Liquid crystal screen anomaly detection method and system based on deep learning

Abstract

The invention discloses a liquid crystal screen anomaly detection method and system based on deep learning. The liquid crystal screen anomaly detection method comprises the following steps: collecting a liquid crystal screen image, and carrying out preprocessing and enhancement processing on the collected image to obtain image data after enhancement processing; performing preliminary feature extraction on the image data after enhancement processing, and then performing feature optimization processing to obtain a comprehensive feature set; and constructing an anomaly detection model based on deep learning, and performing automatic classification detection processing on the comprehensive feature set to obtain a detection result, thereby realizing anomaly detection of the liquid crystal screen. According to the method, the image is enhanced, so that the brightness, color and contrast ratio of the image are greatly improved, especially the expressive force on a fine abnormal region is improved, and the subsequent feature extraction and classification are more accurate; and the preliminary features in the image are optimized, so that the abnormity of the liquid crystal screen can be accurately detected.

Description

Technical Field

[0001] The present invention relates to the technical field of liquid crystal screen anomaly detection, and in particular to a liquid crystal screen anomaly detection method and system based on deep learning. Background Art

[0002] With the widespread application of liquid crystal display (LCD) technology, liquid crystal displays have become an important part of modern electronic devices, from smart phones, tablets, laptops to TVs, industrial displays and other devices. The quality of LCD screens directly affects the user experience of the final product. However, due to the complexity of the LCD screen manufacturing process, various defects are prone to occur during the production process, such as bright spots, dark spots, color difference, uneven brightness, etc. These defects not only affect the display effect, but in some cases even cause the entire LCD screen to fail. Therefore, LCD screens need to undergo strict quality inspections before leaving the factory to ensure that their display effects are intact.

[0003] Traditional LCD screen inspection mainly relies on manual visual inspection and simple image processing algorithms. Although manual visual inspection can detect relatively obvious defects, it is inefficient, inaccurate, and easily affected by human factors. Especially in mass production, the efficiency and consistency of manual inspection are difficult to guarantee. Traditional detection algorithms based on image processing usually rely on simple edge detection, color difference analysis and other methods. These methods have limited detection capabilities for complex LCD screen anomalies, especially tiny bright spots, dark spots and subtle color differences. With the improvement of LCD screen resolution and the continuous improvement of user requirements for display effects, traditional detection methods can no longer meet current detection needs.

[0004] However, the above technology has at least the following technical problems: in LCD screen detection, the accuracy of LCD screen image processing is poor and the feature processing optimization is not comprehensive. Summary of the invention

[0005] The technical problem to be solved by the present invention is to overcome the defects of the above-mentioned prior art and provide a method and system for detecting LCD screen abnormalities based on deep learning, which enhances the image to greatly improve the brightness, color and contrast of the image, especially the expressiveness of subtle abnormal areas; optimizes the preliminary features in the image to form a more expressive feature set between the features, so as to realize accurate detection of LCD screen abnormalities.

[0006] To this end, the present invention adopts the following technical solution.

[0007] In a first aspect, the present invention provides a method for detecting abnormalities of a liquid crystal screen based on deep learning, which comprises the steps of: S1. Collecting LCD screen images, preprocessing and enhancing the collected images to obtain enhanced image data; S2. Perform preliminary feature extraction on the enhanced image data, and then perform feature optimization processing to obtain a comprehensive feature set; S3. Construct an anomaly detection model based on deep learning, perform automatic classification and detection processing on the comprehensive feature set, obtain detection results, and realize LCD screen anomaly detection.

[0008] Furthermore, in S1, the pre-processed LCD screen image data is enhanced using a recursive adaptive multi-layer enhancement algorithm to obtain enhanced image data; the recursive adaptive multi-layer enhancement algorithm performs multi-dimensional adjustments to the color and contrast of the LCD screen image to enhance the expressiveness of subtle abnormal areas in the image, providing higher quality input for subsequent processing.

[0009] Furthermore, in the implementation process of the recursive adaptive multi-layer enhancement algorithm, the pre-processed LCD screen image data is subjected to dynamic color mapping and fusion processing; through nonlinear operations of different color channels, the pixel values ​​of the red, green and blue channels are subjected to complex interactive processing to enhance the color level and contrast of the image.

[0010] Furthermore, in the implementation process of the recursive adaptive multi-layer enhancement algorithm, the image data after dynamic color mapping and fusion processing is recursively weighted. Through layer-by-layer recursive calculation, the pixel values ​​are weightedly superimposed layer by layer, and noise amplification is prevented through nonlinear suppression, ensuring that the local information of the image is effectively enhanced without generating excessive artifacts or noise accumulation.

[0011] Furthermore, in the process of implementing the recursive adaptive multi-layer enhancement algorithm, adaptive local contrast enhancement processing is performed on the image after the recursive weighting processing.

[0012] Furthermore, after the recursive weighting and contrast enhancement processing, it is necessary to finally fuse all levels of images to ensure that the pixel information at different levels can complement each other to form a comprehensive enhancement result; the goal of multi-layer fusion is to integrate the pixel values ​​of different enhancement levels into the final output image by weighted averaging.

[0013] Furthermore, the S2 specifically includes: The existing feature engineering technology is used to perform preliminary feature extraction on the enhanced image data to obtain preliminary image feature data, and the preliminary image feature data is subjected to feature analysis and optimization processing using a nonlinear cross-feature interaction and multidimensional projection optimization algorithm to obtain a comprehensive feature set; the nonlinear cross-feature interaction and multidimensional projection optimization algorithm achieves refined processing of image feature data through complex cross-feature interaction processing, multidimensional nonlinear projection and normalization processing, and multidimensional feature cross-optimization.

[0014] Furthermore, in the process of implementing the nonlinear cross-feature interaction and multi-dimensional projection optimization algorithm, the preliminary image feature data is subjected to nonlinear cross-feature interaction processing to form more expressive feature data; after the interaction matrix is ​​generated, in order to map the features that have been interactively processed to a multi-dimensional nonlinear space, the subtle differences in the features are further amplified by nonlinear mapping, and multi-dimensional nonlinear projection processing is performed on them; the projection feature matrix is ​​normalized to eliminate the numerical imbalance between dimensions and ensure that the final output feature matrix has a consistent numerical range in each dimension; and the normalized feature matrix is ​​subjected to multi-dimensional feature cross-optimization processing.

[0015] Furthermore, the S3 specifically includes: using a convolutional neural network to construct an anomaly detection model based on deep learning, processing the comprehensive feature set through the anomaly detection model based on deep learning, and obtaining a detection result, wherein the detection result is a binary classification result, indicating whether the current state of the LCD screen is normal; when an abnormality of the LCD screen is detected, the anomaly detection model based on deep learning further classifies the anomaly type; for the detected anomaly, the anomaly detection model based on deep learning also evaluates its severity.

[0016] In a second aspect, the present invention provides a liquid crystal screen abnormality detection system based on deep learning, which is used to implement the above-mentioned liquid crystal screen abnormality detection method, and comprises: Image processing unit: collects LCD screen images, performs pre-processing and enhancement processing on the collected images, and obtains enhanced image data; Feature extraction and optimization processing unit: perform preliminary feature extraction on the enhanced image data, and then perform feature optimization processing to obtain a comprehensive feature set; Anomaly detection unit: Build an anomaly detection model based on deep learning, perform automatic classification detection processing on the comprehensive feature set, obtain detection results, and realize LCD screen anomaly detection.

[0017] The present invention has the following beneficial effects: 1. Enhance the image, greatly improving the brightness, color and contrast of the image, especially the expression of subtle abnormal areas (such as bright spots, dark spots, color differences), making subsequent feature extraction and classification more accurate.

[0018] 2. Optimize the preliminary features in the image (such as color, texture, edge, etc.) to form a more expressive feature set between the features, ensuring that the comprehensive feature set can cover all the information of the LCD screen image, especially the tiny abnormal areas, to achieve accurate detection of LCD screen abnormalities. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1It is a flow chart of the method for detecting abnormality of a liquid crystal screen based on deep learning of the present invention; Figure 2 This is an architectural diagram of the LCD screen anomaly detection system based on deep learning of the present invention. DETAILED DESCRIPTION

[0020] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0022] See attached Figure 1 , which shows a flow chart of a method for detecting abnormalities of a liquid crystal screen based on deep learning provided by an embodiment of the present invention, the method comprising the following steps: S1. Collecting LCD screen images, preprocessing and enhancing the collected images to obtain enhanced image data.

[0023] Professional technicians select suitable industrial cameras to capture images of LCD screens. The camera should be equipped with appropriate light sources (such as ring light sources, strip light sources) to ensure uniform illumination of the LCD screen image and avoid interference from reflections and shadows. At the same time, full-screen shooting or sub-area shooting should be adopted according to the size of the LCD screen. For large LCD screens, sub-area shooting should be adopted and then image stitching should be performed to ensure complete coverage of the detail area and obtain the LCD screen image.

[0024] The LCD screen image is preprocessed by, for example, denoising, resizing, and standardization to obtain preprocessed LCD screen image data. The preprocessing method is a technical means well known to those skilled in the art and will not be described in detail here.

[0025] The pre-processed LCD screen image data is enhanced using a recursive adaptive multi-layer enhancement algorithm to obtain enhanced LCD screen image data; the recursive adaptive multi-layer enhancement algorithm performs multi-dimensional adjustments on the color and contrast of the LCD screen image to enhance the expressiveness of subtle abnormal areas (such as bright spots, dark spots and color differences) in the image, providing higher quality input for subsequent processing. The specific implementation process is as follows: First, the pre-processed LCD screen image data is dynamically mapped and fused; through nonlinear operations of different color channels, the pixel values ​​of the red, green and blue channels are processed in a complex interactive manner to enhance the color level and contrast of the image, especially in the detection of color difference anomalies. The specific processing formula is as follows:

[0026] in, is the pixel value of the red channel, indicating the pixel at position The red brightness of the image at is the pixel value of the green channel, indicating the pixel at position The green brightness of the image at is the pixel value of the blue channel, indicating the pixel at position The blue brightness of the image at It is the pixel value after dynamic color mapping, which is used to output the enhanced image. The above formula combines logarithmic, square, square root and fractional operations to make the pixel values ​​of different channels produce complex nonlinear interactions, with the purpose of enhancing the color contrast of the image and especially amplifying the changes in color difference.

[0027] Furthermore, the image data after dynamic color mapping and fusion processing is recursively weighted. Through layer-by-layer recursive calculation, the pixel values ​​are weighted and superimposed layer by layer, and nonlinear suppression is used to prevent noise amplification, ensuring that the local information of the image is effectively enhanced without generating excessive artifacts or noise accumulation. The formula for recursive weighted processing is as follows:

[0028] in, It is through The image after layer recursive processing is The pixel value at ; It is The image after layer recursive processing is The pixel value at the first layer is ; It is The weight coefficient of the layer determines the weight of the recursive processing of each layer; It is the first The nonlinear suppression coefficient of the layer controls the strength of recursive enhancement; It is The adjustment coefficient in the sine function of the layer controls the periodic change of the sine wave in each recursive layer; It is The attenuation parameter in the layer exponential function controls the attenuation speed of high-level pixel values ​​in the recursive process; by introducing nonlinear operations (logarithm, sine, exponential), the recursive process is made more layered. Each layer in the recursion is not only weighted according to the pixel value of the previous layer, but also processed by logarithmic and sine functions to make the brightness and color information in the image more delicate; the square in the denominator is combined with the exponential function to ensure that the recursion of each layer will not be infinitely enhanced, avoiding the risk of over-amplification of brightness changes and noise, and at the same time achieving the attenuation of the number of recursive layers through the exponential function. In particular, the specific number of recursive layers is set according to the specific scene, which will not be elaborated here.

[0029] Further, the recursively weighted image Perform adaptive local contrast enhancement. By adaptively adjusting the contrast in the image, areas with large brightness differences (such as bright spots or dark spots on the LCD screen) are made more prominent, while areas with small brightness differences are smoothed to avoid noise amplification caused by over-enhancement. The formula for adaptive local contrast enhancement is as follows:

[0030] in, It is The pixel value of the layer after adaptive contrast enhancement is used to output the enhanced image brightness; It is the average brightness value in the local area of ​​the image, indicating the brightness balance of the local neighborhood; It is the standard deviation of local brightness, which is used to measure the discreteness of brightness values ​​in a local area.

[0031] Furthermore, after multi-layer recursion and contrast enhancement processing, it is necessary to fuse all layers of images to ensure that the pixel information at different levels can complement each other and form a comprehensive enhancement result. The goal of multi-layer fusion is to integrate the pixel values ​​of different enhancement levels into the final output image by weighted averaging, which can maintain the global brightness and contrast while highlighting the local details. The multi-layer fusion formula is as follows:

[0032] in, is the pixel value of the final enhanced image; is the pixel value of each layer of the image after local contrast enhancement; is the weight coefficient of each layer; is the total number of processing layers, used to define the number of recursive layers and the number of fusion layers.

[0033] S2. Perform preliminary feature extraction on the enhanced image data, and then perform feature optimization processing to obtain a comprehensive feature set.

[0034] The existing feature engineering technology is used to perform preliminary feature extraction on the enhanced image data to obtain image feature data such as texture features, color features, and edge features as preliminary image feature data. The preliminary image feature data is subjected to feature analysis and optimization processing using a nonlinear cross-feature interaction and multidimensional projection optimization algorithm to obtain a comprehensive feature set. The nonlinear cross-feature interaction and multidimensional projection optimization algorithm achieves refined processing of image feature data through complex cross-feature interaction processing, multidimensional nonlinear projection and normalization processing, and multidimensional feature cross-optimization.

[0035] The specific implementation process is as follows: First, in order to combine different features and break the independence of each feature, the preliminary image feature data is processed nonlinearly across features to form more expressive feature data. Point-by-point calculations are performed between all preliminary features, and an interaction matrix is ​​constructed through nonlinear operations. Each element in the interaction matrix represents a specific pixel point The specific formula for interaction processing is as follows:

[0036] in, is The interactive feature matrix value of the position is determined by the preliminary feature matrix The nonlinear interaction of the features in It is the first Features at the pixel The value at can be texture, color or edge feature; It is the first Features at the pixel The value at can be texture, color or edge feature; It is a feature With features The interaction coefficient between them controls the strength of the association between these features; is a small constant used to avoid the denominator being zero; is The total number of features at location.

[0037] Furthermore, after the interaction matrix is ​​generated, in order to map the features that have been interactively processed into a multidimensional nonlinear space, nonlinear mapping is used to further amplify the subtle differences of the features, and multidimensional nonlinear projection processing is performed on them. The projection processing formula is as follows:

[0038] in, It is The dimension corresponds to the projected feature matrix at the pixel point The value at is the weight coefficient of the projection dimension, which controls the influence of different projection dimensions on the final eigenvalue; is the scaling parameter of the projection, which controls the expansion of the eigenvalues ​​in different dimensions.

[0039] Furthermore, the projection feature matrix is ​​normalized to eliminate the numerical imbalance between dimensions and ensure that the final output feature matrix has a consistent numerical range in each dimension.

[0040] Furthermore, the normalized feature matrix is ​​subjected to multi-dimensional feature cross optimization. The cross operation maximizes the synergy between features of different dimensions, further improving the distinguishing ability and expression ability of the feature matrix. The cross optimization is performed through a nonlinear cross function, and the formula is as follows:

[0041] in, It is the feature matrix after multi-dimensional optimization at the pixel point The value at It is The feature matrix with enhanced and normalized dimensions is The value at It is The cross-optimization weight of the dimension controls the contribution of the dimension to the final optimized feature; It is Dimensionality parameters controlling the frequency of feature crossovers; is the number of optimized feature dimensions.

[0042] Finally, A comprehensive feature set is constructed to provide feature data for LCD screen anomaly detection.

[0043] S3. Construct an anomaly detection model based on deep learning, perform automatic classification and detection processing on the comprehensive feature set, obtain detection results, and realize LCD screen anomaly detection.

[0044] First, an anomaly detection model based on deep learning is constructed. The construction process is as follows: Model design: A convolutional neural network is used to automatically classify the feature data in the comprehensive feature set of the LCD screen. The comprehensive feature set is used as the input of the anomaly detection model based on deep learning, and the dimension is reduced through the pooling layer, and then classified through the fully connected layer to output the detection results. Model training: In order to allow the anomaly detection model based on deep learning to learn the difference between abnormal and normal LCD screens, a loss function (such as the cross entropy loss function) is defined to measure the gap between the model's predicted value and the true label. Then, the back propagation algorithm is used to update the parameters in the anomaly detection model based on deep learning (such as the weight and bias of the convolution kernel) according to the gradient of the loss value. At the same time, in order to prevent overfitting, a regularization term is added, that is, some neurons are randomly discarded during each layer training to improve the generalization ability of the model. Model evaluation: The model is evaluated by calculating evaluation indicators such as precision, recall rate and F1 score. After the evaluation is passed, the anomaly detection model based on deep learning is built.

[0045] The comprehensive feature set is processed by the anomaly detection model based on deep learning to obtain the test result, which is a binary classification result, indicating whether the current state of the LCD screen is normal. Through the calculation output of the model, the LCD screen image will be classified as "normal" or "abnormal". This result is based on the feature rules learned by the model during training. The input comprehensive feature set, such as color, texture, edge and other features, is analyzed by the anomaly detection model based on deep learning to identify minor defects in the LCD screen. Normal result: If the test result shows that the LCD screen is in normal state, it means that no obvious bright spots, dark spots, color difference and other defects are detected on the screen during the test. After feature classification, the model believes that the comprehensive feature set meets the standard of a normal LCD screen. Abnormal result: If the model judges the result as "abnormal", it means that some features are detected that are different from the normal LCD screen. The abnormality may include bright spots, dark spots or uneven color difference.

[0046] When an LCD screen anomaly is detected, the deep learning-based anomaly detection model further classifies the anomaly type. Based on the multi-classification capabilities of the deep learning anomaly detection model, anomalies can be divided into: Bright spots: The brightness values ​​of certain areas of the LCD screen are significantly higher than normal areas. This phenomenon is caused by pixel failure or backlight anomaly in the liquid crystal layer. The detection results will mark these bright spots and give their specific locations. Dark spots: The brightness values ​​of certain areas on the LCD screen are lower than normal and appear as darker areas. The model can automatically identify these dark spots and locate them by analyzing the brightness distribution of the image. Color difference: Color difference often refers to the color display of certain parts of the LCD screen being inconsistent with the surrounding areas, usually manifested as color shift or saturation abnormality. The model can detect these color difference problems and determine their severity by analyzing the features in the color channel. Uneven brightness: There is a large unevenness in the brightness of different areas on the LCD screen, which is manifested as one area being too bright or too dark. The model will detect uneven brightness and mark the problem areas through a detailed analysis of brightness information.

[0047] Finally, for the detected anomalies, the deep learning-based anomaly detection model will also evaluate their severity. By analyzing the size of the abnormal area, the degree of brightness deviation or color deviation, each anomaly is quantitatively evaluated and the following information is provided to the user: Minor anomalies: Minor bright spots or dark spots. Such problems may have little effect on the display effect of the LCD screen. Moderate anomalies: Obvious bright spots, dark spots or color differences. These problems may have a certain impact on the user experience and require further correction. Severe anomalies: Severe color differences, large areas of uneven brightness or dark spots may cause the LCD screen to not work properly and must be replaced or repaired. Help operators make maintenance or replacement decisions in a timely manner. Realize abnormal detection of LCD screens.

[0048] The beneficial effects of this embodiment are: 1. The image is enhanced by a recursive adaptive multi-layer enhancement algorithm, which greatly improves the brightness, color and contrast of the image, especially the expression of subtle abnormal areas (such as bright spots, dark spots, and color differences). This enhancement process not only strengthens the local details in the image, but also effectively reduces noise and ensures the global consistency of the image, greatly improving the detection ability of tiny defects and making subsequent feature extraction and classification more accurate.

[0049] 2. Through nonlinear cross-feature interaction and multi-dimensional projection optimization algorithms, the preliminary features in the image (such as color, texture, edge, etc.) are deeply interacted and optimized. This algorithm breaks the independence between the features, so that the features form a more expressive feature set through complex nonlinear interaction operations. Further through multi-dimensional projection, normalization and feature cross-optimization, the feature data has higher distinguishing and expressive capabilities, ensuring that the comprehensive feature set can cover all the information of the LCD screen image, especially the tiny abnormal areas.

[0050] Refer to the attached Figure 2 , which shows an architecture diagram of a liquid crystal screen abnormality detection system based on deep learning provided by another embodiment of the present invention, which is used to implement the above-mentioned liquid crystal screen abnormality detection method, and includes: Image processing unit: collects LCD screen images, performs pre-processing and enhancement processing on the collected images, and obtains enhanced image data; Feature extraction and optimization processing unit: perform preliminary feature extraction on the enhanced image data, and then perform feature optimization processing to obtain a comprehensive feature set; Anomaly detection unit: Build an anomaly detection model based on deep learning, perform automatic classification detection processing on the comprehensive feature set, obtain detection results, and realize LCD screen anomaly detection.

[0051] It should be noted that each unit in the above-mentioned LCD screen anomaly detection system based on deep learning can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned units can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules. For the specific definition of an LCD screen anomaly detection system based on deep learning, please refer to the definition of an LCD screen anomaly detection method based on deep learning above. The two have the same functions and effects, which will not be repeated here.

[0052] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A method for detecting abnormalities of a liquid crystal screen based on deep learning, characterized in that: Includes steps: S1. Collecting LCD screen images, preprocessing and enhancing the collected images to obtain enhanced image data; S2. Perform preliminary feature extraction on the enhanced image data, and then perform feature optimization processing to obtain a comprehensive feature set; S3. Construct an anomaly detection model based on deep learning, perform automatic classification and detection processing on the comprehensive feature set, obtain detection results, and realize LCD screen anomaly detection.

2. The method for detecting abnormalities of a liquid crystal screen based on deep learning according to claim 1, characterized in that: In S1, the pre-processed LCD screen image data is enhanced using a recursive adaptive multi-layer enhancement algorithm to obtain enhanced image data; the recursive adaptive multi-layer enhancement algorithm performs multi-dimensional adjustments on the color and contrast of the LCD screen image to enhance the expressiveness of subtle abnormal areas in the image.

3. The method for detecting abnormality of a liquid crystal screen based on deep learning according to claim 2, characterized in that: In the implementation process of the recursive adaptive multi-layer enhancement algorithm, dynamic color mapping and fusion processing are performed on the pre-processed LCD screen image data; through nonlinear operations of different color channels, the pixel values ​​of the red, green and blue channels are processed in a complex interactive manner to enhance the color level and contrast of the image.

4. The method for detecting abnormalities of a liquid crystal screen based on deep learning according to claim 3, characterized in that: In the implementation process of the recursive adaptive multi-layer enhancement algorithm, the image data after dynamic color mapping and fusion processing is recursively weighted. Through layer-by-layer recursive calculation, the pixel values ​​are weightedly superimposed layer by layer, and nonlinear suppression is used to prevent noise amplification, ensuring that the local information of the image is effectively enhanced.

5. The method for detecting abnormalities of a liquid crystal screen based on deep learning according to claim 4, characterized in that: In the process of implementing the recursive adaptive multi-layer enhancement algorithm, adaptive local contrast enhancement is performed on the image after recursive weighting processing.

6. The method for detecting abnormalities of a liquid crystal screen based on deep learning according to claim 5, characterized in that: After recursive weighting and contrast enhancement processing, it is necessary to fuse all levels of images to ensure that the pixel information at different levels can complement each other and form a comprehensive enhancement result; the goal of multi-layer fusion is to integrate the pixel values ​​of different enhancement levels into the final output image by weighted averaging.

7. The method for detecting abnormality of a liquid crystal screen based on deep learning according to claim 1, characterized in that: The S2 specifically includes: The existing feature engineering technology is used to perform preliminary feature extraction on the enhanced image data to obtain preliminary image feature data, and the preliminary image feature data is subjected to feature analysis and optimization processing using a nonlinear cross-feature interaction and multidimensional projection optimization algorithm to obtain a comprehensive feature set; the nonlinear cross-feature interaction and multidimensional projection optimization algorithm achieves refined processing of image feature data through complex cross-feature interaction processing, multidimensional nonlinear projection and normalization processing, and multidimensional feature cross-optimization.

8. The method for detecting abnormalities of a liquid crystal screen based on deep learning according to claim 7, characterized in that: In the process of implementing the nonlinear cross-feature interaction and multi-dimensional projection optimization algorithm, the preliminary image feature data is processed by nonlinear cross-feature interaction to form more expressive feature data; After the interaction matrix is ​​generated, in order to map the features that have been interactively processed to a multidimensional nonlinear space, nonlinear mapping is used to further amplify the subtle differences in the features, and multidimensional nonlinear projection processing is performed on them; the projected feature matrix is ​​normalized to eliminate the numerical imbalance between dimensions and ensure that the final output feature matrix has a consistent numerical range in each dimension; and the normalized feature matrix is ​​subjected to multidimensional feature cross-optimization processing.

9. The method for detecting abnormalities of a liquid crystal screen based on deep learning according to claim 1, characterized in that: The S3 specifically includes: using a convolutional neural network to build an anomaly detection model based on deep learning, processing the comprehensive feature set through the anomaly detection model based on deep learning, and obtaining a detection result, wherein the detection result is a binary classification result, indicating whether the current state of the LCD screen is normal; when an abnormality of the LCD screen is detected, the anomaly detection model based on deep learning further classifies the anomaly type; for the detected anomaly, the anomaly detection model based on deep learning also evaluates its severity.

10. A liquid crystal screen anomaly detection system based on deep learning, used to implement the liquid crystal screen anomaly detection method according to any one of claims 1 to 9, characterized in that: include: Image processing unit: collects LCD screen images, performs pre-processing and enhancement processing on the collected images, and obtains enhanced image data; Feature extraction and optimization processing unit: perform preliminary feature extraction on the enhanced image data, and then perform feature optimization processing to obtain a comprehensive feature set; Anomaly detection unit: Build an anomaly detection model based on deep learning, perform automatic classification detection processing on the comprehensive feature set, obtain detection results, and realize LCD screen anomaly detection.

Images