency approximate subband feature reference vectors, and .

[0034] It should be understood that by randomly sampling the low-frequency sub-band feature map n times to generate a sparse set of low-frequency approximate sub-band feature reference vectors, it is essentially to build a dynamic feature memory library covering different brightness distribution patterns. Specifically, the random design breaks through the sensitivity of fixed sampling (such as sliding windows) to local overfitting, and by introducing spatial diversity, it ensures that the reference set can cover the typical brightness gradient patterns in the feature map (including potential defect areas and normal background areas). The sparse sampling strategy strikes a balance between computational efficiency and information coverage, simulating the "selective attention" mechanism of biological vision, screening representative local patterns from global features, and avoiding interference from noise or redundant features. In other words, this step provides a multi-scale, multi-position contextual reference benchmark for subsequent modulation, enabling the enhancement process to adapt to the brightness distribution characteristics of different regions.

[0035] Specifically, Figure 6 FIG. 1 is a flow chart of step S321-3 in the LCD system fault detection method based on image analysis according to an embodiment of the present application. Figure 6 As shown, the step S321-3 includes: S321-31, calculating the Poincare distance between the feature vector of the low-frequency approximate subband feature channel to be enhanced and each low-frequency approximate subband feature reference vector in the sparse set of the low-frequency approximate subband feature reference vector to obtain a low-frequency approximate subband feature spatial modulation parameter matrix; S321-32, calculating the implicit semantic association between the feature vector of the low-frequency approximate subband feature channel to be enhanced and each low-frequency approximate subband feature reference vector in the sparse set of the low-frequency approximate subband feature reference vector to obtain a low-frequency approximate subband feature semantic association coding matrix A set of matrices; S321-33, using each low-frequency approximate subband feature semantic association coding matrix in the set of low-frequency approximate subband feature semantic association coding matrices as a basic mask control component and using the low-frequency approximate subband feature spatial modulation parameter matrix as a secondary mask control component, the low-frequency approximate subband feature information compensation coding vector between each low-frequency approximate subband feature reference vector in the sparse set of low-frequency approximate subband feature reference vectors and the low-frequency approximate subband feature channel feature vector to be enhanced is explicitly modeled and controlled to obtain an implicit coding vector of the low-frequency approximate subband feature enhancement component.

[0036] More specifically, in the embodiment of the present application, the step S321-31 includes: calculating the Poincare distance between the low-frequency approximate subband feature channel feature vector to be enhanced and each low-frequency approximate subband feature reference vector in the sparse set of the low-frequency approximate subband feature reference vector to obtain a low-frequency approximate subband feature spatial modulation parameter matrix, which can be expressed as the following formula: ;in, To calculate the square of the Euclidean norm of a vector, is the inverse hyperbolic cosine function, for and The Poincare distance between are the eigenvalues ​​in the low-frequency approximate subband feature space modulation parameter matrix, is the low-frequency approximate subband feature spatial modulation parameter matrix.

[0037] It should be understood that the Poincare distance is used to measure the hierarchical association between the feature vector of the low-frequency approximate subband feature channel to be enhanced and the reference vector of the low-frequency approximate subband feature in the hyperbolic space. Compared with the Euclidean distance, it is better at capturing the tree-like or hierarchical data distribution characteristics. Specifically, the brightness gradient pattern in the low-frequency subband (such as the intensity change between the defect and the background) often presents a nonlinear hierarchical relationship (such as a strong gradient in the center to a weak gradient at the edge). The Poincare distance models such a relationship as the intensity of spatial correlation through the geometric characteristics of the hyperbolic space. The generated low-frequency approximate subband feature spatial modulation parameter matrix not only quantifies the spatial similarity between the area to be enhanced and the reference area, but also implicitly encodes the brightness distribution law across regions (such as the difference between the local gradient trend of the defect and the global change of the background), providing a differentiated weight allocation basis for subsequent spatial modulation, thereby suppressing the interference of non-related areas while enhancing the defects.

[0038] More specifically, in the embodiment of the present application, the step S321-32 includes: calculating the implicit semantic association between the low-frequency approximate subband feature channel feature vector to be enhanced and each low-frequency approximate subband feature reference vector in the sparse set of the low-frequency approximate subband feature reference vector to obtain a set of low-frequency approximate subband feature semantic association coding matrices, which can be expressed as follows: ;in, is matrix multiplication, for The corresponding weight matrix is, is the transpose operation, yes and The length of the vector after multiplication, yes function, yes and The low-frequency approximate sub-band feature semantic association coding matrix between , that is, the first in the set of low-frequency approximate sub-band feature semantic association coding matrices A low-frequency approximate sub-band feature semantic association encoding matrix, is the set of low-frequency approximate sub-band feature semantic association coding matrices, and are the first, second and third low-frequency approximate subband feature semantic association coding matrices in the set. A low-frequency approximate sub-band feature semantic association encoding matrix.

[0039] It should be understood that by mining the abstract semantic similarity between the feature vector of the low-frequency approximate sub-band feature channel to be enhanced and the reference vector of the low-frequency approximate sub-band feature (such as the commonalities and differences in the brightness gradient pattern), a low-frequency approximate sub-band feature semantic association coding matrix is ​​generated. This step goes beyond the matching of pixel-level surface features and focuses on high-order semantic patterns (such as "center-edge gradient" and "multi-scale texture similarity"), which can effectively solve the problem of misassociation caused by traditional methods relying on explicit feature similarity. For example, when there are weak defects in the area to be enhanced, semantic association can filter out reference areas with similar gradient directions but different intensities, even if they are far away in space. In other words, the obtained low-frequency approximate sub-band feature semantic association coding matrix is ​​used as a basic mask to realize the semantic-level screening of the reference vector, ensuring that subsequent information compensation only injects feature components that are strongly related to the defect pattern, avoiding interference from irrelevant background information.

[0040] More specifically, in the embodiment of the present application, the step S321-33 includes: calculating the position difference vector between each low-frequency approximate sub-band feature reference vector and the low-frequency approximate sub-band feature channel feature vector to be enhanced to obtain a set of low-frequency approximate sub-band feature information compensation coding vectors, and the process can be expressed as: ;in, yes and The low-frequency approximate sub-band feature information compensation coding vector between.

[0041] Based on the basic mask control component and the secondary mask control component, each low-frequency approximate sub-band feature information compensation coding vector in the set of low-frequency approximate sub-band feature information compensation coding vectors is subjected to mask-compensation optimization based on dual-field cooperative modulation to obtain a set of low-frequency approximate sub-band feature information compensation optimization coding vectors. The process can be expressed as: ;in, yes The low-frequency approximate subband feature information after regulation is compensated and regulated by the coding vector, yes The optimized low-frequency approximate subband feature information is compensated for the optimized coding vector.

[0042] Based on the basic mask control component and the secondary mask control component, the set of the low-frequency approximate sub-band feature information compensation optimization coding vectors is subjected to global explicit modeling modulation aggregation to obtain the low-frequency approximate sub-band feature enhancement component implicit coding vector. The above process can be expressed as follows: ;in, is the number of vectors in the sparse set of low-frequency approximate subband feature reference vectors, It is the implicit coding vector of the low-frequency approximate subband feature enhancement component.

[0043] It should be understood that a hierarchical modulation mechanism is formed by using the semantic association encoding matrix of the low-frequency approximate subband features as the primary mask and the spatial modulation parameter matrix of the low-frequency approximate subband features as the secondary mask. Specifically, the primary mask selects the low-frequency approximate subband feature reference vectors that match the defect pattern (such as areas with similar gradient directions) based on semantic similarity, and the secondary mask strengthens the contribution weight of the low-frequency approximate subband feature reference vectors in the local neighborhood according to spatial correlation. For example, for suspected defect areas, the semantic mask gives priority to reference vectors with similar gradient patterns, and the spatial mask gives higher weights to neighboring areas, thereby dynamically fusing contextual information across regions. This explicit modeling strategy breaks through the reliance of traditional enhancement methods on implicit feature learning, and through an interpretable two-layer control mechanism, accurately injects enhancement components that conform to the physical characteristics of the defect (such as supplementing weak gradient signals), while suppressing pseudo-enhancement caused by noise or background unevenness.

[0044] In particular, in the information coding framework of low-frequency approximate subband feature enhancement, the implementation of field density modulation configuration relies on the synergy between the low-frequency approximate subband feature semantic association encoding matrix and the low-frequency approximate subband feature spatial modulation parameter matrix. Specifically, the low-frequency approximate subband feature semantic association encoding matrix, as a standard field, encodes the implicit semantic association between feature vectors (such as the abstract similarity of brightness gradient patterns), which is expressed by Maintain transformation invariance constraints. This operation essentially projects the information compensation component (i.e., the difference information between the feature to be enhanced and the reference vector) under the control of the semantic mask into the geometric measurement space of the standard field, ensuring the robustness of the enhancement process to the global brightness gradient background. The low-frequency approximate subband feature space modulation parameter matrix is ​​used as a covariant adjustment field to dynamically adjust the local geometric structure of the standard field through the spatial correlation strength quantized by the Poincare distance (i.e., the hierarchical brightness distribution pattern). Based on The covariant adjustment mechanism of the standard field applies the spatial modulation parameters to the curved space topology, so that the derivative hierarchical geometric measure of the standard field can adapt to the spatial distribution characteristics of the feature map. For example, in the defect edge area, the covariant adjustment field corrects the deviation of the standard field in the local curvature by strengthening the contribution weight of the adjacent reference vector, thereby improving the spatial continuity expression of the defect gradient signal. That is, the intensity coupling of the field interaction is quantified by the conventional partial differential form, and its core lies in establishing the differential manifold relationship between semantic association and spatial modulation. Specifically, the partial differential coupling term of the standard field and the covariant adjustment field reflects the joint gradient direction of semantic screening and spatial enhancement, so that the low-frequency approximate subband feature information compensation optimization coding vector can simultaneously meet the semantic consistency and spatial smoothness constraints. This coupling mechanism suppresses high-frequency noise interference (such as artifacts caused by camera noise) while mining cross-region contextual information (such as the long-range correlation between the local gradient pattern of the defect and the reference vector), enhancing the low-frequency subband feature’s ability to characterize weak defects.

[0045] Specifically, in the embodiment of the present application, the step S321-4 is used to: fuse the implicit coding vector of the low-frequency approximate sub-band feature enhancement component and the low-frequency approximate sub-band feature channel feature vector to be enhanced to obtain an enhanced low-frequency approximate sub-band feature channel feature vector, wherein the enhanced low-frequency approximate sub-band feature channel feature vector is the channel feature vector of the pixel position (i, j) of the enhanced low-frequency approximate sub-band feature map, which can be expressed as follows: ;in, and is a weighted hyperparameter, yes The enhanced low-frequency approximate sub-band feature channel feature vector after enhancement, that is, the channel feature vector of the (i, j)th pixel position of the enhanced low-frequency approximate sub-band feature map.

[0046] It should be understood that the fusion of the implicit coding vector of the low-frequency approximate subband feature enhancement component and the feature vector of the low-frequency approximate subband feature channel to be enhanced can achieve the coordinated optimization of local feature expression and global context information. Specifically, the feature vector of the original low-frequency approximate subband feature channel to be enhanced retains the initial morphology and low-frequency characteristics of the defect, while the implicit coding vector of the low-frequency approximate subband feature enhancement component supplements the weak signal (such as the local gradient details of the Mura defect) lost due to background interference or noise suppression through cross-regional semantic association. The fusion process is not a simple weighted superposition, but a reconstruction of the feature space distribution through nonlinear mapping, so that the enhanced features have both sensitivity to tiny defects and robustness to background changes. For example, in areas with uneven backlighting, the fused features can highlight the continuous gradient pattern of the defect, while weakening the interference of global brightness changes, and ultimately improving the ability of the low-frequency subband features to distinguish defects from the background.

[0047] In step S322, the enhanced low-frequency approximate subband feature map, the enhanced horizontal detail subband feature map, the enhanced vertical detail subband feature map and the enhanced diagonal detail subband feature map are fused to obtain the multi-scale fusion feature map of the LCD panel image. Accordingly, considering that the sub-band feature maps of different frequency bands and directions respectively enhance the defect characteristics of specific dimensions through enhancement processing, their isolated analysis still has the limitation of information fragmentation. Although the enhanced low-frequency approximate subband feature map can characterize the overall brightness distribution of the defect, it loses the edge detail information; and the high-frequency detail enhanced sub-band feature map in the horizontal, vertical and diagonal directions captures the local gradient changes, but lacks the contextual perception of the global morphology of the defect. Therefore, in order to make full use of the advantages of each sub-band, integrate the global and local information of the image, and more comprehensively describe the characteristics of the LCD panel image, the present application obtains the multi-scale fusion feature map of the LCD panel image by fusing the enhanced low-frequency approximate sub-band feature map, the enhanced horizontal detail sub-band feature map, the enhanced vertical detail sub-band feature map and the enhanced diagonal detail sub-band feature map. In this way, the obtained multi-scale fusion feature map of LCD panel image can provide a comprehensive feature representation for subsequent defect detection and analysis. It contains both the overall structural information of the image and the detailed texture information, and organically combines the features of different scales and directions, so that the model can analyze and understand the image from multiple angles, so as to more accurately locate and describe the location, shape, size, texture and other features of Mura defects, and provide a more sufficient basis for defect classification and evaluation.

[0048] In step S4, a statistical analysis based on the Mura segmentation region is performed on the multi-scale fusion feature map of the LCD panel image to obtain a Mura region statistical feature encoding vector. Specifically, Figure 7 FIG. 4 is a flow chart of step S4 in the LCD system fault detection method based on image analysis according to an embodiment of the present application. Figure 7 As shown, the step S4 includes: S41, performing image semantic segmentation on the multi-scale fusion feature map of the LCD panel image to obtain a Mura segmentation area image; S42, extracting statistical features of the Mura segmentation area image to obtain a Mura area statistical feature encoding vector.

[0049] In step S41, the multi-scale fusion feature map of the LCD panel image is subjected to image semantic segmentation to obtain a Mura segmentation area image. Accordingly, considering that the multi-scale fusion feature map of the LCD panel image integrates the global brightness distribution and multi-directional edge details, the problem of accurate boundary division of the defect area still needs to be solved. Image semantic segmentation can classify each pixel in the image according to the information in the feature map, and determine whether it belongs to the Mura defect area, thereby accurately outlining the boundary and range of the Mura defect, and providing accurate location information for subsequent defect analysis and processing. Based on this, in the technical solution of the present application, the multi-scale fusion feature map of the LCD panel image is subjected to image semantic segmentation to obtain a Mura segmentation area image. That is, the Mura segmentation area image obtained by image semantic segmentation can accurately identify the Mura defect area on the LCD panel. Each pixel in the image is classified, and the pixel points belonging to the Mura defect are marked to form a clear segmentation area, so that the Mura defect is clearly visible in the image, which is convenient for the operator or the subsequent analysis algorithm to directly locate and observe the defect, and provide a basis for the evaluation and repair of the defect.

[0050] Specifically, first, the multi-scale fusion feature map of the LCD panel image is used as input data. This feature map has integrated the multi-scale information of low-frequency brightness distribution and high-frequency edge details through the previous steps. To ensure that the eigenvalues ​​of each channel are in the same dimension, the input feature map needs to be normalized, such as using the Min-Max scaling method, to avoid subsequent segmentation deviations due to differences in the value range.

[0051] In the semantic segmentation network architecture design stage, an encoder-decoder network model is usually used, such as U-Net or DeepLabv3+, to take into account the fusion of spatial details and high-level semantic information. The encoder part gradually downsamples through convolution and pooling operations to extract high-level semantic features, including the texture pattern and brightness distribution of mura defects. The decoder part restores the spatial resolution through transposed convolution or interpolation upsampling, and combines the jump connection features of the encoder to refine the segmentation boundaries. To further improve the segmentation accuracy, a spatial or channel attention mechanism can be introduced in the decoding stage, such as the CBAM module, to dynamically weight the key areas related to mura defects in the multi-scale feature map while suppressing the interference of background noise.

[0052] During the training phase, it is necessary to prepare labeled data as supervisory signals. Usually, a manually labeled or simulated binary mask of mura defects is used, where the defect area is marked as 1 and the background is marked as 0. The design of the loss function needs to take into account the class imbalance problem where mura defects usually account for a small proportion. Therefore, the main loss function can use Dice Loss or Focal Loss to enhance the model's attention to the defect area. In addition, boundary-aware loss, such as Boundary Loss, can be introduced to enhance the network's ability to recognize the gradual transition area of ​​the edge of mura defects.

[0053] In the inference stage, the pre-processed multi-scale fusion feature map of the LCD panel image is input into the trained semantic segmentation network, and the probability map of each pixel belonging to the mura defect is output, with a value range of 0 to 1. Subsequently, the probability map is converted into a binary segmentation mask through thresholding, and an adaptive threshold method (such as the Otsu algorithm) or a fixed threshold (such as 0.5) can be used for binarization to generate a preliminary mura segmentation area image. To optimize the segmentation results, morphological post-processing is required, including using an open operation to eliminate isolated noise points, and retaining areas larger than the preset threshold through connected domain analysis to filter out tiny pseudo-defects.

[0054] Finally, the optimized binary mask is aligned with the original image to ensure that the boundaries of the defective area are accurately matched, and the mura segmentation area image is output, where the defective area is represented by white and the background is represented by black. To verify the segmentation accuracy, indicators such as IoU (intersection over union) or pixel-level accuracy can be used to evaluate the consistency of the segmentation result with the true annotation to ensure that it meets the accuracy requirements of industrial detection.

[0055] In step S42, the statistical features of the Mura segmented area image are extracted to obtain the Mura area statistical feature encoding vector. Accordingly, it is considered that although semantic segmentation can accurately locate the defect area, relying solely on the segmentation result cannot meet the needs of defect classification and quantitative evaluation. If the defect assessment is performed directly based on pixel-level brightness values ​​or simple morphological parameters, it is difficult to characterize the complex visual characteristics of the defects. For example, the human eye's perception of Mura defects depends not only on the absolute value of the brightness difference, but also on its distribution uniformity, texture coherence and other abstract features. In addition, there are significant differences in the distribution laws of Mura defects of different forms (such as point-shaped, line-shaped, and cloud-shaped) in the statistical feature space, but the existing methods lack systematic modeling of multidimensional features, which makes it difficult for the classification model to establish a mapping relationship between defect morphology and quality level. Based on this, the present application extracts the statistical features of the Mura segmented area image to obtain the Mura area statistical feature encoding vector. In particular, in a specific example of the present application, for the segmented binary mask area, first extract the brightness distribution characteristics of the original enhanced image within its coverage, including grayscale mean, variance, gradient histogram entropy and other indicators, to characterize the overall contrast and local fluctuation characteristics of the defect; at the same time, calculate the morphological characteristics of the defect area, such as area-to-perimeter ratio, compactness, Fourier descriptor, etc., to quantify its geometric distribution law; further combine the texture analysis algorithm to extract the contrast and correlation parameters of the grayscale co-occurrence matrix to capture the micro-texture characteristics of the defect. After normalization and feature selection, these features are fused into a fixed-dimensional statistical feature encoding vector according to preset weights to form a multi-dimensional defect fingerprint that takes into account brightness, morphology, and texture.

[0056] In step S5, based on the Mura region statistical feature coding vector, a Mura defect recognition result is obtained. Specifically, in the embodiment of the present application, step S5 includes: inputting the Mura region statistical feature coding vector into a defect classifier based on the SVM model to obtain the Mura defect recognition result. In particular, the Mura defect recognition result is divided into "no defect", "minor defect", "moderate defect" and "serious defect" levels. It should be understood that the support vector machine (SVM) is a supervised machine learning algorithm that performs well in pattern recognition and classification tasks. It has good performance for processing a limited number of high-dimensional data, and the Mura region statistical feature coding vector is exactly a high-dimensional data representation obtained after feature extraction of the Mura segmentation region image. SVM can find an optimal hyperplane in the high-dimensional space to separate data points of different categories, and has an effective processing method for both linearly separable and nonlinearly separable data (mapping the data to a higher dimensional space through a kernel function to make it linearly separable). In the Mura defect classification task, the statistical feature coding vectors corresponding to different types of Mura defects can be regarded as data points of different categories, and SVM can use the features of these data points to build a classification model. The input feature vector is analyzed and judged by the SVM model, and the Mura defects are classified into corresponding categories according to the pre-trained classification model, such as different types of Mura defects corresponding to different shapes, sizes, grayscale distribution and other characteristics. This helps production personnel understand the specific situation of the defects so that they can take targeted measures to repair or improve the production process.

[0057] In summary, the LCD system fault detection method based on image analysis based on the embodiment of the present application is explained, which first obtains the LCD panel image captured by a high-resolution industrial camera, then performs homomorphic filtering and CLAHE contrast enhancement processing on the image to obtain an enhanced LCD panel image, and then performs multi-scale feature enhancement fusion on the enhanced image to obtain a multi-scale fusion feature map of the LCD panel image, and then performs statistical analysis on the feature map based on the Mura segmentation area to obtain the Mura area statistical feature coding vector, and finally obtains the Mura defect recognition result based on this coding vector. In this way, the characteristics of Mura defects can be captured more accurately, and the misjudgment and missed detection caused by interference factors such as light source and noise can be reduced, thereby improving the detection accuracy of Mura defects. At the same time, the detection system can adapt to different production conditions to ensure the stability of the detection results.

[0058] Figure 8 FIG. 1 is a system block diagram of an LCD system fault detection system based on image analysis according to an embodiment of the present application. Figure 8As shown, according to an embodiment of the present application, an LCD system fault detection system 100 based on image analysis includes: an image acquisition module 110, used to acquire an LCD panel image acquired by a high-resolution industrial camera; an image preprocessing module 120, used to perform homomorphic filtering and CLAHE contrast enhancement on the LCD panel image to obtain an enhanced LCD panel image; an image encoding module 130, used to perform multi-scale feature enhancement fusion on the enhanced LCD panel image to obtain a multi-scale fusion feature map of the LCD panel image; a Mura area analysis module 140, used to perform statistical analysis based on the Mura segmentation area on the multi-scale fusion feature map of the LCD panel image to obtain a Mura area statistical feature coding vector; a defect recognition module 150, used to obtain a Mura defect recognition result based on the Mura area statistical feature coding vector.

[0059] Here, those skilled in the art will appreciate that the specific functions and operations of the various units and modules in the above-mentioned LCD system fault detection system 100 based on image analysis have been described in detail above. Figures 1 to 7 The description of the LCD system fault detection method based on image analysis has been introduced in detail, and therefore, its repeated description will be omitted.

[0060] In summary, the LCD system fault detection system 100 based on image analysis according to the embodiment of the present application is explained, which first obtains the LCD panel image captured by a high-resolution industrial camera, then performs homomorphic filtering and CLAHE contrast enhancement processing on the image to obtain an enhanced LCD panel image, and then performs multi-scale feature enhancement fusion on the enhanced image to obtain a multi-scale fusion feature map of the LCD panel image, and then performs statistical analysis on the feature map based on the Mura segmentation area to obtain the Mura area statistical feature coding vector, and finally obtains the Mura defect recognition result based on this coding vector. In this way, the characteristics of the Mura defect can be captured more accurately, and the misjudgment and missed detection caused by interference factors such as light source and noise can be reduced, thereby improving the detection accuracy of the Mura defect. At the same time, the detection system can adapt to different production conditions to ensure the stability of the detection results.

Claims

1. A method for detecting LCD system faults based on image analysis, characterized in that: include: Get LCD panel images captured by high-resolution industrial cameras; Performing homomorphic filtering and CLAHE contrast enhancement on the LCD panel image to obtain an enhanced LCD panel image; The enhanced LCD panel image is subjected to multi-scale feature enhancement fusion to obtain a multi-scale fusion feature map of the LCD panel image, including: performing wavelet decomposition and feature extraction on the enhanced LCD panel image to obtain a low-frequency approximate sub-band feature map, a horizontal detail sub-band feature map, a vertical detail sub-band feature map, and a diagonal detail sub-band feature map; performing feature enhancement fusion processing based on space-semantic dual dimensions on the low-frequency approximate sub-band feature map, the horizontal detail sub-band feature map, the vertical detail sub-band feature map, and the diagonal detail sub-band feature map to obtain the multi-scale fusion feature map of the LCD panel image; performing statistical analysis based on the Mura segmentation area on the multi-scale fusion feature map of the LCD panel image to obtain a Mura area statistical feature coding vector; and obtaining a Mura defect recognition result based on the Mura area statistical feature coding vector.

2. The LCD system fault detection method based on image analysis according to claim 1, characterized in that: The enhanced LCD panel image is subjected to wavelet decomposition and feature extraction to obtain a low-frequency approximate subband feature map, a horizontal detail subband feature map, a vertical detail subband feature map, and a diagonal detail subband feature map, including: performing wavelet decomposition on the enhanced LCD panel image to obtain a low-frequency approximate subband matrix, a horizontal detail subband matrix, a vertical detail subband matrix, and a diagonal detail subband matrix; and performing feature extraction based on hole convolution coding on the low-frequency approximate subband matrix, the horizontal detail subband matrix, the vertical detail subband matrix, and the diagonal detail subband matrix to obtain the low-frequency approximate subband feature map, the horizontal detail subband feature map, the vertical detail subband feature map, and the diagonal detail subband feature map.

3. The LCD system fault detection method based on image analysis according to claim 1, characterized in that: The low-frequency approximate subband feature map, the horizontal detail subband feature map, the vertical detail subband feature map and the diagonal detail subband feature map are subjected to feature enhancement and fusion processing based on space-semantic dual dimensions to obtain the multi-scale fusion feature map of the LCD panel image, including: performing feature enhancement based on space-semantic dual dimensions on the low-frequency approximate subband feature map, the horizontal detail subband feature map, the vertical detail subband feature map and the diagonal detail subband feature map to obtain an enhanced low-frequency approximate subband feature map, an enhanced horizontal detail subband feature map, an enhanced vertical detail subband feature map and an enhanced diagonal detail subband feature map; fusing the enhanced low-frequency approximate subband feature map, the enhanced horizontal detail subband feature map, the enhanced vertical detail subband feature map and the enhanced diagonal detail subband feature map to obtain the multi-scale fusion feature map of the LCD panel image.

4. The LCD system fault detection method based on image analysis according to claim 3, characterized in that: The low-frequency approximate subband feature map, the horizontal detail subband feature map, the vertical detail subband feature map and the diagonal detail subband feature map are subjected to feature enhancement based on the spatial-semantic dual dimension to obtain an enhanced low-frequency approximate subband feature map, an enhanced horizontal detail subband feature map, an enhanced vertical detail subband feature map and an enhanced diagonal detail subband feature map, including: extracting a channel feature vector at the (i, j)th pixel position from the low-frequency approximate subband feature map as a low-frequency approximate subband feature channel feature vector to be enhanced; performing n random scans on the low-frequency approximate subband feature map to obtain n low-frequency approximate subband feature channels. The feature vector is used as a sparse set of low-frequency approximate subband feature reference vectors; the sparse set of low-frequency approximate subband feature reference vectors and the low-frequency approximate subband feature channel feature vector to be enhanced are implicitly compensated and enhanced based on a mask control component to obtain an implicit coding vector of a low-frequency approximate subband feature enhancement component; the implicit coding vector of the low-frequency approximate subband feature enhancement component and the low-frequency approximate subband feature channel feature vector to be enhanced are fused to obtain an enhanced low-frequency approximate subband feature channel feature vector, wherein the enhanced low-frequency approximate subband feature channel feature vector is a channel feature vector of a pixel position (i, j) of the enhanced low-frequency approximate subband feature map.

5. The LCD system fault detection method based on image analysis according to claim 4, characterized in that: The sparse set of low-frequency approximate subband feature reference vectors and the low-frequency approximate subband feature channel feature vector to be enhanced are implicitly compensated and enhanced based on a mask control component to obtain an implicit coding vector of a low-frequency approximate subband feature enhancement component, including: calculating the Poincare distance between the low-frequency approximate subband feature channel feature vector to be enhanced and each low-frequency approximate subband feature reference vector in the sparse set of the low-frequency approximate subband feature reference vector to obtain a low-frequency approximate subband feature spatial modulation parameter matrix; calculating the Poincare distance between the low-frequency approximate subband feature channel feature vector to be enhanced and each low-frequency approximate subband feature reference vector in the sparse set of the low-frequency approximate subband feature reference vector The implicit semantic association between the quantities is obtained to obtain a set of low-frequency approximate sub-band feature semantic association coding matrices; using each low-frequency approximate sub-band feature semantic association coding matrix in the set of low-frequency approximate sub-band feature semantic association coding matrices as a basic mask control component and using the low-frequency approximate sub-band feature spatial modulation parameter matrix as a secondary mask control component, the low-frequency approximate sub-band feature information compensation coding vector between each low-frequency approximate sub-band feature reference vector in the sparse set of the low-frequency approximate sub-band feature reference vector and the feature vector of the low-frequency approximate sub-band feature channel to be enhanced is explicitly modeled and controlled to obtain the implicit coding vector of the low-frequency approximate sub-band feature enhancement component.

6. The LCD system fault detection method based on image analysis according to claim 5, characterized in that: Using each low-frequency approximate subband feature semantic association coding matrix in the set of the low-frequency approximate subband feature semantic association coding matrices as a basic mask control component and using the low-frequency approximate subband feature spatial modulation parameter matrix as a secondary mask control component, the low-frequency approximate subband feature information compensation coding vector between each low-frequency approximate subband feature reference vector in the sparse set of the low-frequency approximate subband feature reference vector and the low-frequency approximate subband feature channel feature vector to be enhanced is explicitly modeled and controlled to obtain the implicit coding vector of the low-frequency approximate subband feature enhancement component, including: calculating the low-frequency approximate subband feature information compensation coding vector between each low-frequency approximate subband feature reference vector and the low-frequency approximate subband feature channel feature vector to be enhanced Positionally differential vectors between vectors are used to obtain a set of low-frequency approximate sub-band feature information compensation coding vectors; based on the basic mask control component and the secondary mask control component, each low-frequency approximate sub-band feature information compensation coding vector in the set of low-frequency approximate sub-band feature information compensation coding vectors is subjected to mask-compensation optimization based on dual-field cooperative modulation to obtain a set of low-frequency approximate sub-band feature information compensation optimization coding vectors; based on the basic mask control component and the secondary mask control component, the set of low-frequency approximate sub-band feature information compensation optimization coding vectors is subjected to global explicit modeling modulation aggregation to obtain the implicit coding vector of the low-frequency approximate sub-band feature enhancement component.

7. The LCD system fault detection method based on image analysis according to claim 6, characterized in that: The method comprises: performing image semantic segmentation on the multi-scale fusion feature map of the LCD panel image to obtain a Mura segmentation region image; and extracting statistical features of the Mura segmentation region image to obtain the Mura region statistical feature coding vector.

8. The LCD system fault detection method based on image analysis according to claim 7, characterized in that: Based on the Mura region statistical feature coding vector, a Mura defect recognition result is obtained, including: inputting the Mura region statistical feature coding vector into a defect classifier based on a SVM model to obtain the Mura defect recognition result.

9. An LCD system fault detection system based on image analysis, characterized in that: include: An image acquisition module, used to acquire LCD panel images acquired by a high-resolution industrial camera; An image preprocessing module, used for performing homomorphic filtering and CLAHE contrast enhancement on the LCD panel image to obtain an enhanced LCD panel image; An image encoding module, used for performing multi-scale feature enhancement fusion on the enhanced LCD panel image to obtain a multi-scale fusion feature map of the LCD panel image, wherein the image encoding module is used for: performing wavelet decomposition and feature extraction on the enhanced LCD panel image to obtain a low-frequency approximate subband feature map, a horizontal detail subband feature map, a vertical detail subband feature map, and a diagonal detail subband feature map; performing feature enhancement fusion processing based on spatial-semantic dual dimensions on the low-frequency approximate subband feature map, the horizontal detail subband feature map, the vertical detail subband feature map, and the diagonal detail subband feature map to obtain the multi-scale fusion feature map of the LCD panel image; a mura region analysis module, used for performing statistical analysis based on the mura segmentation region on the multi-scale fusion feature map of the LCD panel image to obtain a mura region statistical feature coding vector; The defect recognition module is used to obtain a Mura defect recognition result based on the Mura region statistical feature coding vector.

10. The LCD system fault detection system based on image analysis according to claim 9, characterized in that: The defect recognition module is used to: input the Mura region statistical feature encoding vector into a defect classifier based on a SVM model to obtain the Mura defect recognition result.

Images