A control method for display module, electronic device and chip
By collecting light intensity distribution data, establishing deviation classification models and deep learning analysis, and combining with multi-objective optimization algorithm to optimize the control strategy of display modules, the problems of inaccurate display module control and low defect detection efficiency in the existing technology are solved, and the problem of inaccurate display module control and low defect detection efficiency is achieved. More efficient and accurate display effects and robustness.
Patent Information
- Application Number
- CN202411106846.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-13
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-08-13
AI Technical Summary
The existing display module control methods are difficult to adapt to complex and changeable usage environments, resulting in poor display effects, and are inefficient in handling display defects and are prone to misjudgment, which cannot meet the high-precision requirements.
By creating an initial control strategy, collecting light intensity distribution data, establishing a classification model between classes of relative deviations, performing image capture and preprocessing, using backpropagation neural networks and deep reinforcement learning for defect analysis, combining multi-objective optimization problem decomposition and near-end Strategy optimization algorithm, optimizes the control strategy of display modules.
It significantly improves the control accuracy and uniformity of the display module, improves the accuracy and robustness of defect detection, ensures the optimization of display effects, and enhances the adaptability and stability of the system.
Smart Images

Figure CN118629337B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of display control technology, and in particular to a control method of a display module, an electronic device and a chip. Background Art
[0002] Currently, the control method of display modules usually relies on traditional control strategies to manage display modules through fixed parameter presets. However, this method has certain limitations and is difficult to adapt to complex and changing usage environments and increasing user needs, resulting in unsatisfactory display effects and affecting user experience.
[0003] In addition, existing display module control methods also have obvious deficiencies in handling display defects. Traditional methods usually use simple image processing technology, which is difficult to accurately identify and classify various display defects. This processing method is not only inefficient, but also prone to misjudgment, and cannot meet the needs of high-precision display devices. Summary of the invention
[0004] The present application provides a control method, electronic device and chip of a display module, which are used to improve the control accuracy of the display module.
[0005] In a first aspect, the present application provides a method for controlling a display module, the method for controlling a display module comprising:
[0006] Creating an initial control strategy for the display module, and performing display testing and light intensity distribution data collection on the display module to obtain light intensity distribution data at different viewing angles;
[0007] Establishing an inter-class relative deviation classification model and performing deviation analysis on the light intensity distribution data to obtain a luminous consistency evaluation index of the display module;
[0008] Capturing and preprocessing the display mode of the display module to obtain a display mode image data set;
[0009] Performing dynamic multi-threshold analysis on the display mode image data set and using a back propagation neural network to perform display defect analysis to obtain a display defect classification index;
[0010] Deep reinforcement learning and multi-objective optimization problem decomposition are performed on the luminous consistency evaluation index and the display defect classification index, and a proximal strategy optimization algorithm is used to optimize and solve the control strategy to obtain the optimal control strategy of the display module.
[0011] In a second aspect, the present application provides an electronic device, the electronic device comprising:
[0012] An acquisition module is used to create an initial control strategy for the display module, and perform display testing and light intensity distribution data acquisition on the display module to obtain light intensity distribution data at different viewing angles;
[0013] An analysis module, used to establish an inter-class relative deviation classification model and perform deviation analysis on the light intensity distribution data to obtain a luminous consistency evaluation index of the display module;
[0014] A capture module, used for performing image capture and image preprocessing on the display mode of the display module to obtain a display mode image data set;
[0015] A processing module, used for performing dynamic multi-threshold analysis on the display mode image data set and using a back propagation neural network to perform display defect analysis to obtain a display defect classification index;
[0016] A solution module is used to perform deep reinforcement learning and multi-objective optimization problem decomposition on the luminous consistency evaluation index and the display defect classification index, and use a proximal strategy optimization algorithm to optimize and solve the control strategy to obtain the optimal control strategy of the display module.
[0017] The third aspect of the present application provides a chip, 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 chip executes the above-mentioned display module control method.
[0018] In the technical solution provided by the present application, by collecting light intensity distribution data and analyzing deviations of the display module, and establishing an inter-class relative deviation classification model, the luminous consistency of the display module at different viewing angles can be effectively evaluated, and the uniformity of the display effect can be significantly improved. The display defect analysis of the display mode image data set using dynamic multi-threshold analysis and back propagation neural network can more accurately identify and classify display defects, and improve the accuracy and reliability of defect detection. Through deep reinforcement learning and proximal strategy optimization algorithms, the control strategy of the display module is optimized and solved, and the optimal control strategy can be found, thereby improving the overall performance of the display module and ensuring the optimization of the display effect. The display mode is captured by a high-speed camera, and image correction, background removal, image enhancement and noise reduction are performed to obtain a high-quality image data set, which helps to more accurately analyze the display mode and detect defects. Combined with the luminous consistency evaluation index and the display defect classification index, the multi-objective optimization problem decomposition and non-dominated sorting method are adopted, which can simultaneously consider multiple optimization goals and improve the comprehensiveness and efficiency of the display module control strategy. Through the use of technical means such as gray-level co-occurrence matrix and feature convolution operations, the light intensity distribution data and display mode image data are intelligently analyzed, key features are extracted and deep learning is performed to improve the intelligence and effect of data analysis. Using the Markov decision process model and deep Q network, the control strategy is adjusted according to real-time data, with adaptive capabilities, and can better cope with display requirements under different usage environments and conditions. Through efficient optimization algorithms and intelligent analysis methods, the system's adaptability to light intensity distribution and display defects is enhanced, and the robustness and stability of the display module control system are improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.
[0020] Figure 1 A schematic diagram of an embodiment of a method for controlling a display module in an embodiment of the present application;
[0021] Figure 2 This is a schematic diagram of an embodiment of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION
[0022] The embodiments of the present application provide a control method, an electronic device and a chip for a display module. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0023] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In one embodiment of the present application, a method for controlling a display module includes:
[0024] Step S101, creating an initial control strategy for the display module, and performing display testing and light intensity distribution data collection on the display module to obtain light intensity distribution data at different viewing angles;
[0025] It is understandable that the execution subject of the present application may be an electronic device, 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.
[0026] Specifically, an initial control strategy for the display module is created, and an initial driving circuit model for the display module is established. By analyzing the initial control strategy, an initial driving circuit model that can maximize the performance of the display module is developed. The initial driving circuit model is discretized according to the initial control strategy to obtain an initial driving control sequence. The specific characteristics and requirements of the display module must be considered during the creation of this sequence to ensure the effectiveness and accuracy of subsequent tests. The initial driving control sequence is loaded into the driving circuit of the display module, and the display module is tested for full viewing angle display, and a standard grayscale calibration image is displayed. The standard grayscale calibration image can provide a basic display performance benchmark, which is helpful for accurate correction and analysis of subsequent data. The display module is placed in the spherical coordinate system test position of the multi-dimensional viewing angle measurement platform, and the light intensity distribution image of the display module is collected for each viewing angle position of the spherical coordinate system. After the acquisition is completed, the light intensity distribution image is denoised and corrected to obtain a purified light intensity distribution image set, and a light intensity distribution data cube is constructed based on the purified light intensity distribution image set. In the process of image denoising and correction, image processing algorithms are applied to remove the noise that may be generated during the acquisition process and to correct the irregularities in the light intensity distribution to ensure the accuracy and consistency of the data. After constructing the light intensity distribution data cube, it is sliced to obtain the light intensity distribution matrix of the horizontal and vertical viewing angles. The light intensity distribution matrix of the horizontal and vertical viewing angles is used as the light intensity distribution data at different viewing angles.
[0027] Step S102, establishing an inter-class relative deviation classification model and performing deviation analysis on the light intensity distribution data to obtain a luminous consistency evaluation index of the display module;
[0028] Specifically, the light intensity distribution data is subjected to tensor decomposition to obtain a core tensor and a coefficient matrix. Tensor decomposition is an effective method for processing high-dimensional data. By decomposing the original data into tensors and matrices in multiple low-dimensional subspaces, the potential structure and features of the data can be effectively extracted. An inter-class relative deviation classification model is constructed based on the core tensor and the coefficient matrix. The model is used to analyze the relative deviations between different categories and evaluate the luminous consistency of the display module. A feature convolution operation is performed on the light intensity distribution data to obtain a feature tensor. Feature convolution is a deep learning technology that extracts important features from the data by performing multi-layer convolution operations on the input data to enhance the analysis capability of the model. According to the inter-class relative deviation classification model, the feature tensor is classified to obtain a classification result, and the light intensity distribution features of different categories are identified. The light intensity distribution data is clustered according to the classification results to obtain clustered data. Clustering is an unsupervised learning method that discovers natural groupings and structures in the data by clustering similar data points together. Principal component analysis is performed on the clustered data to extract the principal components, and the display modules are graded according to the principal components to obtain the luminous consistency level. The luminous consistency evaluation index of the display module is output according to the luminous consistency level.
[0029] Step S103, performing image capture and image preprocessing on the display mode of the display module to obtain a display mode image data set;
[0030] Specifically, a preset pattern sequence is displayed by a display module, and a high-speed camera is used to shoot the preset pattern sequence to obtain an original image sequence. The original image sequence is image corrected to obtain a corrected image sequence. In the image correction process, geometric transformation and color correction techniques are applied to eliminate distortion and chromatic aberration in the camera shooting process, so that the image more truly reflects the actual display effect of the display module. The corrected image sequence is background removed to obtain a foreground image sequence. The background removal step adopts an image segmentation algorithm to remove irrelevant background information and retain the main display content of the display module. The foreground image sequence is image enhanced to obtain an enhanced image sequence. Image enhancement technology improves the visual effect of the image by adjusting parameters such as image contrast, brightness and sharpness, making the display details clearer. Image denoising is performed on the enhanced image sequence to obtain a denoised image sequence. The image denoising step improves the signal-to-noise ratio of the image by removing random noise in the image, thereby ensuring the purity and reliability of the data. The denoised image sequence is image binarized to obtain a binary image sequence. The image is converted into an image with only black and white colors, which simplifies the complexity of the image and makes the features of the image more prominent. The binary image sequence is used as a display mode image dataset, which contains key image information of the display module in different display modes.
[0031] Step S104, performing dynamic multi-threshold analysis on the display mode image data set and using a back propagation neural network to perform display defect analysis to obtain a display defect classification index;
[0032] Specifically, a gray level co-occurrence matrix is constructed for the display mode image data set to obtain a gray level co-occurrence matrix set, and a texture feature vector is calculated based on the gray level co-occurrence matrix set. The gray level co-occurrence matrix can reflect the spatial relationship between the gray values of pixels in the image. The texture feature vector obtained by calculation describes the texture features of the display image. The texture feature vector is input into the adaptive threshold segmentation algorithm, and the display mode image data set is subjected to multi-threshold segmentation to obtain a segmented image set. The adaptive threshold segmentation algorithm automatically adjusts the threshold according to the local characteristics of the image, segments different areas in the image, and effectively extracts the display defect area. A morphological opening operation is performed on the segmented image set to obtain an open operation image set. Morphological opening operation is an image processing technology that removes noise points and small areas in the image and retains the main structural features, making the defect area clearer and more coherent. Defect feature descriptors are extracted based on the open operation image set, and the defect feature descriptors are input into the pre-trained convolutional neural network model. The defect feature descriptors are feature extracted and encoded by the convolutional neural network model to obtain an encoded feature vector. The convolutional neural network model extracts deep features from the defect feature descriptor through multi-layer convolution and pooling operations, encodes them, and generates a coded feature vector with high-dimensional feature representation. The coded feature vector is input into the back propagation neural network classifier, and the back propagation neural network classifier is trained for parameter optimization to obtain an optimized classifier model. The back propagation neural network classifier continuously adjusts the model parameters through the back propagation algorithm to minimize the classification error and improve the classification accuracy and robustness. The coded feature vector is classified by the optimized classifier model to obtain the display defect classification index. The optimized classifier model can accurately identify and classify different defect types in the display image and provide detailed display defect classification indicators.
[0033] Step S105: perform deep reinforcement learning and multi-objective optimization problem decomposition on the luminous consistency evaluation index and the display defect classification index, and use the proximal strategy optimization algorithm to optimize and solve the control strategy to obtain the optimal control strategy of the display module.
[0034] Specifically, a Markov decision process model for display module control strategy optimization is constructed, and the luminous consistency evaluation index and display defect classification index are input into the Markov decision process model as state features. The Markov decision process model can effectively describe the problem of display module control strategy optimization through the definition of state, action and reward, and provide a framework for solving the optimal control strategy. According to the Markov decision process model, a reward function for display module control strategy optimization is established, and the reward function is input into the deep Q network to perform deep reinforcement learning training on the control strategy optimization problem of the display module. The reward function defines the pros and cons criteria of the control strategy. Through the deep Q network, the control strategy is continuously tried and adjusted in a simulated environment to maximize the cumulative reward and gradually approach the optimal strategy. During the deep reinforcement learning training process, the deep Q network will evaluate the value of each state-action pair and update the strategy so that the model can better cope with the actual control needs of the display module. According to the output of the deep Q network, a multi-objective optimization model is modeled for the control strategy optimization problem to obtain a multi-objective optimization model, and the multi-objective optimization model is non-dominated sorted by a multi-objective evolutionary algorithm to obtain a Pareto optimal solution set. The multi-objective optimization model takes into account multiple objectives in the display module control (such as luminous consistency and display defect minimization) at the same time. Through the multi-objective evolutionary algorithm, the solution space can be effectively explored to find a set of solutions that perform well on different objectives. Each solution in the Pareto optimal solution set represents a possible control strategy, which reaches a balance between different objectives and does not have the possibility of being improved at the same time. According to the Pareto optimal solution set, the control strategy optimization problem is targeted and planned to obtain the target planning solution, and the target planning solution is input into the proximal strategy optimization algorithm. The target planning solution represents the solution selected from the Pareto optimal solution set that best meets the actual application requirements. The proximal strategy optimization algorithm is a reinforcement learning algorithm that locally updates the strategy to ensure that each update maintains a certain stability and convergence in the strategy space, thereby improving the efficiency and effect of the solution. The control strategy optimization problem is iteratively solved by the proximal strategy optimization algorithm to obtain the optimal control strategy. After multiple rounds of iterative optimization and verification, the optimal control strategy can show the best control effect in practical applications. The optimal control strategy is applied to the driving circuit of the display module to achieve the optimal control of the display module.
[0035] The target planning solution is parameterized to obtain a first control strategy parameter vector, and the display module is simulated according to the first control strategy parameter vector to obtain first simulation state data. The target planning solution is converted into specific control strategy parameters to facilitate actual testing and evaluation in a simulation environment. The first simulation state data is input into a state value function approximator to obtain a current state value estimate. The state value function approximator is a model obtained through historical data training, which can estimate the value of the current state. The advantage function value is calculated according to the current state value estimate, and the first control strategy parameter vector is updated according to the advantage function value to obtain a second control strategy parameter vector. The calculation of the advantage function value measures the pros and cons of the current strategy by comparing the current state value with the expected state value, thereby guiding the update of the parameter vector and gradually approaching the optimal strategy. The second control strategy parameter vector is input into a policy function approximator to obtain an updated control strategy, and the display module is simulated according to the updated control strategy to obtain second simulation state data. The policy function approximator generates a new control strategy by learning the updated control strategy parameters, so that the simulation results are closer to the optimal state in actual applications. It is determined whether the second simulation state data meets the termination condition. If the termination condition is met, the second control strategy parameter vector is determined as the optimal control strategy parameter vector. The setting of the termination condition is usually based on preset performance indicators, such as the convergence and stability of the simulation results, to ensure that the optimization process can be stopped in time when the optimal state is reached. The optimal control strategy of the display module is constructed according to the optimal control strategy parameter vector. Finally, the optimal control strategy is obtained after multiple rounds of iterative optimization and verification, which can show the best control effect in practical applications.
[0036] In the embodiment of the present application, by collecting light intensity distribution data and analyzing deviations of the display module, an inter-class relative deviation classification model is established, which can effectively evaluate the luminous consistency of the display module at different viewing angles and significantly improve the uniformity of the display effect. Using dynamic multi-threshold analysis and back propagation neural network to analyze display defects of display mode image data sets, it is possible to more accurately identify and classify display defects, thereby improving the accuracy and reliability of defect detection. Through deep reinforcement learning and proximal strategy optimization algorithms, the control strategy of the display module is optimized and solved, and the optimal control strategy can be found, thereby improving the overall performance of the display module and ensuring the optimization of the display effect. The display mode is captured by a high-speed camera, and image correction, background removal, image enhancement and noise reduction are performed to obtain a high-quality image data set, which helps to more accurately perform display mode analysis and defect detection. Combining the luminous consistency evaluation index and the display defect classification index, the multi-objective optimization problem decomposition and non-dominated sorting method are adopted, which can simultaneously consider multiple optimization goals and improve the comprehensiveness and efficiency of the display module control strategy. Through the use of technical means such as gray-level co-occurrence matrix and feature convolution operations, the light intensity distribution data and display mode image data are intelligently analyzed, key features are extracted and deep learning is performed to improve the intelligence and effect of data analysis. Using the Markov decision process model and deep Q network, the control strategy is adjusted according to real-time data, with adaptive capabilities, and can better cope with display requirements under different usage environments and conditions. Through efficient optimization algorithms and intelligent analysis methods, the system's adaptability to light intensity distribution and display defects is enhanced, and the robustness and stability of the display module control system are improved.
[0037] In a specific embodiment, the process of executing step S101 may specifically include the following steps:
[0038] (1) Create the initial control strategy of the display module and establish the initial driving circuit model of the display module;
[0039] (2) Discretize the initial drive circuit model according to the initial control strategy to obtain the initial drive control sequence;
[0040] (3) Loading the initial drive control sequence to the drive circuit of the display module, and performing a full-viewing angle display test on the display module to display a standard grayscale calibration image;
[0041] (4) Placing the display module at the spherical coordinate system test position of the multi-dimensional viewing angle measurement platform, and collecting the light intensity distribution image of the display module at each viewing angle position of the spherical coordinate system;
[0042] (5) De-noising and correcting the light intensity distribution image to obtain a purified light intensity distribution image set, and constructing a light intensity distribution data cube based on the purified light intensity distribution image set;
[0043] (6) Slicing the light intensity distribution data cube to obtain the light intensity distribution matrix of the horizontal and vertical viewing angles;
[0044] (7) The light intensity distribution matrices of the horizontal viewing angle and the vertical viewing angle are used as the light intensity distribution data at different viewing angles.
[0045] Specifically, by studying the specific needs and characteristics of the display module, a set of initial control strategies are formulated. The control strategy includes adjusting the brightness, contrast, color balance and other parameters of the display module to ensure the best performance of the display effect under various environmental conditions. According to the initial control strategy, the initial driving circuit model of the display module is designed and established. The model determines the specific parameters and layout of each component in the driving circuit through circuit design and simulation. The initial control strategy is applied to the model for discretization. The initial driving control sequence is obtained by converting the continuous control signal into a discrete control sequence. After the initial driving control sequence is loaded into the driving circuit of the display module, a full-viewing angle display test is performed to display a standard grayscale calibration image. By using the standard grayscale calibration image, the basic display performance of the display module can be evaluated to ensure that the driving circuit can correctly execute the initial control strategy. The display module is placed in the spherical coordinate system test position of the multi-dimensional viewing angle measurement platform for testing. On this platform, the display module can perform light intensity distribution tests at various viewing angles in the spherical coordinate system to obtain a full range of display performance data. At each viewing angle position, a high-precision camera device is used to capture the light intensity distribution image of the display module. These images reflect the brightness and color distribution of the display module at different viewing angles. The collected light intensity distribution image may contain some noise and distortion, which needs to be denoised and corrected. By applying image processing algorithms, the noise in the image is removed, and the image distortion caused by the camera equipment or environmental factors is corrected to obtain a purified light intensity distribution image set. According to the purified light intensity distribution image set, a light intensity distribution data cube is constructed. The light intensity distribution data at different viewing angles are integrated into a multidimensional data structure to provide comprehensive light intensity distribution information. The light intensity distribution data cube not only contains the light intensity data of the display module at different viewing angles, but also can show the law and characteristics of the light intensity changing with the viewing angle. In order to better analyze the light intensity distribution data, the light intensity distribution data cube is sliced to obtain the light intensity distribution matrix of the horizontal viewing angle and the vertical viewing angle, and the complex multidimensional data is converted into a two-dimensional matrix form. The light intensity distribution matrix of the horizontal viewing angle and the vertical viewing angle records the light intensity distribution of the display module at these viewing angles. For example, suppose that during the test, it is found that at certain viewing angles, the brightness distribution of the display module is uneven, and obvious bright spots and dark spots appear. By analyzing the light intensity distribution matrix at these viewing angles, the specific cause of the problem can be found. It may be that the driving circuit cannot provide enough current at these viewing angles or due to material problems of the display module itself. After determining the cause of the problem, the display performance can be improved by adjusting the control strategy of the driving circuit, increasing the current supply at these viewing angles, or improving the material and structural design of the display module. Finally, the light intensity distribution matrix of the horizontal and vertical viewing angles is used as the light intensity distribution data at different viewing angles, combined with the optimization of the initial control strategy and the driving circuit model, to achieve efficient control and optimization of the display module.
[0046] In a specific embodiment, the process of executing step S102 may specifically include the following steps:
[0047] (1) Perform tensor decomposition on the light intensity distribution data to obtain the kernel tensor and coefficient matrix, and construct an inter-class relative deviation classification model based on the kernel tensor and coefficient matrix;
[0048] (2) Perform feature convolution operation on the light intensity distribution data to obtain a feature tensor;
[0049] (3) Classify the feature tensor according to the inter-class relative deviation classification model to obtain the classification result;
[0050] (4) Clustering the light intensity distribution data according to the classification results to obtain cluster data;
[0051] (5) Performing principal component analysis on the clustered data, extracting the principal components, and classifying the display modules according to the principal components to obtain the luminous consistency grade, and outputting the luminous consistency evaluation index of the display module according to the luminous consistency grade.
[0052] Specifically, the light intensity distribution data can be represented as a high-dimensional tensor , which contains the light intensity distribution information of the display module at different viewing angles. Perform tensor decomposition on the light intensity distribution data to obtain the core tensor and coefficient matrix, decomposing the high-dimensional data into a more manageable low-dimensional form. Tensor decomposition can use methods such as CP decomposition or Tucker decomposition. For CP decomposition, the tensor Represented as multiple core tensors The product of the coefficient matrix, that is
[0053] ;
[0054] in, , and is the coefficient matrix, represents the outer product operation, is the rank of the decomposition. The kernel tensor and the coefficient matrix , B, Together they form the result of tensor decomposition. By decomposing, the original high-dimensional tensor Represented as the interaction between low-dimensional core information and patterns. According to the decomposition of the core tensor and the coefficient matrix , , , a classification model of relative deviation between classes can be constructed. By analyzing the relative deviation between different classes, the luminous consistency of the display module can be evaluated. The calculation formula of relative deviation is:
[0055] ;
[0056] in, It is Class and The relative deviation between classes, and They are the corresponding elements in the core tensor. Through calculation, a relative deviation matrix is obtained for the construction of the classification model. The light intensity distribution data is subjected to feature convolution operation. The decomposed light intensity distribution data is input into the convolutional neural network, and the important features in the data are extracted through multi-layer convolution and pooling operations to obtain the feature tensor F. The feature convolution operation can be expressed as:
[0057] F ;
[0058] in, is the convolution kernel, * indicates the convolution operation, is the bias term, is an activation function. According to the inter-class relative deviation classification model, the feature tensor is classified to obtain the classification result. The classification process is implemented by a pre-trained classification model, which can accurately assign the data in the feature tensor to different categories, thereby identifying the different luminous characteristics of the display module. The light intensity distribution data is clustered according to the classification result to obtain clustered data. Clustering is an unsupervised learning method that discovers natural groupings and structures in the data by clustering similar data points together. Common clustering algorithms include K-means clustering and hierarchical clustering. Through clustering, data points with similar luminous characteristics in the light intensity distribution data are classified into the same category. Principal component analysis is performed on the clustered data to extract the principal components, and the display modules are graded according to the principal components to obtain the luminous consistency level. Principal component analysis extracts the main features in the data by projecting high-dimensional data into low-dimensional space, reduces the dimension of the data, and improves the efficiency and accuracy of the analysis. The luminous consistency evaluation index of the display module is output according to the luminous consistency level. For example, suppose that during the test, through tensor decomposition and feature convolution operations, it is found that there is a significant deviation in the light intensity distribution under certain viewing angles. After classification and cluster analysis, these data are assigned to the same category. Through principal component analysis, the main deviation features in these data are extracted, and the display module is graded according to these features, and it is found that its luminous consistency level is low. This evaluation index suggests that the driving circuit and materials of the display module need to be improved to improve its luminous consistency.
[0059] In a specific embodiment, the process of executing step S103 may specifically include the following steps:
[0060] (1) Displaying a preset pattern sequence through a display module and photographing the preset pattern sequence using a high-speed camera to obtain an original image sequence;
[0061] (2) performing image correction on the original image sequence to obtain a corrected image sequence, and performing background removal on the corrected image sequence to obtain a foreground image sequence;
[0062] (3) performing image enhancement on the foreground image sequence to obtain an enhanced image sequence, and performing image denoising on the enhanced image sequence to obtain a denoised image sequence;
[0063] (4) Binarizing the denoised image sequence to obtain a binarized image sequence, and using the binarized image sequence as a display mode image data set.
[0064] Specifically, the display module displays a preset pattern sequence to comprehensively evaluate the performance of the display module under different conditions. The preset patterns may include standard grayscale patterns, color gradient patterns, and complex images, which can reveal the performance of the display module in terms of brightness, contrast, color reproduction, and response time. The preset pattern sequence is photographed using a high-speed camera to obtain an original image sequence. The original image sequence is image corrected to obtain a corrected image sequence. Image errors caused by factors such as camera position, lens distortion, and illumination changes during the shooting process are eliminated. Common correction methods include geometric correction and color correction. Geometric correction eliminates image distortion through mapping transformation to restore the geometric shape of the image to normal. Color correction eliminates color deviation caused by illumination changes by adjusting the color distribution of the image. After image correction is completed, background removal is performed on the corrected image sequence to obtain a foreground image sequence. Background information irrelevant to the display module is removed from the image, and only the display content of the display module is retained. Common methods include background subtraction and image segmentation. Background subtraction removes background by subtracting a pre-acquired background image from each frame of the image. Image segmentation separates foreground and background pixels by classifying the pixels in the image. Image enhancement is performed on the foreground image sequence to obtain an enhanced image sequence. The purpose of image enhancement is to improve the visual effect of the image and make the display characteristics of the display module more prominent. Common image enhancement methods include histogram equalization and contrast enhancement. Histogram equalization increases the dynamic range and contrast of the image by adjusting the grayscale distribution of the image. Contrast enhancement improves the contrast and detail of the image by stretching the grayscale range of the image. Image denoising is performed on the enhanced image sequence to obtain a denoised image sequence. Random noise in the image is removed to improve the signal-to-noise ratio of the image. Common image denoising methods include mean filtering, median filtering and Gaussian filtering. Mean filtering smoothes the noise in the image by locally averaging the pixels of the image. Median filtering effectively removes salt and pepper noise by locally sorting the pixels of the image and selecting the median as the new pixel value. Gaussian filtering removes high-frequency noise by performing Gaussian smoothing on the image. The denoised image sequence is binarized to obtain a binary image sequence, and the binary image sequence is used as a display mode image data set. The image is converted into a binary image with only black and white colors to simplify the complexity of the image and highlight the display characteristics of the display module. Common image binarization methods include global thresholding and local thresholding. Global thresholding selects a global threshold and sets pixels above the threshold to white and pixels below the threshold to black. Local thresholding selects a threshold according to the local characteristics of the image and performs binarization. Image binarization can be expressed as:
[0065] ;
[0066] in, is the pixel value of the image after binarization, is the binarization threshold.
[0067] In a specific embodiment, the process of executing step S104 may specifically include the following steps:
[0068] (1) Constructing the gray-level co-occurrence matrix of the display mode image data set to obtain a gray-level co-occurrence matrix set, and calculating the texture feature vector based on the gray-level co-occurrence matrix set;
[0069] (2) Inputting the texture feature vector into the adaptive threshold segmentation algorithm, performing multi-threshold segmentation on the display mode image data set, and obtaining a segmented image set;
[0070] (3) Perform morphological opening operation on the segmented image set to obtain an opening operation image set;
[0071] (4) Extracting defect feature descriptors based on the open operation image set and inputting the defect feature descriptors into the pre-trained convolutional neural network model;
[0072] (5) Extract and encode the defect feature descriptor through the convolutional neural network model to obtain the encoded feature vector;
[0073] (6) Inputting the encoded feature vector into the back propagation neural network classifier, and performing parameter optimization training on the back propagation neural network classifier to obtain an optimized classifier model;
[0074] (7) The display defect classification is performed on the encoded feature vector by optimizing the classifier model to obtain the display defect classification index.
[0075] Specifically, the gray level co-occurrence matrix is constructed for the display mode image data set. The gray level co-occurrence matrix is an important tool for describing the spatial relationship between the gray values of an image. The gray value in the image is recorded The pixel and gray value are The frequency of simultaneous occurrence of pixels in a specific direction and distance. The formula for constructing the gray level co-occurrence matrix is:
[0076] ;
[0077] in, Indicates the distance and direction The gray value is and The number of times the pixel pair appears, is the image at position The gray value at and Combine and construct a set of gray-level co-occurrence matrices. Calculate the texture feature vector based on the gray-level co-occurrence matrix set. The texture feature vector can extract the texture information of the image, such as contrast, energy, entropy, and homogeneity. These features can be calculated by the statistics of the gray-level co-occurrence matrix. For example, contrast can be expressed as:
[0078] ;
[0079] in, is the number of gray levels. By calculating these statistics for each gray-level co-occurrence matrix, the texture feature vector is obtained The texture feature vector is input into the adaptive threshold segmentation algorithm, and the display mode image data set is segmented by multiple thresholds to obtain a segmented image set. The adaptive threshold segmentation algorithm automatically adjusts the threshold according to the local characteristics of the image, thereby accurately segmenting different areas in the image. The process can be expressed as:
[0080] ;
[0081] in, is the segmented image, It is an adaptive threshold calculated based on local characteristics. Perform morphological opening operation on the segmented image set to obtain the open operation image set. Morphological opening operation is an image processing technology that can remove noise and small objects in the image by performing erosion operation and then dilation operation on the image, while retaining the main structure of the image. The opening operation can be expressed as:
[0082] ;
[0083] in, is the image after opening operation, is a structural element, represents the corrosion operation, Denotes the expansion operation. Extract the defect feature descriptor based on the open operation image set, and input the defect feature descriptor into the pre-trained convolutional neural network model. The defect feature descriptor can describe the defect features in the displayed image, such as shape, size, and position. The pre-trained convolutional neural network model extracts the deep features in the defect feature descriptor through multi-layer convolution and pooling operations, encodes it, and generates an encoded feature vector The feature extraction process of convolutional neural network can be expressed as:
[0084] ;
[0085] in, is the convolution kernel, * indicates the convolution operation, is the bias term, is the activation function. The encoded feature vector is input into the back propagation neural network classifier, and the back propagation neural network classifier is trained to optimize the parameters to obtain the optimized classifier model. The back propagation neural network classifier continuously adjusts the model parameters through the back propagation algorithm to minimize the classification error and improve the classification accuracy and robustness. The training process of the classifier can be expressed as:
[0086] ;
[0087] in, are the parameters of the classifier, is the learning rate, is the loss function, is the true label. Through training, an optimized classifier model is obtained. The display defect classification of the encoded feature vector is performed by the optimized classifier model to obtain the display defect classification index. The optimized classifier model can accurately identify and classify different defect types in the display image and provide detailed display defect classification indicators. For example, by processing the display mode image data set, it can be found that some display modules have bright spot and dark spot defects. The texture feature vectors of these defects are extracted by constructing the grayscale co-occurrence matrix and calculating the texture features. The defect area is segmented by using adaptive threshold segmentation and morphological opening operation, and the defect features are deeply extracted and encoded by the pre-trained convolutional neural network model. The defects are classified by the back-propagation neural network classifier to determine their type and severity, and the display defect classification index is output.
[0088] In a specific embodiment, the process of executing step S105 may specifically include the following steps:
[0089] (1) Construct a Markov decision process model for optimizing the display module control strategy, and input the luminous consistency evaluation index and display defect classification index into the Markov decision process model as state features;
[0090] (2) Based on the Markov decision process model, a reward function for optimizing the control strategy of the display module is established, and the reward function is input into the deep Q network to perform deep reinforcement learning training on the control strategy optimization problem of the display module;
[0091] (3) Based on the output of the deep Q network, a multi-objective optimization model is constructed for the control strategy optimization problem to obtain a multi-objective optimization model. The multi-objective optimization model is then non-dominatedly sorted using a multi-objective evolutionary algorithm to obtain a Pareto optimal solution set.
[0092] (4) Perform target planning on the control strategy optimization problem according to the Pareto optimal solution set, obtain the target planning solution, and input the target planning solution into the proximal strategy optimization algorithm;
[0093] (5) Iteratively solve the control strategy optimization problem through the proximal strategy optimization algorithm to obtain the optimal control strategy;
[0094] (6) Apply the optimal control strategy to the driving circuit of the display module to achieve optimal control of the display module.
[0095] Specifically, a Markov decision process model for optimizing the control strategy of the display module is constructed. In the Markov decision process, the state Indicates the current display module's luminous consistency and display defect status, action Indicates the control strategy adopted for the display module, the transfer probability Indicates that from the current state After action Then transfer to the next state The probability of reward function Indicates in status Take action The instant reward obtained. The luminous consistency evaluation index and the display defect classification index are input into the Markov decision process model as state features. The specific state is expressed as ,in It is an evaluation index of luminescence consistency. It is a display defect classification indicator. Through these state characteristics, the current performance of the display module can be reflected, providing data support for the optimization of the control strategy. The reward function for the optimization of the display module control strategy is established according to the Markov decision process. , and input the reward function into the deep Q network for deep reinforcement learning training. The definition of the reward function needs to comprehensively consider the luminous consistency and display defects to encourage the system to achieve the best display effect. The specific reward function can be expressed as:
[0096] ;
[0097] in, and is the weight parameter, It represents the luminous consistency evaluation index. Indicates the defect classification indicator. and The value of can balance the relationship between luminous consistency and display defects. In deep reinforcement learning training, the deep Q network uses the Q value function The approximate optimal strategy is are the parameters of the neural network. The Q-value function is updated through the Bellman equation:
[0098] ;
[0099] in, is the discount factor, Indicates that in the next state Select the best action The obtained Q value. By repeatedly iteratively updating the Q value function, the optimal control strategy can be approached. According to the output of the deep Q network, the control strategy optimization problem is modeled with multi-objective optimization to obtain a multi-objective optimization model. The multi-objective optimization model needs to consider the optimization objectives of luminous consistency and display defects at the same time, which can be expressed as:
[0100] ;
[0101] The multi-objective optimization model is non-dominatedly sorted by a multi-objective evolutionary algorithm to obtain the Pareto optimal solution set. Each solution in the Pareto optimal solution set represents a control strategy that achieves a balance between different objectives, and it is impossible to improve another objective without sacrificing one objective. According to the Pareto optimal solution set, the control strategy optimization problem is targeted and the target planning solution is obtained. The target planning solution is input into the proximal strategy optimization algorithm. By limiting the step size of the strategy update, the stability and efficiency of the strategy optimization are improved, thereby maximizing the expected reward of the strategy. The control strategy optimization problem is iteratively solved by the proximal strategy optimization algorithm, and the optimal control strategy is finally obtained. The optimal control strategy can show the best effect in practical applications through multiple rounds of optimization and verification. The optimal control strategy is applied to the driving circuit of the display module to achieve the optimal control of the display module.
[0102] In a specific embodiment, the execution step iteratively solves the control strategy optimization problem through the proximal strategy optimization algorithm, and the process of obtaining the optimal control strategy may specifically include the following steps:
[0103] (1) Parameterizing the target planning solution to obtain a first control strategy parameter vector, and simulating the display module according to the first control strategy parameter vector to obtain first simulation state data;
[0104] (2) inputting the first simulation state data into a state value function approximator to obtain a current state value estimate;
[0105] (3) estimating and calculating the advantage function value according to the current state value, and updating the first control strategy parameter vector according to the advantage function value to obtain the second control strategy parameter vector;
[0106] (4) inputting the second control strategy parameter vector into the strategy function approximator to obtain an updated control strategy, and simulating the display module according to the updated control strategy to obtain second simulation state data;
[0107] (5) determining whether the second simulation state data satisfies a termination condition. If the termination condition is satisfied, determining the second control strategy parameter vector as the optimal control strategy parameter vector;
[0108] (6) Construct the optimal control strategy of the display module based on the optimal control strategy parameter vector.
[0109] Specifically, the target planning solution is converted into control strategy parameters. Assume that the target planning solution is expressed as , through the parameterization process, the first control strategy parameter vector is obtained The purpose of the parameterization process is to transform the high-level planning solution into specific control instructions for execution in the simulation. The display module is simulated to obtain the first simulation state data. The simulation process simulates the operation state of the display module under the control strategy and generates a series of state data. , these data include the luminous consistency evaluation and display defect situation of the display module at different time points. Input to the state value function approximator to get the current state value estimate. The state value function approximator is a model used to estimate the current state value, which can be trained with historical data. The estimation formula of the state value function is:
[0110] ;
[0111] in, is the state value estimate, is the weight vector, is the current state data. The advantage function value is estimated and calculated based on the current state value. Advantage function value It reflects the advantage of the current action relative to the average level and can be calculated by the following formula:
[0112] ;
[0113] in, is the state-action value function, which means taking action in state s The expected return, is the state value estimation. According to the advantage function value, the first control strategy parameter vector Update and obtain the second control strategy parameter vector The parameter update process can be achieved by gradient ascent method to maximize the advantage function value:
[0114] ;
[0115] in, is the learning rate, is the advantage function value pair parameter The second control strategy parameter vector Input to the policy function approximator to obtain the updated control strategy. The policy function approximator is a model used to generate a specific control strategy. Input, generate specific control instructions. The updated control strategy is expressed as According to the update control strategy Simulate the display module to obtain the second simulation state data Through simulation, verify the effect of the updated control strategy in actual application. Determine the second simulation state data Whether the termination condition is met. If the termination condition is met, the second control strategy parameter vector The optimal control strategy parameter vector is determined. The termination condition can be that the simulation state data reaches the expected performance index, such as the luminous consistency and the display defect situation reach the set standard. Construct the optimal control strategy for the display module. The optimal control strategy is optimized and verified through multiple rounds of iterations to ensure its best effect in practical applications.
[0116] The above describes the control method of the display module in the embodiment of the present application. The following describes the electronic device in the embodiment of the present application. Figure 2 In the embodiments of the present application, an electronic device includes:
[0117] The acquisition module 201 is used to create an initial control strategy for the display module, and perform display testing and light intensity distribution data acquisition on the display module to obtain light intensity distribution data at different viewing angles;
[0118] An analysis module 202 is used to establish an inter-class relative deviation classification model and perform deviation analysis on the light intensity distribution data to obtain a luminous consistency evaluation index of the display module;
[0119] The capture module 203 is used to capture and preprocess the display mode of the display module to obtain a display mode image data set;
[0120] Processing module 204, used to perform dynamic multi-threshold analysis on the display mode image data set and use a back propagation neural network to perform display defect analysis to obtain a display defect classification index;
[0121] The solution module 205 is used to perform deep reinforcement learning and multi-objective optimization problem decomposition on the luminous consistency evaluation index and the display defect classification index, and use the proximal strategy optimization algorithm to optimize the control strategy to obtain the optimal control strategy of the display module.
[0122] Through the collaboration of the above components, by collecting light intensity distribution data and analyzing deviations of the display module, and establishing an inter-class relative deviation classification model, the luminous consistency of the display module at different viewing angles can be effectively evaluated, and the uniformity of the display effect can be significantly improved. The display defect analysis of the display mode image data set using dynamic multi-threshold analysis and back propagation neural network can more accurately identify and classify display defects, improving the accuracy and reliability of defect detection. Through deep reinforcement learning and proximal strategy optimization algorithms, the control strategy of the display module is optimized and solved, and the optimal control strategy can be found, thereby improving the overall performance of the display module and ensuring the optimization of the display effect. The display mode is captured by a high-speed camera, and image correction, background removal, image enhancement and noise reduction are performed to obtain a high-quality image data set, which is helpful for more accurate display mode analysis and defect detection. Combining the luminous consistency evaluation index and the display defect classification index, the multi-objective optimization problem decomposition and non-dominated sorting method are adopted, which can simultaneously consider multiple optimization objectives and improve the comprehensiveness and efficiency of the display module control strategy. Through the use of technical means such as gray-level co-occurrence matrix and feature convolution operations, the light intensity distribution data and display mode image data are intelligently analyzed, key features are extracted and deep learning is performed to improve the intelligence and effect of data analysis. Using the Markov decision process model and deep Q network, the control strategy is adjusted according to real-time data, with adaptive capabilities, and can better cope with display requirements under different usage environments and conditions. Through efficient optimization algorithms and intelligent analysis methods, the system's adaptability to light intensity distribution and display defects is enhanced, and the robustness and stability of the display module control system are improved.
[0123] The present application also provides a chip, which includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the control method of the display module in the above-mentioned embodiments.
[0124] 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.
[0125] 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 a computer device (which can be a personal computer, server, or network device, etc.) to execute 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 and other media that can store program codes.
[0126] 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 control method for a display module, It is characterized in that The control method of the display module includes: Create an initial control strategy for the display module, and perform display testing and light intensity distribution data collection on the display module to obtain light intensity distribution data at different viewing angles; specifically include: create an initial control strategy for the display module, and establish an initial driving circuit model for the display module; discretize the initial driving circuit model according to the initial control strategy to obtain an initial driving control sequence; load the initial driving control sequence to the driving circuit of the display module, and perform a full-viewing angle display test on the display module to display a standard grayscale calibration image; place the display module at a spherical coordinate system test position of a multi-dimensional viewing angle measurement platform, and collect a light intensity distribution image of the display module at each viewing angle position of the spherical coordinate system; perform denoising and correction processing on the light intensity distribution image to obtain a purified light intensity distribution image set, and construct a light intensity distribution data cube based on the purified light intensity distribution image set; slice the light intensity distribution data cube to obtain light intensity distribution matrices at horizontal and vertical viewing angles; use the light intensity distribution matrices at horizontal and vertical viewing angles as light intensity distribution data at different viewing angles; Establishing an inter-class relative deviation classification model and performing deviation analysis on the light intensity distribution data to obtain a luminous consistency evaluation index of the display module; specifically comprising: performing tensor decomposition on the light intensity distribution data to obtain a core tensor and a coefficient matrix, and constructing an inter-class relative deviation classification model based on the core tensor and the coefficient matrix; performing a feature convolution operation on the light intensity distribution data to obtain a feature tensor; classifying the feature tensor according to the inter-class relative deviation classification model to obtain a classification result; clustering the light intensity distribution data according to the classification result to obtain cluster data; performing principal component analysis on the cluster data to extract the principal component, and grading the display module according to the principal component to obtain a luminous consistency grade, and outputting a luminous consistency evaluation index of the display module according to the luminous consistency grade; Capturing and preprocessing the display mode of the display module to obtain a display mode image data set; Performing dynamic multi-threshold analysis on the display mode image data set and using a back propagation neural network to perform display defect analysis to obtain a display defect classification index; The luminous consistency evaluation index and the display defect classification index are subjected to deep reinforcement learning and multi-objective optimization problem decomposition, and the proximal strategy optimization algorithm is used to optimize and solve the control strategy to obtain the optimal control strategy of the display module; specifically, the method comprises: constructing a Markov decision process model for display module control strategy optimization, and inputting the luminous consistency evaluation index and the display defect classification index as state features into the Markov decision process model; establishing a reward function for display module control strategy optimization according to the Markov decision process model, and inputting the reward function into a deep Q network, and performing deep reinforcement learning training on the control strategy optimization problem of the display module; performing multi-objective optimization modeling on the control strategy optimization problem according to the output of the deep Q network to obtain a multi-objective optimization model, and performing non-dominated sorting on the multi-objective optimization model through a multi-objective evolutionary algorithm to obtain a Pareto optimal solution set; performing target planning on the control strategy optimization problem according to the Pareto optimal solution set to obtain a target planning solution, and inputting the target planning solution into a deep Q network. to the proximal strategy optimization algorithm; parameterize the target planning solution to obtain a first control strategy parameter vector, and simulate the display module according to the first control strategy parameter vector to obtain first simulation state data; input the first simulation state data to a state value function approximator to obtain a current state value estimate; calculate an advantage function value according to the current state value estimate, and update the first control strategy parameter vector according to the advantage function value to obtain a second control strategy parameter vector; input the second control strategy parameter vector to a strategy function approximator to obtain an updated control strategy, and simulate the display module according to the updated control strategy to obtain second simulation state data; determine whether the second simulation state data satisfies a termination condition, and if so, determine the second control strategy parameter vector as an optimal control strategy parameter vector; construct an optimal control strategy for the display module according to the optimal control strategy parameter vector; apply the optimal control strategy to a driving circuit of the display module to achieve optimal control of the display module.
2. The control method of the display module according to claim 1, It is characterized in that The step of performing image capture and image preprocessing on the display mode of the display module to obtain a display mode image data set includes: Displaying a preset pattern sequence through the display module, and photographing the preset pattern sequence with a high-speed camera to obtain an original image sequence; Performing image correction on the original image sequence to obtain a corrected image sequence, and performing background removal on the corrected image sequence to obtain a foreground image sequence; Performing image enhancement on the foreground image sequence to obtain an enhanced image sequence, and performing image denoising on the enhanced image sequence to obtain a denoised image sequence; The denoised image sequence is subjected to image binarization to obtain a binarized image sequence, and the binarized image sequence is used as a display mode image data set.
3. The control method of the display module according to claim 1, It is characterized in that The dynamic multi-threshold analysis is performed on the display mode image data set and the display defect analysis is performed using a back propagation neural network to obtain a display defect classification index, including: Constructing a gray-level co-occurrence matrix for the display mode image data set to obtain a gray-level co-occurrence matrix set, and calculating a texture feature vector according to the gray-level co-occurrence matrix set; Inputting the texture feature vector into an adaptive threshold segmentation algorithm, performing multi-threshold segmentation on the display mode image data set to obtain a segmented image set; Performing a morphological opening operation on the segmented image set to obtain an opening operation image set; Extracting defect feature descriptors according to the open operation image set, and inputting the defect feature descriptors into a pre-trained convolutional neural network model; Extracting and encoding the defect feature descriptor through the convolutional neural network model to obtain an encoded feature vector; Inputting the encoded feature vector into a back propagation neural network classifier, and performing parameter optimization training on the back propagation neural network classifier to obtain an optimized classifier model; The display defect classification is performed on the encoded feature vector using the optimized classifier model to obtain a display defect classification index.
4. An electronic device, used to execute the control method of the display module according to claim 1, It is characterized in that The electronic device comprises: The acquisition module is used to create an initial control strategy for the display module, and perform display testing and light intensity distribution data acquisition on the display module to obtain light intensity distribution data at different viewing angles; specifically, the module includes: creating an initial control strategy for the display module, and establishing an initial driving circuit model for the display module; discretizing the initial driving circuit model according to the initial control strategy to obtain an initial driving control sequence; loading the initial driving control sequence into the driving circuit of the display module, and performing a full-viewing angle display test on the display module to display a standard grayscale calibration image; placing the display module at a spherical coordinate system test position of a multi-dimensional viewing angle measurement platform, and acquiring a light intensity distribution image of the display module at each viewing angle position of the spherical coordinate system; performing denoising and correction processing on the light intensity distribution image to obtain a purified light intensity distribution image set, and constructing a light intensity distribution data cube based on the purified light intensity distribution image set; slicing the light intensity distribution data cube to obtain light intensity distribution matrices at horizontal and vertical viewing angles; using the light intensity distribution matrices at horizontal and vertical viewing angles as light intensity distribution data at different viewing angles; An analysis module is used to establish an inter-class relative deviation classification model and perform deviation analysis on the light intensity distribution data to obtain a luminous consistency evaluation index of the display module; specifically comprising: performing tensor decomposition on the light intensity distribution data to obtain a core tensor and a coefficient matrix, and constructing an inter-class relative deviation classification model based on the core tensor and the coefficient matrix; performing a feature convolution operation on the light intensity distribution data to obtain a feature tensor; classifying the feature tensor according to the inter-class relative deviation classification model to obtain a classification result; clustering the light intensity distribution data according to the classification result to obtain cluster data; performing principal component analysis on the cluster data to extract the principal component, and grading the display module according to the principal component to obtain a luminous consistency grade, and outputting the luminous consistency evaluation index of the display module according to the luminous consistency grade; A capture module, used for performing image capture and image preprocessing on the display mode of the display module to obtain a display mode image data set; A processing module, configured to perform dynamic multi-threshold analysis on the display mode image data set and use a back propagation neural network to perform display defect analysis to obtain a display defect classification index; A solution module is used to perform deep reinforcement learning and multi-objective optimization problem decomposition on the luminous consistency evaluation index and the display defect classification index, and adopt a proximal strategy optimization algorithm to perform control strategy optimization solution to obtain the optimal control strategy of the display module; specifically includes: constructing a Markov decision process model for display module control strategy optimization, and inputting the luminous consistency evaluation index and the display defect classification index as state features into the Markov decision process model; establishing a reward function for display module control strategy optimization according to the Markov decision process model, and inputting the reward function into a deep Q network, and performing deep reinforcement learning training on the control strategy optimization problem of the display module; based on the output of the deep Q network, performing multi-objective optimization modeling on the control strategy optimization problem to obtain a multi-objective optimization model, and performing non-dominated sorting on the multi-objective optimization model through a multi-objective evolutionary algorithm to obtain a Pareto optimal solution set; performing target planning on the control strategy optimization problem according to the Pareto optimal solution set to obtain a target planning solution, and inputting the target planning into the deep Q network. The solution is input into the proximal strategy optimization algorithm; the target planning solution is parameterized to obtain a first control strategy parameter vector, and the display module is simulated according to the first control strategy parameter vector to obtain first simulation state data; the first simulation state data is input into the state value function approximator to obtain a current state value estimate; the advantage function value is calculated according to the current state value estimate, and the first control strategy parameter vector is updated according to the advantage function value to obtain a second control strategy parameter vector; the second control strategy parameter vector is input into the strategy function approximator to obtain an updated control strategy, and the display module is simulated according to the updated control strategy to obtain second simulation state data; it is determined whether the second simulation state data satisfies the termination condition, and if the termination condition is satisfied, the second control strategy parameter vector is determined as the optimal control strategy parameter vector; the optimal control strategy of the display module is constructed according to the optimal control strategy parameter vector; the optimal control strategy is applied to the driving circuit of the display module to achieve optimal control of the display module.
5. A chip, It is characterized in that The chip includes: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instruction in the memory so that the chip executes the control method of the display module according to any one of claims 1 to 3.
