A display screen light source uniformity testing method and system

Through multiple image acquisition and average processing, distortion correction, adaptive cropping and grayscale conversion, combined with intelligent segmentation and multi-parameter measurement, the neural network model is used to solve the uniformity test of traditional test methods on display surface light source on display screen The problem of accuracy and inefficiency achieves higher test accuracy and adaptability.

CN118735919BActive Publication Date: 2024-11-12SHENZHEN SUNDENAL TECH CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411224549.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-03
Publication Date
2024-11-12
Estimated Expiration
2044-09-03

AI Technical Summary

Technical Problem

Traditional display surface light source uniformity testing methods have insufficient testing accuracy, low efficiency and poor adaptability, making it difficult to deal with display screens of different types and specifications, especially on large-size or high-resolution display screens, and neglecting the backlight module structure, resulting in Test results deviation.

Method used

Multiple image acquisition and average processing, combined with distortion correction and adaptive cropping, adaptive grayscale conversion and dynamic window scanning, multi-level heat map analysis and intelligent segmentation, optimize test partitions according to the backlight module layout, and perform multi-parameters Combined sampling measurement and high-dimensional feature data dimensionality reduction, use dynamically optimized neural network models for failure mode recognition.

Benefits of technology

It significantly improves the accuracy and efficiency of the uniformity test of display surface light source, reduces manual intervention, improves the interpretability and accuracy of the test results, and adapts to different brightness distribution characteristics and backlight module design.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118735919B_ABST
    Figure CN118735919B_ABST
Patent Text Reader

Abstract

The present application relates to the technical field of display screen testing, and discloses a display screen surface light source uniformity testing method and system, the method comprising: performing multiple image acquisitions on the display screen and taking an average value to obtain an original image, and performing distortion correction and adaptive cropping processing to obtain a standardized image; performing adaptive grayscale conversion and dynamic window scanning to obtain target brightness distribution data and performing multi-level thermal map analysis to obtain a hierarchical brightness distribution result; performing intelligent segmentation to obtain an initial partitioning scheme and optimizing it to obtain an adaptive test partition; performing multi-parameter joint sampling measurement on the adaptive test partition to obtain high-dimensional feature data, and performing dimensionality reduction processing on the high-dimensional feature data to obtain a target performance indicator set; inputting the target performance indicator set into a dynamically optimized neural network model to perform fault mode recognition to obtain a fault mode recognition result, thereby improving the accuracy of the display screen surface light source uniformity test.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of display screen testing, and in particular to a method and system for testing the uniformity of a light source on a display screen. Background Art

[0002] With the continuous development of display technology, high-quality display screens are increasingly used in all walks of life. Among them, the uniformity of the light source on the display screen directly affects the display effect and user experience. Traditional display screen light source uniformity testing methods often have problems such as insufficient test accuracy, low efficiency and poor adaptability. These methods usually use fixed test parameters and simple image processing algorithms, which are difficult to deal with display screens of different types and specifications. In addition, traditional methods often require a lot of time and human resources when dealing with large-size or high-resolution display screens, and cannot meet the needs of modern industries for high-efficiency and high-precision testing.

[0003] Another prominent problem is that existing test methods often lack consideration of the display backlight module structure. Different backlight module designs will lead to differences in light source distribution, and ignoring this may lead to deviations in test results. At the same time, traditional methods are often difficult to accurately identify and quantify when faced with complex optical phenomena, such as local highlights and dark corners, which limits the reliability and practicality of test results. Summary of the invention

[0004] The present application provides a display screen light source uniformity testing method and system, thereby improving the accuracy of display screen light source uniformity testing.

[0005] The first aspect of the present application provides a display screen light source uniformity testing method, the display screen light source uniformity testing method comprising:

[0006] Capturing images of the display screen multiple times and taking an average value to obtain an original image, and performing distortion correction and adaptive cropping processing on the original image to obtain a standardized image;

[0007] Performing adaptive grayscale conversion and dynamic window scanning on the standardized image to obtain target brightness distribution data, and performing multi-level heat map analysis on the target brightness distribution data to obtain a hierarchical brightness distribution result;

[0008] Intelligently segmenting the hierarchical brightness distribution result to obtain an initial partitioning scheme, and optimizing the initial partitioning scheme according to the display screen backlight module layout information to obtain an adaptive test partition;

[0009] Performing multi-parameter joint sampling measurement on the adaptive test partition to obtain high-dimensional feature data, and performing dimensionality reduction processing on the high-dimensional feature data to obtain a target performance indicator set;

[0010] The target performance indicator set is input into a dynamically optimized neural network model to perform fault mode recognition to obtain a fault mode recognition result.

[0011] The second aspect of the present application provides a display screen light source uniformity testing system, the display screen light source uniformity testing system comprising:

[0012] The acquisition module is used to acquire multiple images of the display screen and take an average value to obtain an original image, and to perform distortion correction and adaptive cropping processing on the original image to obtain a standardized image;

[0013] An analysis module, used for performing adaptive grayscale conversion and dynamic window scanning on the standardized image to obtain target brightness distribution data, and performing multi-level thermal map analysis on the target brightness distribution data to obtain a hierarchical brightness distribution result;

[0014] A segmentation module, used for intelligently segmenting the hierarchical brightness distribution result to obtain an initial partitioning scheme, and optimizing the initial partitioning scheme according to the display screen backlight module layout information to obtain an adaptive test partition;

[0015] A processing module, used for performing multi-parameter joint sampling measurement on the adaptive test partition to obtain high-dimensional feature data, and performing dimensionality reduction processing on the high-dimensional feature data to obtain a target performance indicator set;

[0016] The identification module is used to input the target performance indicator set into the dynamically optimized neural network model to perform fault mode identification and obtain a fault mode identification result.

[0017] The third aspect of the present application provides an electronic device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the electronic device executes the above-mentioned display screen light source uniformity test method.

[0018] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned display screen light source uniformity testing method.

[0019] Compared with the prior art, the present application has the following beneficial effects: through multiple image acquisition and averaging processing, combined with distortion correction and adaptive cropping, the quality and accuracy of the original image are significantly improved; the use of adaptive grayscale conversion and dynamic window scanning technology enables the test method to adapt to display screens with different brightness distribution characteristics; combined with intelligent segmentation algorithms and display screen backlight module layout information, a more reasonable and accurate test partition is achieved, and the pertinence and efficiency of the test are improved. Through multi-parameter joint sampling measurement and dimensionality reduction processing of high-dimensional feature data, the various performance indicators of the display screen are comprehensively evaluated; the dynamically optimized neural network model is used for fault mode recognition, which improves the accuracy and efficiency of fault diagnosis; and by introducing automated and intelligent processing procedures, manual intervention is greatly reduced, and test efficiency is improved. Through multi-level thermal map analysis and visualization technology, the interpretability of the test results is improved; through standardized data processing procedures and intelligent parameter adjustment, the accuracy of the display screen light source uniformity test is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0021] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical contents disclosed by the present invention.

[0022] Figure 1 It is a flow chart of a method for testing uniformity of light source of a display screen provided by an embodiment of the present invention;

[0023] Figure 2 The figure is a schematic block diagram of the structure of a display screen light source uniformity testing system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of 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.

[0025] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0026] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0027] It should be further understood that the term "and / or" used in the specification and appended claims of this application refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Figure 1 In the embodiment of the present application, an embodiment of a method for testing uniformity of a display light source includes:

[0028] Step 100: Capture images of the display screen multiple times and take an average value to obtain an original image, and perform distortion correction and adaptive cropping processing on the original image to obtain a standardized image;

[0029] It is understandable that the execution subject of the present application may be a display screen light source uniformity test system, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0030] Specifically, N images are collected from the display screen to obtain N original images. The N original images collected are subjected to pixel-level arithmetic averaging to obtain a comprehensive average original image, thereby reducing the influence of random noise and obtaining more accurate information on the brightness distribution of the display screen. The average original image is subjected to edge detection to obtain the edge contour of the display screen. The perspective transformation matrix is ​​calculated through the detected edge contour, and the matrix is ​​used to describe the perspective distortion in the image. The homography matrix is ​​solved by using the four-point correspondence method to correct the geometric deformation in the image and obtain the distortion correction parameters. Based on the distortion correction parameters, the average original image is subjected to affine transformation to correct the distorted part in the image and obtain a corrected image. The corrected image is subjected to adaptive threshold segmentation to generate a display area mask, which is used to distinguish between the display area and the non-display area. The corrected image is subjected to region growing processing through the display area mask, which helps to accurately determine the boundary of the display area. The minimum bounding rectangle is calculated based on the precise display area boundary, and the specific coordinates of the cropping area are determined to ensure that the image area to be processed subsequently completely covers the display area without omission. The corrected image is cropped using the calculated cropping region coordinates to generate a preliminary standardized image. The preliminary standardized image is resampled by bilinear interpolation to adjust the image resolution to meet the requirements of subsequent processing, and to smooth the image to eliminate pixel discontinuity problems caused by cropping and other processing. Through the above steps, a standardized image is obtained.

[0031] Step 200, performing adaptive grayscale conversion and dynamic window scanning on the standardized image to obtain target brightness distribution data, and performing multi-level heat map analysis on the target brightness distribution data to obtain a hierarchical brightness distribution result;

[0032] Specifically, the standardized image is subjected to histogram equalization to obtain an initial grayscale image, and the contrast of the image is enhanced to make the brightness distribution more uniform and obvious. According to the brightness distribution characteristics of the initial grayscale image, an adaptive gamma value is calculated, and the value is used to control the nonlinear brightness transformation of the image. By performing a nonlinear transformation on the initial grayscale image based on the adaptive gamma value, the brightness difference in the image is enhanced to obtain an enhanced grayscale image. The enhancement process helps to highlight the area with uneven brightness on the display screen, making it easier to detect in subsequent analysis. The initial window size and step size are set for the enhanced grayscale image to obtain initial scanning parameters. Based on these parameters, the enhanced grayscale image is scanned with a sliding window, and the average brightness value in each window is calculated to obtain rough brightness distribution data. The rough brightness distribution data is subjected to variance analysis to obtain refined brightness distribution data, which helps to identify the detailed changes in the brightness distribution and makes the brightness data more accurate. The refined brightness distribution data is organized into a multidimensional array to obtain target brightness distribution data, which represents the brightness characteristics of each area of ​​the display screen. The target brightness distribution data is subjected to multi-scale Gaussian filtering to obtain smooth brightness data of different scales. The role of multi-scale Gaussian filtering is to smooth data, remove noise, and retain the main features of brightness distribution. The difference map between adjacent scales is calculated to obtain multi-scale difference data, which reflects the brightness changes at different scales. A multi-layer heat map is generated based on the multi-scale difference data, and the multi-layer heat map is superimposed by scale to obtain a hierarchical brightness distribution result. In this way, the brightness distribution on the display screen is displayed at different levels, making the detection and analysis of uneven brightness areas more intuitive and accurate.

[0033] Step 300: Intelligently segment the hierarchical brightness distribution result to obtain an initial partitioning scheme, and optimize the initial partitioning scheme according to the display backlight module layout information to obtain an adaptive test partition;

[0034] It should be noted that the hierarchical brightness distribution result is subjected to multi-scale superpixel segmentation, and the image is divided into multiple regions with similar features to obtain an initial superpixel set. The average brightness value and brightness variance of each superpixel in the initial superpixel set are calculated to obtain a superpixel feature vector set. The feature vector set includes the brightness features of each superpixel, which helps to identify brightness differences. K-means clustering is performed on the superpixel feature vector set to divide the superpixels into K brightness categories. Through clustering analysis, superpixels with different brightness features are assigned to different categories, which helps to better understand the overall structure of the brightness distribution. According to the brightness category, the initial superpixel set is merged to obtain a rough partitioning result. Regional growth is performed on the rough partitioning result to refine the partition boundary so that it is more in line with the actual brightness change, and a more refined optimized partitioning result is obtained. The shape complexity of each partition in the optimized partitioning result is calculated, and the ratio of the square of the perimeter to the area is used as an indicator of shape complexity to obtain a shape complexity set. Through this indicator, it is evaluated whether the shape of the partition is too complex or irregular. According to the shape complexity set, the boundary of the optimized partitioning result is simplified to obtain a simplified partitioning result. Boundary simplification helps to eliminate unnecessary details and make the partition boundaries smoother and more regular. Read the layout information of the display backlight module, including the position of LED lamp beads and the parameters of the optical diffusion film. Based on this information, a theoretical light intensity distribution model is constructed to obtain a theoretical partition template. The theoretical light intensity distribution model simulates the light output characteristics of the backlight module and provides an ideal light intensity distribution. The simplified partition results are subjected to morphological operations with the theoretical partition template to obtain an adaptive test partition. Morphological operations include morphological expansion, corrosion and other operations, which help to adjust the shape of the partition to make it more consistent with the theoretical light intensity distribution.

[0035] Step 400: Perform multi-parameter joint sampling measurement on the adaptive test partition to obtain high-dimensional feature data, and perform dimensionality reduction processing on the high-dimensional feature data to obtain a target performance indicator set;

[0036] Specifically, a stratified random sampling strategy is performed on the adaptive test partitions to obtain a sampling point set. The stratified random sampling strategy is intended to ensure that the sampling points are evenly distributed in different areas of the display screen to fully reflect the optical performance of each area. A series of measurements are performed on the sampling point set, including brightness measurement, chromaticity measurement, contrast measurement, flicker frequency measurement, and response time measurement. In the brightness measurement, the brightness value of each sampling point is obtained, and the brightness uniformity index is calculated based on these data to reflect the uniformity of the brightness of the display screen; in the chromaticity measurement, the chromaticity data set is collected, and the chromaticity uniformity index is calculated by analyzing these data to evaluate the consistency of the color display of the display screen; in the contrast measurement, the contrast data set is obtained, and the contrast uniformity index is calculated to measure the stability of the contrast between different areas of the display screen. At the same time, in the flicker frequency measurement, the flicker data set is collected, and the flicker index is calculated, which is used to judge the degree of flicker phenomenon of the display screen; in the response time measurement, the response time data set is recorded, and the average response time is calculated to evaluate the response speed of the display screen to the signal input. These measurement results are combined to form a comprehensive high-dimensional feature vector, which contains the performance indicators of the display screen in multiple dimensions. Perform principal component analysis on the high-dimensional feature vector, reduce the dimension of the feature through dimensionality reduction processing, and obtain the reduced feature vector. The reduced feature vector represents the condensed information of the main performance indicators of the display. Standardize the reduced feature vector so that different features have the same dimension and scale, and obtain a standardized feature vector. Perform Fisher discriminant analysis on the standardized feature vector to obtain the target performance indicator set. Fisher discriminant analysis is a linear discriminant analysis method used to maximize the ratio of between-class variance to within-class variance, thereby achieving the best class separation effect.

[0037] Step 500: Input the target performance indicator set into the dynamically optimized neural network model to perform fault mode recognition and obtain a fault mode recognition result.

[0038] Specifically, the target performance indicator set is normalized to convert indicators of different dimensions into the same standardized input vector. The standardized input vector is input into the input layer of the dynamically optimized neural network model, and the input layer output is obtained after processing. The input layer output is passed to the first hidden layer, and the weight matrix and bias vector are processed to obtain the linear output. The ReLU (Rectified Linear Unit) activation function is applied to the linear output to obtain the activation output of the first hidden layer. The ReLU function enhances the nonlinear ability of the model by zeroing negative values, so that the neural network can better learn complex patterns. The activation output of the first hidden layer is passed to the dynamic attention layer, which identifies important features in the input data by calculating the attention weight. The attention weight reflects the model's attention to different features, and then the activation output of the first hidden layer and the attention weight are weighted and summed to obtain the attention layer output. The weighting mechanism helps the model focus on key features and improve the accuracy of fault mode recognition. The output of the attention layer is input into the second hidden layer, and the output of the second hidden layer is obtained by processing through linear transformation and ReLU activation function. The linear transformation further explores the relationship between features, while the ReLU activation function continues to improve the nonlinear representation ability of the model. Perform a dynamic residual connection on the output of the second hidden layer and add it to the activation output of the first hidden layer to obtain the residual output. The residual connection reduces the gradient vanishing problem by adding input information while retaining the previous layer information, effectively improving training efficiency and performance. The residual output is input to the output layer and the original output vector is obtained after linear transformation. The Softmax function is applied to the original output vector to obtain the fault mode probability distribution. The Softmax function converts the original output into a probability distribution, and each element represents the probability that the input data belongs to a certain fault mode. According to the fault mode probability distribution, the fault mode recognition result is generated.

[0039] The activation output of the first hidden layer is input to the dynamic attention layer. The activation output is transposed to obtain a transposed matrix. The transposed matrix retains the feature information of the original activation output, but changes its arrangement to make it suitable for subsequent matrix operations. The transposed matrix is ​​multiplied by the first attention weight matrix to obtain an intermediate result, which preliminarily extracts and weights the important information in the input features. The intermediate result is analyzed by the hyperbolic tangent function to map the input data to the range of -1 to 1, enhance the model's ability to capture the nonlinear characteristics of the input data, and obtain a nonlinear transformation result. The nonlinear transformation result is dot-producted with the attention vector to calculate the original attention score. The original attention score reflects the importance of each feature in the final decision. The original attention score is normalized to ensure that the proportional relationship between the score values ​​is preserved, and at the same time it is converted into a more standardized range to obtain the attention weight. The attention weight is expanded into a diagonal matrix to obtain the second attention weight matrix. The activation output of the first hidden layer is matrix multiplied by the second attention weight matrix to obtain a weighted feature matrix. The input features are reweighted by the attention weights to enhance the model's attention to key features. The weighted feature matrix is ​​summed along the feature dimension to obtain the context vector. The context vector is a compact representation of the weighted feature information. The context vector is concatenated with the activation output of the first hidden layer to obtain the enhanced feature vector, which allows the model to use both the original information and the information enhanced by the attention mechanism. The enhanced feature vector is transformed into a fully connected layer to obtain the attention layer output.

[0040] In the embodiment of the present application, the quality and accuracy of the original image are significantly improved through multiple image acquisition and averaging processing, combined with distortion correction and adaptive cropping. The adaptive grayscale conversion and dynamic window scanning technology are used to enable the test method to adapt to display screens with different brightness distribution characteristics. The intelligent segmentation algorithm and the display screen backlight module layout information are combined to achieve a more reasonable and accurate test partition, which improves the pertinence and efficiency of the test. Through multi-parameter joint sampling measurement and dimensionality reduction processing of high-dimensional feature data, the various performance indicators of the display screen are comprehensively evaluated. The dynamically optimized neural network model is used for fault mode recognition to improve the accuracy and efficiency of fault diagnosis. By introducing automated and intelligent processing procedures, manual intervention is greatly reduced and test efficiency is improved. Through multi-level thermal map analysis and visualization technology, the interpretability of the test results is improved. Through standardized data processing procedures and intelligent parameter adjustment, the accuracy of the display screen surface light source uniformity test is improved.

[0041] In a specific embodiment, the process of executing step 100 may specifically include the following steps:

[0042] Capture images of the display screen N times to obtain N original images, and perform pixel-level arithmetic averaging on the N original images to obtain an average original image;

[0043] Perform edge detection on the average original image to obtain the edge contour of the display screen, and calculate the perspective transformation matrix based on the edge contour of the display screen to obtain the distortion correction parameters, and use the four-point correspondence method to solve the homography matrix;

[0044] Performing affine transformation on the average original image based on the distortion correction parameter to obtain a corrected image, and performing adaptive threshold segmentation on the corrected image to obtain a display area mask;

[0045] Performing region growth on the rectified image according to the display region mask to obtain the precise display region boundary, and calculating the minimum circumscribed rectangle based on the precise display region boundary to obtain the cropping region coordinates;

[0046] The rectified image is cropped according to the coordinates of the cropping region to obtain a preliminary standardized image, and the preliminary standardized image is resampled by bilinear interpolation to obtain a standardized image.

[0047] Specifically, the display screen is imaged N times to obtain N original images. Image acquisition can be achieved by setting fixed lighting conditions and the same camera position to reduce the impact of external factors on image quality. The N original images are processed by pixel-level arithmetic averaging to reduce noise and other random errors and obtain an average original image. Assume that the pixel points of each image are , the pixel value of the i-th image is , then the average pixel value of the original image It can be expressed as: ;

[0048] in, is the number of images, is the i-th image at coordinates The pixel value at . Perform edge detection on the average original image to determine the edge contour of the display screen. Commonly used image processing algorithms such as Canny, Sobel or Laplacian can be used for edge detection. After the edge of the display screen is detected, the perspective transformation matrix is ​​calculated according to the edge contour. The four corner points of the display screen are selected as control points, and the homography matrix H is solved by the four-point correspondence method. The transformation matrix is ​​used to correct the perspective deformation caused by the camera angle or the installation position of the display screen. The calculation of the homography matrix H can be performed by the least squares method, and the formula is: ;

[0049] in, and are the coordinates of the original image and the corrected image, s is the scale factor, and H is the 3x3 homography matrix. Using the obtained perspective transformation matrix H, the average original image is affine transformed to obtain the corrected image. The affine transformation corrects the geometric distortion of the image so that the spatial position of each pixel in the image is more consistent with the position on the actual display screen. The corrected image is adaptively thresholded and the threshold is dynamically adjusted according to the local information of the image to segment the image into foreground (display area) and background (non-display area). The corrected image is region grown according to the display area mask. Starting from the known seed point, the area is expanded according to the similarity between pixels until no more pixels meet the conditions, thereby obtaining the display area boundary, which can well adapt to the actual shape of the display screen. Based on this boundary, the minimum circumscribed rectangle is calculated to determine the specific coordinates of the cropping area. According to the obtained cropping area coordinates, the corrected image is cropped to generate a preliminary standardized image, removing irrelevant parts of the image and retaining only the effective display area of ​​the display screen. The cropped image may have inconsistent resolution or pixel discontinuity due to the transformation and cropping process, and bilinear interpolation resampling is required. Bilinear interpolation is an image resampling technique that calculates new pixel values ​​by weighted averaging the values ​​of adjacent pixels to smooth the image and improve the resolution. The standardized image is obtained to ensure that the image has consistent resolution and quality for subsequent analysis and processing.

[0050] In a specific embodiment, the process of executing step 200 may specifically include the following steps:

[0051] Perform histogram equalization on the standardized image to obtain an initial grayscale image, and calculate the adaptive gamma value based on the brightness distribution characteristics of the initial grayscale image;

[0052] Based on the adaptive gamma value, the initial grayscale image is nonlinearly transformed to obtain an enhanced grayscale image;

[0053] The initial window size and step size are set for the enhanced grayscale image to obtain the initial scanning parameters, and the enhanced grayscale image is scanned by a sliding window according to the initial scanning parameters, and the average brightness value in each window is calculated to obtain the rough brightness distribution data;

[0054] Perform variance analysis on the rough brightness distribution data to obtain refined brightness distribution data, and organize the refined brightness distribution data into a multidimensional array to obtain target brightness distribution data;

[0055] Perform multi-scale Gaussian filtering on the target brightness distribution data to obtain smooth brightness data of different scales, and calculate the difference map between adjacent scales to obtain multi-scale difference data;

[0056] A multi-layer heat map is generated based on the multi-scale differential data, and the multi-layer heat map is superimposed according to the scale to obtain a hierarchical brightness distribution result.

[0057] Specifically, the standardized image is processed by histogram equalization. By mapping the pixel values ​​of the image to a new brightness range, the contrast of the image is adjusted so that the brightness values ​​in the image are evenly distributed and the details of the image are enhanced. Assuming that the gray value in the standardized image is L, the formula for histogram equalization is: ;

[0058] in, is the new pixel value, and are the number of rows and columns of the image, respectively. The original gray level The frequency distribution of is the gray level of the current pixel. The adaptive gamma value is calculated based on the brightness distribution characteristics of the initial grayscale image. The calculation of the adaptive gamma value usually depends on the distribution of the brightness histogram of the image. The purpose of gamma correction is to perform nonlinear transformation on the image to better adapt to the visual characteristics of the human eye. The basis for selecting the adaptive gamma value is to enhance the details and contrast of the image so that the information of the dark and bright parts of the image can be appropriately enhanced. Generally, a gamma value less than 1 will brighten the image, and a gamma value greater than 1 will darken the image. The formula is expressed as: ;

[0059] in, is the pixel value after gamma correction, is the pixel value of the initial grayscale image, is the adaptive gamma value. Set the initial window size and step size for the enhanced grayscale image obtained for sliding window scanning. Perform local analysis in the image to capture the details of local brightness changes. The choice of window size and step size affects the fineness and amount of calculation of the scan. Generally, the window size can be set to 8x8 or 16x16 pixels, and the step size is set to 1 / 2 or 1 / 4 of the window size according to demand. During the sliding window scanning process, the average brightness value in each window is calculated to generate rough brightness distribution data. Perform variance analysis on the rough brightness distribution data to reveal the brightness changes in the data. By calculating the variance of the brightness in each window, the areas with large brightness changes are identified. The variance calculation formula is: ;

[0060] in, is the variance, is the number of pixels in the window, is the brightness value of each pixel, is the average brightness value in the window. The refined brightness distribution data obtained by variance analysis can more accurately describe the brightness changes in the image. The refined brightness distribution data is organized into a multidimensional array to obtain the target brightness distribution data. Multi-scale Gaussian filtering is performed on the target brightness distribution data. Gaussian filtering is a smoothing operation used to remove noise from an image while retaining the main features of the image. The application of Gaussian filters of different scales can produce brightness data with different degrees of smoothness, which reflects the characteristics of the image at different resolutions. ;

[0061] in, is a Gaussian filter, is the standard deviation, which determines the scale of the filter. The difference map between adjacent scales is calculated. Multi-scale difference data can reveal the brightness changes at different scales. The difference map is generated by subtracting the smooth brightness data of adjacent scales. These difference data help to identify edges and other important features in the image. Multi-layer heat maps are generated based on multi-scale difference data, and these multi-layer heat maps are superimposed by scale to obtain hierarchical brightness distribution results. The generation of multi-layer heat maps can represent the brightness differences in different areas through color coding. The visual representation helps to observe and analyze the brightness uniformity of the display.

[0062] In a specific embodiment, the process of executing step 300 may specifically include the following steps:

[0063] Perform multi-scale superpixel segmentation on the hierarchical brightness distribution results to obtain an initial superpixel set;

[0064] The average brightness value and brightness variance of each superpixel are calculated for the initial superpixel set to obtain a superpixel feature vector set, and K-means clustering is performed on the superpixel feature vector set to obtain K brightness categories;

[0065] The initial superpixel set is merged according to the brightness category to obtain a rough partitioning result, and the rough partitioning result is subjected to region growing to obtain an optimized partitioning result;

[0066] Calculate the shape complexity of each partition in the optimized partition result, use the ratio of the square of the perimeter to the area as the complexity index, and obtain the shape complexity set;

[0067] The optimized partitioning result is simplified according to the shape complexity set to obtain a simplified partitioning result, and the display backlight module layout information is read, including the LED lamp bead position and optical diffusion film parameters;

[0068] According to the position of LED lamp beads and the parameters of optical diffusion film, a theoretical light intensity distribution model is constructed to obtain a theoretical partition template. The simplified partition result is subjected to morphological operation with the theoretical partition template to obtain an adaptive test partition.

[0069] Specifically, the hierarchical brightness distribution results are segmented by multi-scale superpixels to divide the image into a set of tightly connected pixels to obtain an initial superpixel set. Superpixels have similar color, brightness or texture features, which can effectively reduce the computational complexity of image processing while retaining key image information. Common superpixel generation algorithms include SLIC (Simple Linear Iterative Clustering), which generates superpixel sets by iteratively optimizing the location of cluster centers. During the segmentation process, multiple scale parameters are selected to cover the image with superpixels of different sizes to capture brightness changes at different levels. The resulting initial superpixel set provides a hierarchical structure of the image, which is helpful for subsequent analysis. The average brightness value and brightness variance are calculated for each superpixel to form a superpixel feature vector set. The average brightness value and brightness variance of each superpixel represent the overall brightness level and brightness consistency of the superpixel area, respectively. Through these features, the brightness distribution characteristics of the superpixel are identified. K-means clustering is used to classify the superpixel feature vector set and divide the superpixels into K brightness categories. The K-means clustering algorithm classifies similar feature vectors into one category by minimizing the objective function of intra-class variance. The choice of K value depends on the actual application scenario and can be determined through experiments. After clustering, the initial superpixel set is merged according to the brightness category to form a rough partition result. These partitions reflect the brightness characteristics of different areas of the display screen and provide preliminary regional divisions. Based on the rough partition results, regional growth is performed to obtain more accurate partitions. The region growing method uses the similarity between pixels to expand the region from the boundary of the initial partition until a preset similarity threshold is reached. In this way, an optimized partition result is obtained, and the partition boundary is clearer and more accurate. The shape complexity of each partition in the optimized partition result is calculated. The shape complexity can be measured by the ratio of the square of the perimeter to the area. The shape complexity reflects the complexity of the partition boundary, and a higher complexity usually means that the boundary is more irregular. According to the shape complexity set, the boundary of the optimized partition result is simplified, and redundant details are removed to make the partition boundary more regular and smooth. Read the layout information of the display backlight module, including the position of the LED lamp beads and the parameters of the optical diffusion film. This information helps to understand the light source distribution and optical characteristics of the display. Based on this information, a theoretical light intensity distribution model is constructed, which simulates the light output characteristics and diffusion effect of the LED light source and generates a theoretical partition template. The theoretical partition template provides a reference for the ideal light intensity distribution. The simplified partition results are subjected to morphological operations with the theoretical partition template to obtain adaptive test partitions. Morphological operations include morphological dilation and erosion operations, which can adjust the shape and size of the partitions to make the partitions more consistent with the actual light intensity distribution. The generation of adaptive test partitions is to ensure that the test can accurately reflect the uniformity of the light source of the display.

[0070] In a specific embodiment, the process of executing step 400 may specifically include the following steps:

[0071] Perform stratified random sampling strategy analysis on the adaptive test partitions to obtain a set of sampling points;

[0072] The brightness of the sampling point set is measured to obtain a brightness data set, and a brightness uniformity index is calculated; the chromaticity of the sampling point set is measured to obtain a chromaticity data set, and a chromaticity uniformity index is calculated; the contrast of the sampling point set is measured to obtain a contrast data set, and a contrast uniformity index is calculated; the flicker frequency of the sampling point set is measured to obtain a flicker data set, and a flicker index is calculated; the response time of the sampling point set is measured to obtain a response time data set, and an average response time is calculated;

[0073] Combining the brightness uniformity index, the chromaticity uniformity index, the contrast uniformity index, the flicker index and the average response time into a high-dimensional feature vector;

[0074] Perform principal component analysis on the high-dimensional feature vector to obtain a feature vector after dimension reduction, and perform standardization on the feature vector after dimension reduction to obtain a standardized feature vector;

[0075] Fisher discriminant analysis is performed on the standardized feature vector to obtain the target performance indicator set.

[0076] Specifically, a stratified random sampling strategy is analyzed for the adaptive test partitions, and representative sampling points are selected from each test partition to form a sampling point set. The implementation of the stratified random sampling strategy includes determining the number of sampling points in each partition and then randomly selecting these points within each partition, which helps to capture the optical characteristics of different areas on the display screen and avoid data deviation and non-uniformity. The sampling point set is measured for brightness to obtain a brightness data set. The brightness uniformity index is calculated. The brightness uniformity index is used to evaluate the uniformity of brightness distribution between different areas of the display screen. The sampling point set is measured for chromaticity to obtain a chromaticity data set. Chromaticity measurement involves the color accuracy of the display screen, and a colorimeter is usually used to measure the color coordinates on the CIE 1931 chromaticity diagram. These data are analyzed to calculate the chromaticity uniformity index. The chromaticity uniformity index can be calculated by measuring the deviation of the chromaticity of each sampling point from the target chromaticity, and then taking the average or maximum value of these deviations to evaluate the consistency of the color performance of the display screen. The sampling point set is measured for contrast. Contrast determines the visibility and image clarity of the display screen under different light. The contrast data set can be obtained by measuring the brightness ratio of the white and black areas on the screen. The contrast uniformity index can be obtained by calculating the uniformity of these ratios. Common calculation methods include the ratio of the maximum contrast to the minimum contrast, or the calculation of the contrast standard deviation. In order to detect the flicker phenomenon of the display screen, the flicker frequency of the sampling point set is measured. The flicker frequency measurement can use a high frame rate camera or a flicker meter to capture the flicker characteristics of the display screen, and the flicker index can be calculated by analyzing the changes in the measured flicker frequency. At the same time, the response time is measured. The response time refers to the time required for the display screen to change from one state to another, and the grayscale response time (GTG) is usually used as a measurement standard. The response time data set can be obtained by measuring the response speed of the display screen under different brightness changes, and the average response time is the average of these data. The brightness uniformity index, chromaticity uniformity index, contrast uniformity index, flicker index and average response time are combined to form a high-dimensional feature vector. The high-dimensional feature vector is subjected to principal component analysis. By converting the original feature vector into a set of linearly unrelated new features (principal components), the dimension is reduced while retaining the main information of the data. The reduced feature vectors are standardized so that each feature has the same dimension and range, usually standardized to zero mean and unit variance. Fisher discriminant analysis (FDA) is performed on the standardized feature vectors to obtain the target performance indicator set. Fisher discriminant analysis is a linear discriminant method used to maximize the ratio of between-class variance to within-class variance, thereby achieving optimal classification of data. The formula for FDA is: ;

[0077] in, is the projection vector, is the between-class scatter matrix, is the intra-class scatter matrix. By optimizing the projection vector ,maximize , to achieve the best classification effect.

[0078] In a specific embodiment, the process of executing step 500 may specifically include the following steps:

[0079] Normalizing the target performance indicator set to obtain a standardized input vector, and inputting the standardized input vector into the input layer of the dynamically optimized neural network model for processing to obtain an input layer output;

[0080] The input layer output is processed by the weight matrix and bias vector of the first hidden layer to obtain the linear output of the first hidden layer, and the activation function ReLU operation is performed on the linear output of the first hidden layer to obtain the activation output of the first hidden layer;

[0081] Input the activation output of the first hidden layer into the dynamic attention layer to obtain the attention weight, and perform weighted summation of the activation output of the first hidden layer and the attention weight to obtain the output of the attention layer;

[0082] The output of the attention layer is input into the second hidden layer, and after linear transformation and ReLU activation, the output of the second hidden layer is obtained;

[0083] Perform a dynamic residual connection on the output of the second hidden layer and add it to the activation output of the first hidden layer to obtain the residual output, and then input the residual output into the output layer and obtain the original output vector after linear transformation;

[0084] The Softmax function is performed on the original output vector to obtain the fault mode probability distribution, and the fault mode recognition result is generated according to the fault mode probability distribution.

[0085] Specifically, the target performance indicator set is normalized to convert the performance indicators of different scales into a relatively consistent range. Usually, the data is standardized to the interval [0, 1] or [-1, 1] to obtain a standardized input vector. The standardized input vector is input to the input layer of the dynamically optimized neural network model. The input layer of the neural network is responsible for receiving and processing all input features and passing them to the subsequent hidden layers. The output of the input layer will continue to be passed to the first hidden layer. The first hidden layer obtains a linear output through the operation of the weight matrix and the bias vector, and its calculation formula is: ;

[0086] in, is the weight matrix is the normalized input vector, is the bias vector, is the linear output. The linear output is the result of the initial weighting of the input data, which retains the main information structure of the input data. Apply the ReLU (Rectified Linear Unit) activation function to the linear output for nonlinear transformation to obtain the activation output of the first hidden layer. The formula of ReLU is: ;

[0087] This function sets all negative values ​​to zero and leaves positive values ​​unchanged, thereby introducing nonlinear features and enhancing the expressiveness of the model. The activation output of the first hidden layer is passed to the dynamic attention layer, which is designed to weight the input features and highlight important features. The dynamic attention layer identifies the relative importance of each input feature by calculating the attention weight, as follows: ;

[0088] in, is the query matrix, is the key matrix, and the attention weight is calculated by the dot product between these matrices. The attention weight is multiplied by the activation output of the first hidden layer, and the weighted sum is performed to obtain the output of the attention layer to strengthen the key feature information. The output of the attention layer is input into the second hidden layer, and the output of the second hidden layer is obtained by linear transformation and ReLU activation function. In order to enhance the stability and performance of the model, the output of the second hidden layer is dynamically connected with residual connection. By adding the output to the activation output of the previous layer, the gradient vanishing problem is avoided and the training efficiency of the deep neural network is improved. The formula is: ;

[0089] in, is the transformation through the hidden layer, As input, the residual connection can retain the original information and avoid information loss. The residual output is input to the output layer, and the original output vector is obtained after linear transformation. The output vector contains the prediction information for each fault mode. In order to convert it into a probability distribution, the Softmax function is applied to the original output vector. The formula of the Softmax function is: ;

[0090] in, is the i-th element in the original output vector, is the probability of the i-th fault mode. The Softmax function normalizes the output vector to a probability distribution in the interval [0, 1], with a sum of 1. Based on the fault mode probability distribution output by Softmax, the final fault mode recognition result is generated.

[0091] In a specific embodiment, the execution step inputs the activation output of the first hidden layer into the dynamic attention layer to obtain the attention weight, and performs weighted summation of the activation output of the first hidden layer and the attention weight to obtain the attention layer output, which may specifically include the following steps:

[0092] Input the activation output of the first hidden layer into the dynamic attention layer, transpose the activation output of the first hidden layer to obtain a transposed matrix, and multiply the transposed matrix by the first attention weight matrix to obtain an intermediate result;

[0093] Perform hyperbolic tangent function analysis on the intermediate result to obtain the nonlinear transformation result, and perform dot product operation on the nonlinear transformation result and the attention vector to obtain the original attention score;

[0094] Normalize the original attention scores to obtain the attention weights, and expand the attention weights into a diagonal matrix to obtain the second attention weight matrix;

[0095] Perform matrix multiplication on the activation output of the first hidden layer and the second attention weight matrix to obtain a weighted feature matrix;

[0096] The weighted feature matrix is ​​summed along the feature dimension to obtain a context vector, and the context vector is concatenated with the activation output of the first hidden layer to obtain an enhanced feature vector;

[0097] The enhanced feature vector is transformed into a fully connected layer to obtain the attention layer output.

[0098] Specifically, the activation output of the first hidden layer is input to the dynamic attention layer. The activation output can be expressed as a matrix , where each row represents the feature vector of a sample. In order to apply the attention mechanism, Perform a transpose operation to obtain the transposed matrix The purpose of the transposition operation is to facilitate subsequent matrix operations, especially operations related to the calculation of attention weights. With the first attention weight matrix Multiply to get the intermediate result . This process can be expressed as: ;

[0099] in, is the intermediate result matrix, is the first attention weight matrix, which is responsible for the preliminary calculation of the weight of the attention score. Perform hyperbolic tangent (tanh) function analysis to obtain nonlinear transformation results The formula for the hyperbolic tangent function is: ;

[0100] This function will Each element in is mapped to the range of (-1, 1), which increases the nonlinear expression ability of the model and enables the attention mechanism to capture more complex feature relationships. With attention vector Perform dot product operation to calculate the original attention score , the formula is: ;

[0101] in, is the attention vector, usually a learned weight vector, used to further adjust the attention score. Normalize and get the attention weight . Normalization usually uses the Softmax function, whose formula is: ;

[0102] in, is the i-th score in the original attention score, is the normalized attention weight. The application of the Softmax function makes the sum of the attention weights equal to 1, which is interpreted as a probability distribution indicating the importance of each feature vector. Expand to a diagonal matrix , that is, each attention weight is located on the diagonal position, and the other elements are zero. Diagonal matrix The formula for generating is: Where diag represents the operation of converting a vector into a diagonal matrix. And the second attention weight matrix Perform matrix multiplication to obtain the weighted feature matrix . Each feature vector is weighted and adjusted by the attention weight, so that the influence of important features in subsequent processing is strengthened. The formula is expressed as: ;

[0103] here, is the weighted feature matrix. Perform a sum operation along the feature dimension to obtain the context vector The context vector is a summary of all weighted feature vectors, providing a comprehensive feature representation. The sum is taken over the feature dimension, and the value at each position represents the contribution of the feature to the whole. The formula is: ;

[0104] in, is the number of features, is the i-th weighted feature vector. and the activation output of the first hidden layer Concatenate to get the enhanced feature vector . Integrate the original information and the attention-weighted information together to ensure that the model can make full use of all available information when outputting. The formula is expressed as: ;

[0105] in, Represents the concatenation operation. Perform the full connection layer transformation to obtain the attention layer output. The function of the full connection layer is to map the input feature vector to the output space through linear transformation. The formula is: ;

[0106] in, is the weight matrix of the fully connected layer, is the bias vector, is the final output vector.

[0107] The above describes the display screen light source uniformity testing method in the embodiment of the present application. The following describes the display screen light source uniformity testing system 10 in the embodiment of the present application. Figure 2 In the embodiment of the present application, a display screen light source uniformity testing system 10 includes:

[0108] The acquisition module 11 is used to acquire images of the display screen multiple times and take an average value to obtain an original image, and to perform distortion correction and adaptive cropping processing on the original image to obtain a standardized image;

[0109] The analysis module 12 is used to perform adaptive grayscale conversion and dynamic window scanning on the standardized image to obtain target brightness distribution data, and perform multi-level heat map analysis on the target brightness distribution data to obtain hierarchical brightness distribution results;

[0110] The segmentation module 13 is used to intelligently segment the hierarchical brightness distribution result to obtain an initial partitioning scheme, and optimize the initial partitioning scheme according to the display backlight module layout information to obtain an adaptive test partition;

[0111] The processing module 14 is used to perform multi-parameter joint sampling measurement on the adaptive test partition to obtain high-dimensional feature data, and obtain a target performance indicator set by performing dimensionality reduction processing on the high-dimensional feature data;

[0112] The identification module 15 is used to input the target performance indicator set into the dynamically optimized neural network model to perform fault mode identification and obtain a fault mode identification result.

[0113] Through the collaboration of the above components, multiple image acquisition and average processing, combined with distortion correction and adaptive cropping, the quality and accuracy of the original image are significantly improved. The use of adaptive grayscale conversion and dynamic window scanning technology enables the test method to adapt to display screens with different brightness distribution characteristics. Combined with the intelligent segmentation algorithm and the layout information of the display backlight module, a more reasonable and accurate test partition is achieved, which improves the pertinence and efficiency of the test. Through multi-parameter joint sampling measurement and dimensionality reduction processing of high-dimensional feature data, the various performance indicators of the display screen are comprehensively evaluated. The dynamic optimization neural network model is used for fault mode recognition to improve the accuracy and efficiency of fault diagnosis. By introducing automated and intelligent processing procedures, manual intervention is greatly reduced and test efficiency is improved. Through multi-level thermal map analysis and visualization technology, the interpretability of the test results is improved. Through standardized data processing procedures and intelligent parameter adjustment, the accuracy of the display surface light source uniformity test is improved.

[0114] The present application also provides an electronic device, which includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes the steps of the display screen light source uniformity testing method in the above-mentioned embodiments.

[0115] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the display screen light source uniformity testing method.

[0116] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0117] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable an electronic device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0118] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these 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 application.

Claims

1. A method for testing the uniformity of light source on a display screen, characterized in that: The display screen light source uniformity testing method comprises: Capturing images of the display screen multiple times and taking an average value to obtain an original image, and performing distortion correction and adaptive cropping processing on the original image to obtain a standardized image; specifically comprising: capturing images of the display screen N times to obtain N original images, and performing pixel-level arithmetic averaging on the N original images to obtain an average original image; performing edge detection on the average original image to obtain a display screen edge contour, and calculating a perspective transformation matrix based on the display screen edge contour to obtain distortion correction parameters, and using a four-point correspondence method to solve a homography matrix; performing an affine transformation on the average original image based on the distortion correction parameters to obtain a corrected image, and performing adaptive threshold segmentation on the corrected image to obtain a display area mask; performing regional growth on the corrected image based on the display area mask to obtain an accurate display area boundary, and calculating a minimum circumscribed rectangle based on the accurate display area boundary to obtain a cropping area coordinate; cropping the corrected image based on the cropping area coordinates to obtain a preliminary standardized image, and performing bilinear interpolation resampling on the preliminary standardized image to obtain a standardized image; The standardized image is subjected to adaptive grayscale conversion and dynamic window scanning to obtain target brightness distribution data, and the target brightness distribution data is subjected to multi-level heat map analysis to obtain hierarchical brightness distribution results; specifically comprising: performing histogram equalization processing on the standardized image to obtain an initial grayscale image, and calculating an adaptive gamma value according to the brightness distribution characteristics of the initial grayscale image; performing nonlinear transformation on the initial grayscale image based on the adaptive gamma value to obtain an enhanced grayscale image; setting an initial window size and step size for the enhanced grayscale image to obtain initial scanning parameters, and performing a multi-level heat map analysis on the target brightness distribution data to obtain a hierarchical brightness distribution result; and performing a multi-level heat map analysis on the target brightness distribution data to obtain a hierarchical brightness distribution result; and performing a multi-level heat map analysis on the standardized image to obtain an initial grayscale image, and calculating an adaptive gamma value according to the brightness distribution characteristics of the initial grayscale image; performing a nonlinear transformation on the initial grayscale image based on the adaptive gamma value to obtain an enhanced grayscale image; and performing a multi-level heat map analysis on the target brightness distribution data to obtain a hierarchical brightness distribution result ... The enhanced grayscale image is subjected to sliding window scanning according to the initial scanning parameters, and the average brightness value in each window is calculated to obtain rough brightness distribution data; the rough brightness distribution data is subjected to variance analysis to obtain refined brightness distribution data, and the refined brightness distribution data is organized into a multidimensional array to obtain target brightness distribution data; the target brightness distribution data is subjected to multi-scale Gaussian filtering to obtain smooth brightness data of different scales, and a difference map between adjacent scales is calculated to obtain multi-scale difference data; a multi-layer heat map is generated according to the multi-scale difference data, and the multi-layer heat map is superimposed according to the scale to obtain a hierarchical brightness distribution result; The hierarchical brightness distribution result is intelligently segmented to obtain an initial partitioning scheme, and the initial partitioning scheme is optimized according to the display backlight module layout information to obtain an adaptive test partition; specifically comprising: performing multi-scale superpixel segmentation on the hierarchical brightness distribution result to obtain an initial superpixel set; calculating the average brightness value and brightness variance of each superpixel in the initial superpixel set to obtain a superpixel feature vector set, and performing K-means clustering on the superpixel feature vector set to obtain K brightness categories; merging the initial superpixel set according to the brightness category to obtain a coarse partitioning result, and clustering the coarse partitioning. The result is subjected to regional growing to obtain an optimized partition result; the shape complexity of each partition in the optimized partition result is calculated, and the ratio of the square of the perimeter to the area is used as the complexity index to obtain a shape complexity set; the boundary of the optimized partition result is simplified according to the shape complexity set to obtain a simplified partition result, and the layout information of the display backlight module is read, including the position of LED lamp beads and the parameters of the optical diffusion film; according to the position of the LED lamp beads and the parameters of the optical diffusion film, a theoretical light intensity distribution model is constructed to obtain a theoretical partition template, and the simplified partition result is subjected to morphological operation with the theoretical partition template to obtain an adaptive test partition; The adaptive test partition is subjected to multi-parameter joint sampling measurement to obtain high-dimensional feature data, and the high-dimensional feature data is subjected to dimensionality reduction processing to obtain a target performance index set; specifically, the method comprises: performing stratified random sampling strategy analysis on the adaptive test partition to obtain a sampling point set; performing brightness measurement on the sampling point set to obtain a brightness data set, and calculating a brightness uniformity index; performing chromaticity measurement on the sampling point set to obtain a chromaticity data set, and calculating a chromaticity uniformity index; performing contrast measurement on the sampling point set to obtain a contrast data set, and calculating a contrast uniformity index; performing flash measurement on the sampling point set The method comprises the following steps: measuring the flicker frequency to obtain a flicker data set, and calculating a flicker index; measuring the response time of the sampling point set to obtain a response time data set, and calculating an average response time; combining the brightness uniformity index, the chromaticity uniformity index, the contrast uniformity index, the flicker index and the average response time into a high-dimensional feature vector; performing principal component analysis on the high-dimensional feature vector to obtain a feature vector after dimension reduction, and performing standardization processing on the feature vector after dimension reduction to obtain a standardized feature vector; performing Fisher discriminant analysis on the standardized feature vector to obtain a target performance indicator set; The target performance indicator set is input into a dynamically optimized neural network model to perform fault mode recognition to obtain a fault mode recognition result.

2. The display screen light source uniformity testing method according to claim 1, characterized in that: The step of inputting the target performance indicator set into the dynamically optimized neural network model for fault mode recognition to obtain a fault mode recognition result includes: Normalizing the target performance indicator set to obtain a standardized input vector, and inputting the standardized input vector into an input layer in a dynamically optimized neural network model for processing to obtain an input layer output; Performing weight matrix and bias vector processing of the first hidden layer on the input layer output to obtain a linear output of the first hidden layer, and performing activation function ReLU operation on the linear output of the first hidden layer to obtain an activation output of the first hidden layer; Inputting the activation output of the first hidden layer into the dynamic attention layer to obtain the attention weight, and performing weighted summation of the activation output of the first hidden layer and the attention weight to obtain the attention layer output; Input the output of the attention layer into the second hidden layer, and obtain the output of the second hidden layer after linear transformation and ReLU activation; Performing a dynamic residual connection on the output of the second hidden layer and adding it to the activation output of the first hidden layer to obtain a residual output, and inputting the residual output into the output layer to obtain an original output vector after linear transformation; A Softmax function operation is performed on the original output vector to obtain a fault mode probability distribution, and a fault mode recognition result is generated according to the fault mode probability distribution.

3. The method for testing the uniformity of light source of a display screen according to claim 2, characterized in that: The step of inputting the activation output of the first hidden layer into the dynamic attention layer to obtain an attention weight, and performing a weighted summation of the activation output of the first hidden layer and the attention weight to obtain an attention layer output includes: Inputting the activation output of the first hidden layer into the dynamic attention layer, performing a transposition operation on the activation output of the first hidden layer to obtain a transposed matrix, and multiplying the transposed matrix by the first attention weight matrix to obtain an intermediate result; Performing a hyperbolic tangent function analysis on the intermediate result to obtain a nonlinear transformation result, and performing a dot product operation on the nonlinear transformation result and the attention vector to obtain an original attention score; Normalizing the original attention scores to obtain attention weights, and expanding the attention weights into a diagonal matrix to obtain a second attention weight matrix; Performing a matrix multiplication operation on the activation output of the first hidden layer and the second attention weight matrix to obtain a weighted feature matrix; Performing a sum operation on the weighted feature matrix along the feature dimension to obtain a context vector, and concatenating the context vector with the activation output of the first hidden layer to obtain an enhanced feature vector; The enhanced feature vector is transformed into a fully connected layer to obtain an attention layer output.

4. A display screen light source uniformity testing system, characterized in that: The display screen light source uniformity testing system comprises: The acquisition module is used to acquire images of the display screen for multiple times and take an average value to obtain an original image, and perform distortion correction and adaptive cropping processing on the original image to obtain a standardized image; specifically, the acquisition module comprises: acquiring images of the display screen for N times to obtain N original images, and performing pixel-level arithmetic averaging on the N original images to obtain an average original image; performing edge detection on the average original image to obtain an edge contour of the display screen, and calculating a perspective transformation matrix based on the edge contour of the display screen to obtain distortion correction parameters, and solving a homography matrix using a four-point correspondence method; performing an affine transformation on the average original image based on the distortion correction parameters to obtain a corrected image, and performing adaptive threshold segmentation on the corrected image to obtain a display area mask; performing region growth on the corrected image based on the display area mask to obtain an accurate display area boundary, and calculating a minimum circumscribed rectangle based on the accurate display area boundary to obtain a cropping area coordinate; cropping the corrected image based on the cropping area coordinates to obtain a preliminary standardized image, and performing bilinear interpolation resampling on the preliminary standardized image to obtain a standardized image; The analysis module is used to perform adaptive grayscale conversion and dynamic window scanning on the standardized image to obtain target brightness distribution data, and perform multi-level heat map analysis on the target brightness distribution data to obtain hierarchical brightness distribution results; specifically comprising: performing histogram equalization processing on the standardized image to obtain an initial grayscale image, and calculating an adaptive gamma value based on the brightness distribution characteristics of the initial grayscale image; performing nonlinear transformation on the initial grayscale image based on the adaptive gamma value to obtain an enhanced grayscale image; setting an initial window size and step size for the enhanced grayscale image to obtain initial scanning parameters, and calculating the adaptive gamma value based on the adaptive gamma value; The initial scanning parameters perform sliding window scanning on the enhanced grayscale image, calculate the average brightness value in each window, and obtain rough brightness distribution data; perform variance analysis on the rough brightness distribution data to obtain refined brightness distribution data, and organize the refined brightness distribution data into a multidimensional array to obtain target brightness distribution data; perform multi-scale Gaussian filtering on the target brightness distribution data to obtain smooth brightness data of different scales, and calculate the difference map between adjacent scales to obtain multi-scale difference data; generate a multi-layer heat map according to the multi-scale difference data, and superimpose the multi-layer heat map according to scale to obtain a hierarchical brightness distribution result; The segmentation module is used to intelligently segment the hierarchical brightness distribution result to obtain an initial partitioning scheme, and optimize the initial partitioning scheme according to the display backlight module layout information to obtain an adaptive test partition; specifically comprising: performing multi-scale superpixel segmentation on the hierarchical brightness distribution result to obtain an initial superpixel set; calculating the average brightness value and brightness variance of each superpixel in the initial superpixel set to obtain a superpixel feature vector set, and performing K-means clustering on the superpixel feature vector set to obtain K brightness categories; merging the initial superpixel set according to the brightness category to obtain a coarse partitioning result, and Performing regional growth on the rough partitioning results to obtain optimized partitioning results; calculating the shape complexity of each partition in the optimized partitioning results, using the ratio of the square of the perimeter to the area as the complexity index, and obtaining a shape complexity set; performing boundary simplification on the optimized partitioning results according to the shape complexity set to obtain a simplified partitioning result, and reading the display backlight module layout information, including the LED lamp bead position and the optical diffusion film parameters; constructing a theoretical light intensity distribution model according to the LED lamp bead position and the optical diffusion film parameters to obtain a theoretical partitioning template, and performing morphological operations on the simplified partitioning results and the theoretical partitioning template to obtain an adaptive test partition; The processing module is used to perform multi-parameter joint sampling measurement on the adaptive test partition to obtain high-dimensional feature data, and obtain a target performance indicator set by performing dimensionality reduction processing on the high-dimensional feature data; specifically comprising: performing stratified random sampling strategy analysis on the adaptive test partition to obtain a sampling point set; performing brightness measurement on the sampling point set to obtain a brightness data set, and calculating a brightness uniformity index; performing chromaticity measurement on the sampling point set to obtain a chromaticity data set, and calculating a chromaticity uniformity index; performing contrast measurement on the sampling point set to obtain a contrast data set, and calculating a contrast uniformity index; performing chromaticity measurement on the sampling point set to obtain a contrast data set, and calculating a contrast uniformity index; performing chromaticity measurement on the sampling point set to obtain a contrast data set, and calculating a contrast uniformity index; performing chromaticity measurement on the sampling point set to obtain a contrast data set, and calculating a contrast uniformity index; performing chromaticity measurement on the sampling point set to obtain a contrast data set, and calculating a contrast uniformity index. Perform flicker frequency measurement to obtain a flicker data set, and calculate a flicker index; perform response time measurement on the sampling point set to obtain a response time data set, and calculate an average response time; combine the brightness uniformity index, the chromaticity uniformity index, the contrast uniformity index, the flicker index, and the average response time into a high-dimensional feature vector; perform principal component analysis on the high-dimensional feature vector to obtain a feature vector after dimension reduction, and perform standardization on the feature vector after dimension reduction to obtain a standardized feature vector; perform Fisher discriminant analysis on the standardized feature vector to obtain a target performance indicator set; The identification module is used to input the target performance indicator set into the dynamically optimized neural network model to perform fault mode identification and obtain a fault mode identification result.

5. An electronic device, characterized in that: The electronic device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the electronic device to execute the display screen light source uniformity testing method as described in any one of claims 1-3.

6. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the display screen light source uniformity testing method according to any one of claims 1 to 3 is implemented.