Production detection method for dynamic refresh rate of display screen

Abstract

The invention provides a production detection method for the dynamic refresh rate of a display screen, and the method comprises the steps: employing a heuristic optimization algorithm, such as a genetic algorithm, for different regions of the display screen, determining an optimal arrangement scheme of an optical sensor array through iterative search under the condition that the overall layout uniformity constraint condition is satisfied, and achieving the optimal arrangement scheme of the optical sensor array while controlling the number of sensors. The key area of the display screen is covered to the maximum degree, and global optimization of sensor arrangement is achieved; a data acquisition method based on compressed sensing is adopted, sparse representation and compression are performed on high-dimensional optical signal data at a sensor array end, the data transmission bandwidth requirement is greatly reduced, high-precision signal recovery and restoration are realized at a background processing system end through an L1 norm minimization reconstruction algorithm, and the real-time performance and reliability of data transmission are ensured.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a production detection method for the dynamic refresh rate of a display screen. Background Art

[0002] In the production process of display screens, in order to fully evaluate the dynamic refresh rate performance of the display screen, it is necessary to simulate various actual usage scenarios. However, different dynamic image contents vary greatly in parameters such as frame rate, color depth and resolution. How to build a comprehensive test scenario that is close to the actual usage environment within a limited test time and with limited image generators is a technical problem that needs to be solved urgently.

[0003] At the same time, a high-precision, high-frequency response light sensor array is the key to accurately capture the changes in the light signal of the display. However, how to reasonably arrange the sensor array, fully consider the different areas of the display, take into account the uniformity of the overall layout, and control the number of sensors to reduce costs. This is a complex optimization problem.

[0004] In addition, during real-time detection, the drift of the optical sensor array and the errors of the data analysis algorithm will accumulate over time, affecting the accuracy of detection. Therefore, how to dynamically adjust the parameters of the sensor array and analysis algorithm to achieve real-time calibration without interrupting detection using a standard refresh rate signal source is also a technical challenge.

[0005] Finally, how to transmit massive amounts of test data in real time and reliably to the background processing system under limited transmission bandwidth, and quickly compare and analyze it in the background to generate intuitive test reports for real-time monitoring and decision-making by production site staff also requires an efficient and stable data transmission and processing solution. Summary of the invention

[0006] The present invention provides a production detection method for the dynamic refresh rate of a display screen, which mainly includes:

[0007] Step S101, according to a preset dynamic refresh rate test scene library, multiple test image sequences covering different frame rates, color depths and resolutions are obtained, the test images are generated in real time by an image generator, and the output parameters of the image generator are controlled to dynamically adjust the frame rate, color depth and resolution of the generated images to simulate various dynamic image contents in a real usage environment;

[0008] Step S102, for different areas of the display screen, a heuristic optimization algorithm, such as a genetic algorithm, is used to determine the optimal layout of the light sensor array through iterative search under the condition of satisfying the overall layout uniformity constraint, while controlling the number of sensors and maximizing the coverage of the key areas of the display screen, thereby achieving global optimization of the sensor layout;

[0009] Step S103, during the dynamic refresh rate test, periodically switch to the preset standard refresh rate signal source, obtain the real-time output data of the light sensor array, fit the sensor output curve by the least squares method, determine the drift degree of the curve, and if the drift exceeds the preset threshold, dynamically adjust the calibration parameters of the sensor array, and update the drift compensation coefficient in the data analysis algorithm to achieve real-time drift calibration;

[0010] Step S104, using a data acquisition method based on compressed sensing to sparsely represent and compress high-dimensional optical signal data at the sensor array end, greatly reducing the data transmission bandwidth requirement, and at the background processing system end, using an L1 norm minimization reconstruction algorithm to achieve high-precision signal recovery and restoration, ensuring the real-time and reliability of data transmission;

[0011] Step S105: In the background processing system, a parallel computing framework based on GPU acceleration is used to process and analyze the received light sensor array data in real time, and a multi-scale similarity matching is performed with a preset standard refresh rate template to quickly determine whether the dynamic refresh rate performance of the display screen meets the standard, and generate an intuitive test report and visualization results;

[0012] Step S106, based on the dynamic refresh rate detection result of the display screen, an incremental learning method is used to continuously optimize and update the parameter control strategy of the image generator. Through a reinforcement learning algorithm, such as Qlearning, the image generator can adaptively adjust the parameter combination of the generated image according to the detection feedback, thereby continuously improving the comprehensiveness and authenticity of the test scene coverage;

[0013] Step S107, establish an evaluation index system for the dynamic refresh rate performance of the display screen, comprehensively consider multiple indicators such as frame rate stability, color reproduction, motion blur, etc. under different test scenarios, and form a quantitative performance score through a weighted fusion method. Perform a correlation analysis with the subjective evaluation results of the human eye, and continuously optimize and improve the evaluation index system to make it closer to the actual perception of users.

[0014] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0015] The present invention discloses a method for testing the dynamic refresh rate performance of a display screen. The method generates a test image sequence covering different frame rates, color depths and resolutions in real time by presetting a dynamic refresh rate test scene library and an image generator. A heuristic optimization algorithm is used to determine the optimal arrangement of a light sensor array to maximize coverage of key areas of the display screen. During the test, the standard refresh rate signal source is periodically switched to realize real-time drift calibration of the sensor array. The data transmission bandwidth is reduced by compressed sensing technology, and real-time processing and analysis are performed using a parallel computing framework accelerated by a GPU. According to the test results, an incremental learning method is used to optimize the parameter control strategy of the image generator. Finally, a comprehensive evaluation index system is established to form a quantitative performance score. The present invention realizes comprehensive, accurate and efficient testing of the dynamic refresh rate performance of a display screen, providing strong support for improving display quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The present invention is a flow chart of a production detection method for the dynamic refresh rate of a display screen. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] like Figure 1 In this embodiment, a production detection method for the dynamic refresh rate of a display screen may specifically include:

[0019] Step S101, according to the preset dynamic refresh rate test scene library, obtain multiple test image sequences covering different frame rates, color depths and resolutions, generate test images in real time through an image generator, and control the output parameters of the image generator to dynamically adjust the frame rate, color depth and resolution of the generated image to simulate various dynamic image contents in a real usage environment.

[0020] According to the preset dynamic refresh rate test scene library, multiple test image sequence data covering different frame rates, color depths and resolutions are obtained; the obtained test image sequence data is input into the image generator, and the test image is generated in real time by the image generator; the output parameters of the image generator are obtained, and the frame rate, color depth and resolution of the image generated by the image generator are dynamically adjusted according to the output parameters; it is determined whether the adjusted frame rate, color depth and resolution meet the requirements in the preset dynamic refresh rate test scene library, if so, the adjusted parameters are applied to the image generator, if not, the parameters are continued to be adjusted; according to the adjusted parameters, the image generator is controlled to continuously generate test images to obtain a dynamically adjusted test image sequence; the dynamically adjusted test image sequence is input into the display device, and the test image sequence is presented on the display device to simulate various dynamic image contents in a real usage environment; a machine learning algorithm is used to evaluate the quality of the image presented by the display device, and the parameters of the image generator are further optimized according to the evaluation results to improve the quality of the test image sequence and make it closer to the dynamic image content in a real usage environment.

[0021] Step S102, for different areas of the display screen, a heuristic optimization algorithm, such as a genetic algorithm, is used to determine the optimal layout of the light sensor array through iterative search under the constraint of overall layout uniformity, while controlling the number of sensors and maximizing the coverage of key areas of the display screen to achieve global optimization of sensor layout.

[0022] According to the different areas of the display screen, the position coordinates and area size of each area are obtained as the input of the genetic algorithm; the genetic algorithm is used to encode the sensor layout plan into chromosomes, and new layout plans are generated through operations such as crossover and mutation; for each layout plan, the area ratio of the key area of ​​the display screen covered and the overall uniformity index are calculated as the fitness function; through iterative search, the sensor layout plan is continuously optimized until the constraints are met or the maximum number of iterations is reached; the layout plan with the highest fitness is selected from the final population and decoded into the specific position coordinates of the sensor; according to the sensor position coordinates, it is calibrated on the display screen to ensure that the sensor can accurately collect the light information of the area where it is located; the light information collected by the sensor is transmitted to the controller, and the brightness and color temperature of the display screen are dynamically adjusted after algorithm processing to realize intelligent backlight control.

[0023] Specifically, the core of intelligent backlight control of display screens is to dynamically adjust screen brightness and color temperature according to ambient lighting conditions, thereby improving the visual experience and saving energy. To achieve this goal, light sensors need to be arranged reasonably to ensure that they can accurately perceive the changes in lighting in different areas of the screen. The following will analyze in detail how to use genetic algorithms to optimize sensor layout schemes. First, it is necessary to obtain information about different areas of the display screen. For example, a 15.6-inch laptop screen is divided into 9 3x3 areas, and the center coordinates and area size of each area are recorded respectively. For example, the coordinates of the center area can be (7.8 inches, 5.2 inches), and the area is approximately 2 square inches. This information will be used as the input parameters of the genetic algorithm. The core idea of ​​the genetic algorithm is to simulate the biological evolution process and continuously optimize the solution through operations such as selection, crossover, and mutation.

[0024] In this example, each sensor layout scheme is encoded into a chromosome. For example, assuming that 4 sensors are used and the coordinates of each sensor can be represented by two genes, then a chromosome contains 8 genes, such as (x1, y1, x2, y2, x3, y3, x4, y4). The initial population can randomly generate multiple such chromosomes, representing different sensor layout schemes. The fitness function is used to evaluate the pros and cons of each scheme. In this example, the fitness function consists of two parts: the area ratio of the key area covered and the overall uniformity index. The key area can be defined as the area where the user's line of sight is usually concentrated, such as the central area of ​​the screen. The larger the coverage ratio, the better the scheme. The uniformity index measures whether the sensors are evenly distributed on the screen to avoid too dense sensors in some areas and lack of sensors in other areas. For example, four sensors are located in the four corners of the screen. Although the coverage rate is high, the uniformity is poor. The genetic algorithm continuously optimizes the scheme through iterative search. In each iteration, the excellent chromosomes are selected according to the fitness function for crossover and mutation operations to generate new chromosomes. The crossover operation can simulate gene recombination, such as exchanging some genes of two chromosomes. Mutation operations can simulate gene mutations, such as randomly changing the value of a gene. Through continuous iterations, the average fitness of the population will gradually increase, and eventually a better sensor layout solution will be obtained. For example, after 100 iterations, a solution with the highest fitness is obtained, and the sensor coordinates after chromosome decoding are: (2,2), (2,10), (13,2), (13,10). This shows that the four sensors are located near the four corners of the screen, which not only ensures coverage of most areas of the screen, but also takes into account the uniformity of distribution.

[0025] Finally, the sensor needs to be calibrated. After installing the sensor in the predetermined position, use a standard light source to illuminate different areas of the screen, record the sensor output value, and establish a mapping relationship between light intensity and sensor output value. In this way, the sensor can accurately measure the light information of the area. Based on this information, the controller dynamically adjusts the brightness and color temperature of the display. For example, in a bright environment, the controller will increase the screen brightness, while in a dim environment, it will reduce the brightness and adjust the color temperature to make it closer to warm tones, thereby reducing eye fatigue. Through the above method, the sensor layout scheme can be optimized, more accurate intelligent backlight control can be achieved, and the user's visual experience can be improved.

[0026] Step S103, during the dynamic refresh rate test, periodically switch to the preset standard refresh rate signal source, obtain the real-time output data of the light sensor array, fit the sensor output curve by the least squares method, and determine the drift degree of the curve. If the drift exceeds the preset threshold, dynamically adjust the calibration parameters of the sensor array, and update the drift compensation coefficient in the data analysis algorithm to achieve real-time drift calibration.

[0027] According to the preset standard refresh rate signal source, the real-time output data of the optical sensor array is obtained and the data is transmitted to the data processing module. The least squares method is used to perform curve fitting on the sensor output data to obtain the fitting curve equation, and the fitting curve parameters are transmitted to the drift judgment module. In the drift judgment module, the deviation degree between the fitting curve and the standard curve is calculated to determine whether the sensor output has drifted. If the deviation degree exceeds the preset threshold, the calibration parameter adjustment is triggered. When the sensor output is detected to drift, the calibration parameters of the sensor array are dynamically adjusted, and the sensor output is restored to the vicinity of the standard curve by modifying parameters such as sensor sensitivity and bias voltage. In the data analysis algorithm, the drift compensation coefficient is introduced, and the sensor output data is corrected by updating the compensation coefficient in real time to eliminate the drift effect and improve the test accuracy. The Kalman filter algorithm is used to filter the sensor output data to remove high-frequency noise interference, improve signal quality, and provide high-quality input data for subsequent data analysis. By constructing an adaptive learning model, the sensor drift trend is predicted based on historical data and current sensor output, and the calibration parameters are adjusted in advance to achieve predictive compensation of sensor output, further improving the stability and reliability of dynamic refresh rate testing.

[0028] Specifically, the refresh rate test of the display screen requires accurate light information collection. In order to ensure the stability and accuracy of the light sensor array, it is necessary to judge and compensate for the drift of the sensor output. The light sensor array works according to the preset standard refresh rate signal source, such as 60Hz, 120Hz or 240Hz. The sensor converts the perceived light intensity into a digital signal and transmits it to the data processing module in real time. Assuming a 60Hz refresh rate signal source, the light sensor array outputs data every 16.67ms. These data reflect the brightness changes of the display screen at different time points. The least squares method is used to fit the curve of the sensor output data. For example, collect 1 second of data and get 60 data points. With time as the horizontal axis and light intensity as the vertical axis, a curve is fitted using the least squares method. The curve equation can approximately represent the law of sensor output change over time.

[0029] The fitted curve parameters, such as slope and intercept, will be transmitted to the drift judgment module. The drift judgment module compares the fitted curve with the standard curve. The standard curve can be the calibration curve of the sensor when it leaves the factory, or it can be a reference curve generated based on the expected light intensity value of a specific area of ​​the display. Assume that the standard curve is a horizontal straight line, representing an ideal stable light output. By calculating the degree of deviation between the fitted curve and the standard curve, such as the root mean square error, it can be determined whether the sensor output has drifted.

[0030] If the degree of deviation exceeds a preset threshold, such as 5%, the calibration parameter adjustment is triggered. When sensor drift is detected, the sensor calibration parameters need to be adjusted dynamically. For example, if the sensor output value is low, the output value can be increased by increasing the sensor sensitivity or reducing the bias voltage. Conversely, if the output value is high, the output value can be reduced by reducing the sensitivity or increasing the bias voltage. These adjustments are all to restore the sensor output to the vicinity of the standard curve to ensure measurement accuracy. In the data analysis algorithm, the introduction of a drift compensation coefficient can further improve the test accuracy. For example, if it is found that the sensor output value is continuously 10% higher, all output values ​​can be multiplied by a compensation coefficient of 0.9 in subsequent measurements. This compensation coefficient can be updated in real time according to the drift situation to eliminate the impact of drift.

[0031] To improve signal quality, the sensor output data can be filtered using a Kalman filter algorithm. Kalman filtering can effectively remove high-frequency noise interference, such as noise caused by power supply fluctuations or changes in ambient light. This can provide cleaner and more reliable data for subsequent data analysis. By building an adaptive learning model, the sensor drift trend can be predicted and the calibration parameters can be adjusted in advance. For example, based on historical data, if a sensor is found to be prone to output value drift in a high temperature environment, then when it is known in advance that the ambient temperature will rise, the calibration parameters of the sensor can be adjusted in advance to achieve predictive compensation. This will further improve the stability and reliability of the dynamic refresh rate test.

[0032] In step S104, a data acquisition method based on compressed sensing is used to sparsely represent and compress high-dimensional optical signal data at the sensor array end, thereby significantly reducing the data transmission bandwidth requirement; at the background processing system end, an L1 norm minimization reconstruction algorithm is used to achieve high-precision signal recovery and restoration, thereby ensuring the real-time and reliability of data transmission.

[0033] Acquire high-dimensional optical signal data collected by the sensor array. Preprocess the acquired optical signal data, including noise removal and normalization. If the signal-to-noise ratio of the optical signal data is lower than the preset threshold, perform noise removal; if the amplitude difference of the optical signal data is too large, perform normalization. Use the compressed sensing method to sparsely represent the preprocessed optical signal data. Obtain the sparse basis of the optical signal through the dictionary learning algorithm, and project the optical signal onto the sparse basis to obtain the sparse coefficient. Compress the optical signal data according to the sparse coefficient to obtain compressed data. Adjust the number of sparse coefficients according to the preset compression rate to control the amount of compressed data. Transmit the compressed data to the background processing system. Adaptively adjust the data packet size according to the network bandwidth to ensure the real-time data transmission. After receiving the compressed data, the background system uses the L1 norm minimization algorithm to reconstruct the compressed data. Solve the L1 norm minimization problem according to the pre-established dictionary and sparse coefficient to obtain the reconstructed optical signal data. L1=||y-Φx||22+λ||x||1, where y is the measurement value, Φ is the measurement matrix, x is the sparse coefficient, and λ is the regularization parameter. Perform quality assessment on the reconstructed optical signal data. By calculating the mean square error between the reconstructed signal and the original signal, determine whether the reconstruction accuracy meets the preset requirements.

[0034] Specifically, the optical sensor array can capture rich optical signal information, which usually exists in the form of high-dimensional data. For example, an array containing 100x100 sensors can obtain 10,000 data points in one sampling, forming a high-dimensional vector. The acquired optical signal data is often interfered by various noises, such as the influence of ambient light and the thermal noise of the sensor itself. In addition, there may be response differences between different sensors, resulting in inconsistent data amplitudes. Therefore, the raw data needs to be preprocessed. If the measured optical signal-to-noise ratio is only 10dB, which is lower than the preset 20dB threshold, a noise removal operation is required, such as using a wavelet denoising algorithm. If the output values ​​of each sensor in the sensor array are very different, such as the maximum value is 1V and the minimum value is 0.1V, a normalization operation is required to scale all data to the same range, such as between 0 and 1. Compressed sensing technology can effectively reduce the storage and transmission costs of high-dimensional data. Its core idea is to use the sparsity of the signal to represent the complete signal with a small number of measured values. Take a face image as an example. Although a face image has many pixels, its main features can be represented by a small number of sparse bases, such as the shape of the eyes, nose, and mouth. Through the dictionary learning algorithm, these sparse bases can be learned from a large amount of sample data and used to represent new light signal data. Assuming that the dictionary learning algorithm obtains 100 sparse bases, and the dimension of the original light signal data is 10,000, the light signal data can be projected onto these 100 bases to obtain 100 sparse coefficients, thus achieving data dimensionality reduction. Among the coefficients obtained by sparse representation, many coefficients have very small values ​​close to zero. According to the preset compression rate, only a part of the larger coefficients can be retained, and the remaining coefficients can be set to zero, thereby further compressing the data. For example, if the compression rate is set to 0.1, only the 10 largest sparse coefficients are retained, and the remaining 90 coefficients are set to zero. In this way, the amount of data is reduced from 100 coefficients to 10 coefficients. The compressed data volume is smaller and easier to transmit.

[0035] According to the network bandwidth, the size of the data packet can be adaptively adjusted. For example, when the network bandwidth is low, the data can be divided into multiple small data packets for transmission to ensure the real-time data transmission. Assuming that the network bandwidth is 1Mbps and the compressed data size is 1kb, the data transmission can be completed within 1ms. After the background system receives the compressed data, it needs to reconstruct it to restore the original optical signal data. Using the L1 norm minimization algorithm, the L1 norm minimization problem can be solved according to the pre-established dictionary and sparse coefficient to obtain the reconstructed optical signal data. The L1 norm minimization algorithm can promote sparse solutions, thereby more accurately restoring the original signal. Finally, the quality of the reconstructed optical signal data needs to be evaluated to determine whether the reconstruction accuracy meets the preset requirements. For example, the mean square error between the reconstructed signal and the original signal is calculated. If the mean square error is less than a preset threshold, such as 0.01, it is considered that the reconstruction accuracy meets the requirements. This can effectively ensure the accuracy of subsequent data processing.

[0036] Step S105, in the background processing system, a parallel computing framework based on GPU acceleration is used to perform real-time processing and analysis on the received light sensor array data, and multi-scale similarity matching is performed with a preset standard refresh rate template to quickly determine whether the dynamic refresh rate performance of the display screen meets the standard, and generate an intuitive test report and visualization results.

[0037] The processing platform is built using the GPU parallel computing framework. Obtain the light sensor array data. If the data is empty, re-acquire it; if the data is not empty, preprocess it, including noise removal and data normalization. Extract multi-scale features based on the preprocessed light sensor array data. Obtain the preset standard refresh rate template. For different refresh rates, establish a corresponding standard template library. Perform multi-scale similarity matching between the extracted multi-scale features and the standard refresh rate template to obtain a similarity score. If the similarity score is higher than the preset threshold, it is judged that the dynamic refresh rate performance of the display screen meets the standard; if the similarity score is lower than the preset threshold, it is judged that the dynamic refresh rate performance of the display screen does not meet the standard. Generate a test report and visualization results based on the judgment results, where the visualization results include a similarity score curve and a display refresh rate performance index diagram.

[0038] Specifically, the GPU parallel computing framework can greatly improve the data processing speed. You can choose mature frameworks such as CUDA or OpenCL to build a processing platform. Using the multi-core architecture of the GPU, a large amount of sensor data can be processed in parallel, thereby speeding up the calculation speed of steps such as preprocessing, feature extraction, and similarity matching, and ultimately shortening the time of the entire detection process. For example, a GPU with thousands of CUDA cores can process data from thousands of sensors at the same time, which can be tens or even hundreds of times more efficient than the traditional CPU serial processing method. Acquiring light sensor array data is the first step in the detection process. You can use a high-speed data acquisition card or other interface to acquire data from the sensor array. In order to prevent data acquisition failure from causing program exceptions, an empty data check is required. If the acquired data is empty, it needs to be reacquired until valid data is acquired. This ensures the smooth progress of subsequent processing steps. For example, you can set up a loop that tries to acquire sensor data each time until the data is not empty before exiting the loop. The preprocessing stage includes noise removal and data normalization.

[0039] Noise removal can use methods such as wavelet transform to filter out noise components in the signal and improve the signal-to-noise ratio. For example, if there is high-frequency noise in the sensor data, wavelet transform can be used to decompose and remove the high-frequency components, thereby retaining useful low-frequency signals. Data normalization can scale the data of different sensors to the same numerical range, such as between 0 and 1. This can avoid the impact of data amplitude differences on subsequent feature extraction and similarity matching. For example, if different sensors have different sensitivities, resulting in large differences in output data amplitudes, all data can be scaled to the same range through normalization operations, thereby eliminating the impact of sensor sensitivity differences.

[0040] Multi-scale feature extraction can more comprehensively describe the dynamic refresh rate performance of the display. Wavelet transform or other multi-scale analysis methods can be used to extract features of different scales. Features of different scales can reflect the refresh characteristics of the display at different time scales. For example, small-scale features can reflect the fast refresh characteristics of the display, while large-scale features can reflect the overall refresh characteristics of the display. Combining multi-scale features for analysis can more accurately evaluate the dynamic refresh rate performance of the display. For example, the energy distribution of the brightness signal of the display at different frequencies can be extracted as multi-scale features. The establishment of a standard refresh rate template library is crucial. For different refresh rates, corresponding standard templates need to be established in advance. These templates can be the refresh data of the pre-measured standard display or theoretical data generated according to the refresh rate standard. The degree of perfection of the standard template library directly affects the accuracy of the test results. For example, corresponding standard templates can be established for common refresh rates such as 60 Hz, 120 Hz, and 144 Hz. Multi-scale similarity matching is a key step in judging the refresh rate performance of the display.

[0041] The extracted multi-scale features are compared with the standard refresh rate template to obtain a similarity score. The higher the similarity score, the closer the refresh rate performance of the display is to the standard template. For example, a cross-correlation algorithm can be used to calculate the similarity between the multi-scale features and the standard template. The comparison of the similarity score with the preset threshold determines the final detection result. If the similarity score is higher than the preset threshold, the dynamic refresh rate performance of the display is judged to be up to standard; otherwise, it is not up to standard. The selection of the threshold needs to be adjusted according to the actual application scenario and accuracy requirements. For example, for application scenarios with high requirements for refresh rate performance, a higher threshold can be set. The detection report and visualization results can clearly show the detection results. The detection report can include refresh rate performance indicators of the display, such as refresh rate, response time, etc. The visualization results can include a similarity score curve and a display refresh rate performance indicator graph, so that users can intuitively understand the performance of the display. For example, a curve showing the change of the similarity score over time and a comparison graph of the display refresh rate and the standard refresh rate can be drawn. This can help users quickly determine whether the refresh rate performance of the display meets the standard.

[0042] Step S106, based on the dynamic refresh rate detection result of the display screen, an incremental learning method is used to continuously optimize and update the parameter control strategy of the image generator. Through a reinforcement learning algorithm, such as Qlearning, the image generator can adaptively adjust the parameter combination of the generated image according to the detection feedback, thereby continuously improving the comprehensiveness and authenticity of the test scene coverage.

[0043] According to the pre-processed dynamic refresh rate detection result data, the image generator parameter control strategy model is constructed. The image is generated by using the image generator parameter combination, and the coverage of the image in the test scene is evaluated to obtain the coverage evaluation result. If the coverage evaluation result does not reach the preset threshold, the parameters of the parameter control strategy model are adjusted according to the coverage evaluation result and the feedback mechanism of the reinforcement learning algorithm.

[0044] Specifically, the dynamic refresh rate of the display screen affects the user experience, especially in fast-changing scenarios, such as games or video playback. Dynamic refresh rate detection is to evaluate the refresh rate performance of the display screen in different scenarios. The purpose of obtaining these test result data is to subsequently analyze and improve the performance of the display screen, and ultimately improve the user experience. For example, test data of different types of display screens (such as LCD, OLED, Mini LED) at different refresh rates (such as 60Hz, 120Hz, 144Hz) can be collected, including response time, flicker, degree of smear, etc. The dynamic refresh rate detection result data usually contains a variety of information, such as timestamp, refresh rate value, test scenario, etc. There may be noise or outliers in these data, which need to be cleaned and denoised. For example, a filtering algorithm can be used to remove noise, or outliers can be identified and removed through statistical analysis. Data normalization is to convert data of different dimensions into the same range, such as between 0 and 1, for subsequent model training. For example, the response time can be normalized to between 0 and 1, and the smaller the value, the faster the response speed. The goal of the image generator parameter control strategy model is to find the best parameter combination so that the generated images can cover various test scenarios as much as possible and have high authenticity. Using the incremental learning method, the model can continuously update parameters according to new data, thereby improving the performance of the model. The initial parameters of the model can be set randomly, such as the learning rate, the number of layers and nodes of the neural network, etc.

[0045] Reinforcement learning algorithms (such as Ql earning) can select parameter combinations for image generators based on the current parameter control strategy model. For example, different parameter combinations can be used as different actions, and coverage and authenticity evaluation results can be used as rewards. By continuously trying different parameter combinations, the best strategy can be learned. The generated images need to be evaluated for coverage in the test scene. The test scene can include different refresh rates, different content types (such as games, videos, static pictures), etc. Coverage refers to how many test scenes the generated image can cover. For example, if the test scene includes three refresh rates of 60Hz, 120Hz and 144Hz, and the generated image can cover these three refresh rates, the coverage is 100%. The coverage evaluation result is used to update the parameter control strategy model. If the coverage is lower than a preset threshold, such as 90%, the parameters of the parameter control strategy model need to be adjusted, such as increasing the number of layers or nodes of the neural network. For example, if the coverage of the image generated at a refresh rate of 144Hz is low, the learning weight of the model for high refresh rate scenes can be increased. The authenticity evaluation of the generated image is to determine whether the generated image is consistent with the actual display refresh rate effect. If the authenticity assessment result is lower than a preset threshold, such as 80%, the parameters of the parameter control strategy model need to be adjusted and the image needs to be regenerated. For example, if the generated image appears blurred or smeared in a high-speed motion scene, the parameters of the model need to be adjusted so that it can generate a clearer and smoother image. The ultimate goal is to find a parameter combination that can generate high-coverage, high-authenticity images, thereby improving the efficiency and accuracy of dynamic refresh rate detection of display screens.

[0046] Step S107, establish an evaluation index system for the dynamic refresh rate performance of the display screen, comprehensively consider multiple indicators such as frame rate stability, color reproduction, motion blur, etc. under different test scenarios, and form a quantitative performance score through a weighted fusion method. Perform a correlation analysis with the subjective evaluation results of the human eye, and continuously optimize and improve the evaluation index system to make it closer to the actual perception of users.

[0047] Obtain multiple performance index data of the display screen. Collect the frame rate fluctuation data, color difference value data and motion blur MTF value data of the display screen under different test scenarios to construct a data set. Preprocess the obtained performance index data. Standardize the frame rate fluctuation data and calculate the frame rate stability index; normalize the color difference value data and calculate the color reproduction index; filter the motion blur MTF value data and calculate the motion blur index. Determine the weights of various performance indicators. According to the visual characteristics of the human eye and the results of user surveys, the weights of indicators such as frame rate stability, color reproduction and motion blur are determined by using the hierarchical analysis method or entropy weight method to form a weight vector. Construct a dynamic refresh rate performance evaluation model for the display screen. According to the idea of ​​weighted fusion, the preprocessed performance index data and the determined weight vector are weighted summed to obtain the quantitative score of the dynamic refresh rate performance of the display screen. Conduct a subjective evaluation experiment of the human eye. Invite multiple subjects to conduct subjective evaluations of different display screens, score the refresh rate performance, and collect subjective scoring data. Analyze the correlation between objective and subjective scores. Calculate the Pearson correlation coefficient or Spearman rank correlation coefficient between the quantitative score of the dynamic refresh rate performance of the display and the subjective evaluation results of the human eye to determine the degree of correlation between the objective evaluation index system and user perception. Optimize and improve the evaluation index system. If the correlation coefficient is lower than the preset threshold, adjust the index weight or add new indicators and repeat the process until the correlation coefficient reaches the preset threshold, so as to obtain an evaluation index system that is closer to the actual user perception.

[0048] The above description is merely a preferred embodiment of one or more embodiments of the present specification and is not intended to limit one or more embodiments of the present specification. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of the present specification shall be included in the scope of protection of one or more embodiments of the present specification.

Claims

1. A production detection method for the dynamic refresh rate of a display screen, characterized in that: The method comprises: Step S101, according to a preset dynamic refresh rate test scene library, multiple test image sequences covering different frame rates, color depths and resolutions are obtained, the test images are generated in real time by an image generator, and the output parameters of the image generator are controlled to dynamically adjust the frame rate, color depth and resolution of the generated images to simulate various dynamic image contents in a real usage environment; Step S102, for different areas of the display screen, a genetic algorithm of a heuristic optimization algorithm is used to determine the optimal layout of the light sensor array through iterative search under the condition of satisfying the overall layout uniformity constraint, while controlling the number of sensors and maximizing the coverage of the key areas of the display screen, so as to achieve global optimization of the sensor layout; Step S103, during the dynamic refresh rate test, periodically switch to the preset standard refresh rate signal source, obtain the real-time output data of the light sensor array, fit the sensor output curve by the least squares method, determine the drift degree of the curve, and if the drift exceeds the preset threshold, dynamically adjust the calibration parameters of the sensor array, and update the drift compensation coefficient in the data analysis algorithm to achieve real-time drift calibration; Step S104, using a data acquisition method based on compressed sensing to sparsely represent and compress high-dimensional optical signal data at the sensor array end, greatly reducing the data transmission bandwidth requirement, and at the background processing system end, using an L1 norm minimization reconstruction algorithm to achieve high-precision signal recovery and restoration, ensuring the real-time and reliability of data transmission; Step S105: In the background processing system, a parallel computing framework based on GPU acceleration is used to process and analyze the received light sensor array data in real time, and a multi-scale similarity matching is performed with a preset standard refresh rate template to quickly determine whether the dynamic refresh rate performance of the display screen meets the standard, and generate an intuitive test report and visualization results; Step S106, according to the dynamic refresh rate detection result of the display screen, the incremental learning method is used to continuously optimize and update the parameter control strategy of the image generator, and the image generator is enabled to adaptively adjust the parameter combination of the generated image according to the detection feedback by using the reinforcement learning algorithm and Q-learning, so as to continuously improve the comprehensive coverage and authenticity of the test scene; Step S107, establish an evaluation index system for the dynamic refresh rate performance of the display screen, comprehensively consider the frame rate stability, color reproduction and motion blur indicators under different test scenarios, form a quantitative performance score through a weighted fusion method, and perform correlation analysis with the subjective evaluation results of the human eye, continuously optimize and improve the evaluation index system, and make it closer to the actual perception of users.

2. The method according to claim 1, characterized in that The step S101 also includes: According to the preset dynamic refresh rate test scene library, multiple test image sequence data covering different frame rates, color depths and resolutions are obtained; Inputting the acquired test image sequence data into an image generator, and generating a test image in real time through the image generator; Obtain output parameters of the image generator, and dynamically adjust the frame rate, color depth and resolution of the image generated by the image generator according to the output parameters; Determine whether the adjusted frame rate, color depth and resolution meet the requirements in the preset dynamic refresh rate test scene library. If so, apply the adjusted parameters to the image generator. If not, continue to adjust the parameters. Controlling the image generator to continuously generate test images according to the adjusted parameters to obtain a dynamically adjusted test image sequence; Input the dynamically adjusted test image sequence into the display device, and present the test image sequence on the display device to simulate various dynamic image contents in a real usage environment; A machine learning algorithm is used to evaluate the quality of images presented by the display device. The parameters of the image generator are further optimized based on the evaluation results to improve the quality of the test image sequence and make it closer to the dynamic image content in the actual usage environment.

3. The method according to claim 1, characterized in that The step S102 also includes: According to different areas of the display screen, the position coordinates and area size parameters of each area are obtained as the input of the genetic algorithm; Genetic algorithm is used to encode the sensor layout scheme into chromosomes, and new layout schemes are generated through crossover and mutation operations; For each layout scheme, calculate the area ratio of the key area of ​​the display screen and the overall uniformity index as the fitness function; Through iterative search, the sensor layout scheme is continuously optimized until the constraints are met or the maximum number of iterations is reached; Select the arrangement scheme with the highest fitness from the final population and decode it into the specific location coordinates of the sensor; Calibrate on the display screen according to the sensor position coordinates to ensure that the sensor can accurately collect light information in the area; The light information collected by the sensor is transmitted to the controller, and the brightness and color temperature of the display are dynamically adjusted after algorithm processing to achieve intelligent backlight control.

4. The method according to claim 1, characterized in that: The step S103 also includes: According to a preset standard refresh rate signal source, real-time output data of the light sensor array is obtained, and the data is transmitted to a data processing module; The least square method is used to perform curve fitting on the sensor output data to obtain the fitting curve equation, and the fitting curve parameters are transmitted to the drift judgment module; In the drift judgment module, the deviation between the fitting curve and the standard curve is calculated to determine whether the sensor output has drifted; If the degree of deviation exceeds a preset threshold, the calibration parameter adjustment is triggered; When the sensor output is detected to have drifted, the calibration parameters of the sensor array are dynamically adjusted, and the sensor output is restored to the vicinity of the standard curve by modifying the sensor sensitivity and bias voltage parameters; In the data analysis algorithm, the drift compensation coefficient is introduced. By updating the compensation coefficient in real time, the sensor output data is corrected to eliminate the drift effect and improve the test accuracy. The Kalman filter algorithm is used to filter the sensor output data to remove high-frequency noise interference, improve signal quality, and provide high-quality input data for subsequent data analysis; By building an adaptive learning model, the sensor drift trend is predicted based on historical data and current sensor output, and the calibration parameters are adjusted in advance to achieve predictive compensation of sensor output, further improving the stability and reliability of dynamic refresh rate testing.

5. The method according to claim 1, characterized in that The step S104 also includes: Acquire high-dimensional optical signal data collected by the sensor array; Preprocessing the acquired optical signal data, including noise removal and normalization; If the signal-to-noise ratio of the optical signal data is lower than a preset threshold, a noise removal operation is performed; If the amplitude difference of the optical signal data is too large, a normalization operation is performed; The compressed sensing method is used to sparsely represent the preprocessed optical signal data; A sparse basis of the optical signal is obtained by using a dictionary learning algorithm, and the optical signal is projected onto the sparse basis to obtain a sparse coefficient; Compressing the optical signal data according to the sparse coefficient to obtain compressed data; The number of sparse coefficients is adjusted according to the preset compression ratio to control the amount of compressed data; Transmit the compressed data to the background processing system; Adaptively adjust data packets according to network bandwidth to ensure real-time data transmission; After receiving the compressed data, the background system uses the L1 norm minimization algorithm to reconstruct the compressed data; According to the pre-established dictionary and sparse coefficients, the L1 norm minimization problem is solved to obtain the reconstructed optical signal data; L1=||y-Φx||22+λ||x||1, where y is the measurement value, Φ is the measurement matrix, x is the sparse coefficient, and λ is the regularization parameter; Performing quality assessment on the reconstructed optical signal data; By calculating the mean square error between the reconstructed signal and the original signal, it is determined whether the reconstruction accuracy meets the preset requirements.

6. The method according to claim 1, characterized in that The step S105 also includes: Use GPU parallel computing framework to build a processing platform; Acquire light sensor array data; If the data is empty, re-acquire it; If the data is not empty, preprocessing is performed, including noise removal and data normalization; Extracting multi-scale features based on the pre-processed optical sensor array data; Get the preset standard refresh rate template; Establish corresponding standard template libraries for different refresh rates; The similarity score is obtained by calculating the multi-scale similarity matching between the extracted multi-scale features and the standard refresh rate template; If the similarity score is higher than the preset threshold, the dynamic refresh rate performance of the display screen is judged to be up to standard; If the similarity score is lower than the preset threshold, it is judged that the dynamic refresh rate performance of the display screen does not meet the standard; Based on the judgment results, a detection report and visualization results are generated, where the visualization results include a similarity score curve and a display refresh rate performance index graph.

7. The method according to claim 1, characterized in that The step S106 also includes: Constructing the image generator parameter control strategy model according to the preprocessed dynamic refresh rate detection result data; Generate the image by using the image generator parameter combination, and evaluate the coverage of the image in the test scene to obtain the coverage evaluation result; If the coverage evaluation result does not reach the preset threshold, the parameters of the parameter control strategy model are adjusted according to the coverage evaluation result and the feedback mechanism of the reinforcement learning algorithm.

8. The method according to claim 1, characterized in that The step S107 also includes: Obtain multiple performance indicator data of the display screen; Collect the frame rate fluctuation data, color difference value data and motion blur MTF value data of the display screen in different test scenarios to build a data set; Preprocess the acquired performance indicator data; Standardize the frame rate fluctuation data and calculate the frame rate stability index; Normalize the color difference value data and calculate the color restoration index; Filter the motion blur MTF value data and calculate the motion blur index; Determine the weight of each performance indicator; According to the visual characteristics of the human eye and the results of user surveys, the weights of the frame rate stability, color reproduction and motion blur indexes are determined using the hierarchical analysis method or entropy weight method to form a weight vector; Construct a dynamic refresh rate performance evaluation model for display screens; According to the idea of ​​weighted fusion, the preprocessed performance index data is weightedly summed with the determined weight vector to obtain a quantitative score of the dynamic refresh rate performance of the display screen; Conduct subjective evaluation experiments of human eyes; Invite multiple subjects to conduct subjective evaluations of different displays, score the refresh rate performance, and collect subjective scoring data; Analyze the correlation between objective and subjective scores; Calculate the Pearson correlation coefficient or Spearman correlation coefficient between the quantitative score of the dynamic refresh rate performance of the display and the subjective evaluation results of the human eye to determine the degree of correlation between the objective evaluation index system and user perception; Optimize and improve the evaluation indicator system; If the correlation coefficient is lower than the preset threshold, the indicator weight is adjusted or new indicators are added, and the system is rebuilt, evaluated and analyzed until the correlation coefficient reaches the preset threshold, thus obtaining an evaluation indicator system that is closer to the user's actual perception.

Images