A large screen display system based on machine learning

Abstract

The application belongs to the technical field of screen display, and particularly relates to a large-screen display system based on machine learning, which comprises an environment sensing module, a user sensing module, a calculation module, a compensation module and a display control module; the compensation module calculates a distance and a viewing angle according to positioning coordinate parameters, generates a corrected brightness adjustment coefficient by compensating an initial brightness adjustment coefficient according to the distance, the viewing angle and a pupil diameter parameter when the distance and the viewing angle exceed preset distance and viewing angle threshold values, and the corrected brightness adjustment coefficient is adjusted by considering the distance of the viewer from the large screen and the viewing angle, so that the large-screen display system can more accurately dynamically adjust the screen brightness according to the visual comfort level of the viewer and the influence of ambient light.

Description

Technical Field

[0001] This invention belongs to the field of screen display technology, and specifically relates to a large-screen display system based on machine learning. Background Technology

[0002] Large screens are used in transportation, intelligent manufacturing, energy management, and production construction to monitor and display key data in real time. For example, in transportation, large screens monitor traffic flow and vehicle conditions in real time, allowing for timely adjustments to traffic signals and improving road efficiency. In practical use, they provide a wide field of view, making it easy for viewers in different positions to clearly see the displayed content and grasp the information presented. In a large screen display system, the viewer's position determines their viewing angle (i.e., viewing angle). According to optical principles, as light reflects from the screen surface to the human eye, the effective viewing angle of a large screen refers to the maximum angle at which the viewer can clearly see the image. Beyond this range, the image may become blurry or difficult to read due to insufficient brightness or glare.

[0003] Even if the screen's display brightness remains consistent, the actual brightness perceived by the viewer will vary depending on their position. When the viewer is positioned to the side or at an angle to the screen, ambient light may be reflected off the screen surface and into their eyes, causing glare or reflections. This reflection reduces the image's contrast and affects the visual experience.

[0004] Therefore, large screens need to be able to adjust their brightness in a targeted manner based on changes in distance and viewing angle caused by changes in the viewer's position. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, this invention provides a large-screen display system based on machine learning, which solves the problem of blurred images caused by changes in distance and viewing angle due to the viewer's position.

[0006] The objective of this invention can be achieved through the following technical solution: a large-screen display system based on machine learning, characterized in that:

[0007] It includes an environmental perception module, a user perception module, a computing module, a compensation module, and a display control module;

[0008] The environmental sensing module is used to acquire environmental parameters, including ambient light intensity.

[0009] The user perception module is used to obtain the viewer's pupil diameter parameters and positioning coordinate parameters;

[0010] The user perception module and the user perception module respectively transmit the acquired data to the calculation module, and the calculation module calculates the initial brightness adjustment coefficient based on the environmental parameters;

[0011] The compensation module calculates the distance and viewing angle based on the positioning coordinate parameters. When the distance and viewing angle exceed the preset distance threshold and viewing angle threshold, the module compensates the initial brightness adjustment coefficient based on the distance, viewing angle and pupil diameter parameters to generate a corrected brightness adjustment coefficient.

[0012] The display control module dynamically adjusts the screen brightness according to the corrected brightness adjustment coefficient.

[0013] Preferably, the compensation module compensates for the initial brightness parameters by including the following steps:

[0014] Based on the pupil diameter parameter, a pre-trained machine learning model is used to predict the viewer's current visual comfort level V;

[0015] Based on the positioning coordinate parameters, the viewer's viewing angle is calculated, and the ambient light influence factor R is evaluated.

[0016] Based on the viewer's current visual comfort level and viewing angle, the initial brightness adjustment coefficient is compensated to generate a corrected brightness adjustment coefficient;

[0017] Here, compensating for the initial brightness adjustment coefficient includes the following steps:

[0018] C_correction = C_initial ;

[0019] Here, Ccorrected is the corrected brightness adjustment coefficient, Cinitial is the initial brightness adjustment coefficient, V is the visual quantization, and R is the ambient light intensity influence factor. These are the weighting coefficients.

[0020] Preferably, the compensation module includes a model training submodule, which includes a data collection unit, a data preprocessing unit, a data feature unit, and a data training unit.

[0021] The data collection unit is used to obtain historical pupil diameter parameters and ambient light intensity parameters;

[0022] The data preprocessing unit is used to perform initial filtering on the historical pupil diameter parameters and ambient light intensity parameters using the wavelet transform method through the machine learning model, then perform secondary filtering on the data after initial filtering using a Kalman filter, then perform noise reduction on the data after secondary filtering, and supplement missing data using mean padding and KNN interpolation, and then normalize the supplemented data using a linear transformation method.

[0023] The data feature unit is used to determine the original features, use principal component analysis to reduce the dimensionality of the original features, and then classify and encode the original features to convert them into numerical features.

[0024] The data training unit is used to train the neural network model to obtain the preset machine learning model, wherein the neural network model includes one of convolutional neural networks, recurrent neural networks, and graph neural networks.

[0025] Preferably, the model training submodule further includes a data evaluation unit, which is used to evaluate the preset machine learning model using test data. The evaluation metrics include at least one of mean absolute variance, mean square error, root mean square error, and R-variance.

[0026] Preferably, the test data is generated iteratively using a generative adversarial network (GAN), and the training set of the GAN consists of historical pupil diameter parameters and ambient light intensity parameters.

[0027] Preferably, it also includes a display feedback adjustment module, which is used to receive feedback results from the current user after adjusting the screen brightness.

[0028] Preferably, the model training submodule further includes an optimization unit, which is used to optimize the parameters and structure of the preset machine learning model based on the evaluation results and feedback results.

[0029] Preferably, it also includes a display recognition module, which is electrically connected to the display control module. The display recognition module transmits the recognized content to the display control module. The display content includes both dynamic and non-dynamic content. The display control module compensates for the brightness adjustment coefficient based on the display content.

[0030] Preferably, the compensation is that when it is determined that the screen content is in a dynamic state, the display control module increases the display brightness of the screen.

[0031] The beneficial effects of this invention are as follows:

[0032] This invention, through its compensation module in this embodiment, compensates for changes in the user's pupil diameter parameters and position coordinates to obtain a corrected brightness adjustment coefficient. This corrected brightness adjustment coefficient takes into account the viewer's distance from the screen and viewing angle, enabling the large-screen display system to more accurately adjust screen brightness dynamically based on the viewer's visual comfort level and ambient light influence. This effectively reduces viewer eye strain and improves the viewing experience. Furthermore, the machine learning-based approach allows the system to continuously learn and optimize, adapting to different viewing scenarios and user needs. Attached Figure Description

[0033] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0034] Figure 1 This is a system block diagram of the display system of the present invention. Detailed Implementation

[0035] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0036] Please see Figure 1 This embodiment provides a large-screen display system based on machine learning, including an environment perception module, a user perception module, a calculation module, a compensation module, and a display control module.

[0037] The environmental sensing module is used to acquire environmental parameters in real time, including ambient light intensity, ambient color temperature, and the rate of change of both. The environmental sensing module can use light sensors such as photoresistors or photodiodes to measure ambient light intensity and convert the light intensity signal into an electrical signal. Then, it calculates the light intensity value based on the electrical signal. It uses an RGB color sensor or a dedicated color temperature sensor (such as Taos TCS34725) to determine the color composition of the ambient light and derive the color temperature information. It performs time series analysis on the obtained light intensity value and color temperature information, uses differential algorithms to obtain the light intensity value-time curve and the color temperature-time curve, and calculates the rate of change of both.

[0038] The user perception module is used to obtain the viewer's pupil diameter parameters, location coordinate parameters, and viewing time parameters in real time.

[0039] This system combines infrared cameras, infrared sensors, and eye-tracking devices with computer vision technology to capture the user's pupil diameter parameters in real time. It can also use edge detection technology to extract the pupil area from the acquired video stream and further measure its diameter. Positioning sensors are used to detect the position coordinate parameters of multiple viewers.

[0040] The viewing time parameter here uses the duration of continuous eye contact with the large screen as the starting point, calculating the longest viewing time among all viewers. As described above, the viewer's location coordinates are used to first locate the viewer, and then the pupil diameter parameters at different positions are obtained. Since the detection instruments used to measure the viewer's location coordinates and pupil diameter parameters include timestamps when recording the results, the location coordinate parameters here can be obtained using general detection instruments capable of locating the viewer's position.

[0041] Therefore, based on this, Shapelet extraction, downsampling techniques, or other corresponding extraction techniques can be used to extract a continuous sequence of a specified time length from the detection parameters of the infrared recognition sensor. This sequence is then used as the cumulative viewing time on the large screen at the current moment. After the continuous time sequence ends, the previously accumulated viewing time is reset to zero and recalculated. Since the eye's focus adjusts when the user's eyes leave the electronic screen and look at a more distant object, eye fatigue is alleviated. Therefore, only the data of the viewer with the longest viewing time needs to be calculated. That is, after the user perception module detects the parameters of the viewer, subsequent detection data for each viewer carries an independent label, realizing personalized understanding based on multiple viewers. Based on the label and the cumulative viewing time, the calculation module has a preset viewing time threshold. It can also correspond the viewing time threshold and the number of viewers entering the time threshold according to a piecewise function. When the calculation module detects that the viewing time of the viewer with the longest viewing time exceeds the viewing time threshold, it reduces the screen refresh rate when displaying static content to reduce screen flicker; and increases the screen refresh rate when displaying dynamic content to make the picture smoother.

[0042] The user perception module and the user perception module respectively transmit the acquired data to the calculation module. The calculation module calculates the initial brightness adjustment coefficient based on the environmental parameters. The calculation module first uses the ambient light intensity and color temperature parameters provided by the environmental perception module, combined with a predefined ambient light response model (such as a linear or nonlinear function). Here, the calculation module adjusts the display brightness of the large screen based on the parameters obtained by the perception module, using a general large screen brightness adjustment technology that adapts to the environment based on environmental parameters and its own set operating parameters.

[0043] The compensation module calculates the distance and viewing angle based on the positioning coordinate parameters. When the distance and viewing angle exceed preset distance and viewing angle thresholds, it generates a corrected brightness adjustment coefficient after compensating the initial brightness adjustment coefficient based on the distance, viewing angle, and pupil diameter parameters. The compensation based on the initial brightness adjustment coefficient includes the following steps:

[0044] S1: Based on the pupil diameter parameter, use a pre-trained machine learning model to predict the viewer's current visual comfort level V, including the following sub-steps:

[0045] S11: Selection and Training of Machine Learning Model: A neural network model was selected. During the training phase, a large amount of pupil diameter parameters and corresponding viewer visual comfort evaluation data were collected as the training set. Visual comfort evaluations can be collected through subjective questionnaires or objective methods such as eye tracking;

[0046] S12: Model Training Process: The neural network model is trained using the pupil diameter parameter from the training set as input and the visual comfort level as output. By continuously adjusting the model's weights and biases, the model can accurately predict the visual comfort level based on the pupil diameter parameter.

[0047] S13: Predict visual comfort level: Input the currently acquired viewer pupil diameter parameters into the pre-trained machine learning model, and the model outputs the viewer's current visual comfort level.

[0048] S2: Based on the positioning coordinate parameters, calculate the viewer's viewing angle and evaluate the ambient light influence factor R, including the following sub-steps:

[0049] S21: Calculate the viewing angle: Based on the positioning coordinate parameters, calculate the viewer's viewing angle relative to the large screen using geometric calculation methods. The center point of the large screen can be used as the origin, and the viewer's positioning coordinates can be... The viewing angle can be obtained through vector calculations and trigonometric function relationships;

[0050] S22: Assessing the Ambient Light Influence Factor R: Based on viewing angle and ambient light intensity, assess the degree of influence of ambient light on the viewer's vision. For example, when the viewer is positioned to the side of a large screen and the ambient light is strong, the reflection of ambient light may have a significant impact on vision; in this case, the R value is larger. Conversely, when the viewing angle is frontal and the ambient light is dim, the R value is smaller. A large amount of experimental data can be obtained through eye simulation experiments to establish the mapping relationship between viewing angle, ambient light intensity, and the R value.

[0051] S3: Based on the viewer's current visual comfort level and viewing angle, the initial brightness adjustment coefficient is compensated to generate a corrected brightness adjustment coefficient;

[0052] Here, the corrected brightness adjustment factor is calculated using the following formula, in conjunction with the initial brightness adjustment factor:

[0053] C_correction = C_initial ;

[0054] Here, Ccorrected is the corrected brightness adjustment coefficient, Cinitial is the initial brightness adjustment coefficient, V is the visual quantization, and R is the ambient light intensity influence factor. These are the weighting coefficients.

[0055] The compensation module sends the calculated corrected brightness adjustment coefficient C to the display control module, which then dynamically adjusts the screen brightness based on this coefficient to improve the viewer's visual experience.

[0056] In this embodiment, the compensation module compensates for the initial brightness coefficient of the large screen based on ambient brightness data and set parameter data to obtain a correction coefficient. This correction coefficient takes into account the changes in light perception caused by the viewer's viewing angle. This allows the large-screen display system to more accurately adjust the screen brightness dynamically according to the viewer's visual comfort level and the influence of ambient light, effectively reducing viewer eye strain and improving the viewing experience. Simultaneously, the machine learning-based approach enables the system to continuously learn and optimize, adapting to different viewing scenarios and user needs.

[0057] Because large screens display different content in different usage scenarios, including both static and dynamic content—static content such as images and text, and dynamic content such as videos—and because the human eye has different sensitivities to different types of visual information, when faced with complex visual stimuli (such as fast-moving images in a video), the human eye tends to focus on overall dynamic changes rather than local details. Therefore, in such cases, appropriately high brightness helps enhance contrast and make the image more vivid. Conversely, reading static text requires higher text legibility, and moderately reducing background brightness can reduce glare and thus alleviate eye strain.

[0058] Therefore, in order to optimize the user's cognitive process when facing different content displayed on the screen, one embodiment further includes a display recognition module. The display recognition module and the display control module are electrically connected. The display recognition module transmits the recognition content to the display control module for recognizing the display content of the screen. The display content includes dynamic and non-dynamic, that is, whether the current display content of the electronic screen is dynamic or static. When the content transmitted by the display recognition module is a dynamic signal, the display control module compensates the brightness adjustment coefficient. The compensation includes increasing the display brightness accordingly. Specifically, when the display recognition module recognizes that the display content of the electronic screen is dynamic, the display control module increases the screen brightness by a certain percentage. This percentage can be obtained through corresponding experimental research. The display control module displays the adjusted screen brightness on the screen.

[0059] By categorizing the real-time content displayed on the electronic screen into dynamic and static content, the brightness of the screen can be personalized. This ensures that the screen provides sufficient brightness when playing videos, allowing users to easily track multiple visual focuses. In text reading mode, unnecessary visual interference should be avoided, and the original logic for adjusting the brightness should be maintained. This achieves differentiated brightness control based on the displayed content.

[0060] In one embodiment, the compensation module includes a model training submodule, which includes a data collection unit, a data preprocessing unit, a data feature unit, a data evaluation unit, an optimization unit, and a data training unit.

[0061] The data collection unit is used to obtain historical pupil diameter parameters and historical ambient light intensity parameters;

[0062] The data preprocessing unit is used to process the historical pupil diameter parameters and historical ambient light intensity parameters using a machine learning model as follows:

[0063] Initial filtering: Wavelet transform is used to perform initial filtering on the original data. Wavelet transform can effectively remove high-frequency noise while retaining the main features of the data. For example, for instantaneous fluctuations in ambient light intensity data, wavelet transform can filter out these noises, resulting in a relatively smooth ambient light intensity variation curve.

[0064] Secondary filtering: The data after the initial filtering is further filtered using a Kalman filter, an extended Kalman filter (EKF), or an unscented Kalman filter (UKF). The UKF performs exceptionally well in handling nonlinear systems, providing a better estimate of the true state of the data. Taking pupil diameter as an example, since pupil diameter changes are influenced by various factors and exhibit nonlinear characteristics, the UKF can more accurately track the trend of pupil diameter changes, further improving data accuracy.

[0065] Noise reduction and missing data processing: In order to remove residual minor noise, the data after secondary filtering needs to be denoised, and missing data needs to be filled using mean imputation and KNN interpolation. For example, if the pupil diameter data at a certain moment is missing, the KNN algorithm will calculate a reasonable estimate based on multiple data points at adjacent moments to fill the missing value.

[0066] Normalization: The completed data is normalized using a linear transformation method. All data is mapped to the [0, 1] interval, ensuring that data with different features have the same scale, facilitating subsequent model training and analysis. For example, ambient light intensity data may vary over a large range; after normalization, it is processed at the same scale as data such as pupil diameter parameters, improving the stability and accuracy of model training.

[0067] Data feature units are used for the following purposes:

[0068] The initial features are determined, including ambient light intensity, pupil diameter parameters, and viewer position coordinates. These initial features are extracted directly from the collected data and form the basis for subsequent analysis and processing.

[0069] Dimensionality reduction is performed using Principal Component Analysis (PCA) to reduce the dimensionality of the original features. PCA can transform high-dimensional data into low-dimensional data while retaining the main information. For example, when processing large amounts of viewer location information, PCA can extract a few key principal components, reducing the dimensionality of the data, lowering computational complexity, and without losing too much information.

[0070] Feature transformation converts raw features into numerical features through categorical encoding. For example, viewer location information can be categorized into front row, middle row, back row, etc., and then converted into numerical features using methods such as one-hot encoding to facilitate processing by neural network models.

[0071] Data training unit:

[0072] For model selection, a hybrid neural network model was chosen, combining the strengths of Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Graph Neural Networks (GNN). CNN excels at processing image and spatial data, showing excellent performance in extracting spatial distribution features from ambient light intensity data; RNN can handle time-series data, excelling at analyzing the trend of pupil diameter changes over time; and GNN is suitable for processing data with complex relationships, such as the positional relationships between viewers.

[0073] Model training involves training the hybrid neural network model using preprocessed and feature-engineered data. During training, appropriate hyperparameters, such as the learning rate and number of iterations, are set. By continuously adjusting the model's parameters, it is possible to accurately learn the relationship between factors such as ambient light intensity, pupil diameter, and visual comfort, ultimately obtaining the pre-defined machine learning model.

[0074] In one embodiment, the data evaluation unit is used for the following:

[0075] For evaluation metrics, mean squared error (MSE) and accuracy are chosen to measure model performance. MSE reflects the magnitude of the error between the model's predicted values ​​and the true values, while accuracy evaluates the model's accuracy in classification tasks.

[0076] Evaluation involves using a validation set to assess the trained model. The validation set data is input into the model, and metrics such as the mean squared error (MSE) and accuracy between the model's predictions and the actual values ​​are calculated. Based on the evaluation results, it is determined whether the model meets the expected performance requirements. If the model's performance is unsatisfactory, the evaluation results are fed back to the optimization unit for further optimization.

[0077] To maximize the use of existing data and enhance the model's ability to learn from the real distribution, the test data used in the evaluation unit is generated iteratively using a Generative Adversarial Network (GAN) to learn historical environmental parameters, historical pupil diameter parameters, and historical location coordinate parameters. The test data simulates the distribution characteristics of historical data while introducing a certain degree of diversity or perturbation. The generated test dataset is used to validate the main model and evaluate its performance, such as accuracy and loss value. Since the test data is generated by a GAN, it includes samples similar to but slightly different from historical data, while also introducing new sample variants.

[0078] Because measuring pupil dilation in the human eye requires the test subject to walk at different angles or distances in front of a screen, the process is cumbersome. However, existing technologies include bio-simulation of the eyeball that, through experimental data and models, can biologically display specific eye activities or reactions. For example, CN102698442B discloses a medical simulated human eye, in which the drive motors and support plates of the blinking transmission mechanism and pupil dilation mechanism are fixed to a base. The blinking drive motor transmits power to the eyelids via a crank-rocker transmission mechanism to achieve blinking. The pupil drive motor drives the pupil dilation mechanism installed inside the eyeball to achieve pupil dilation. A pupil size and position judgment mechanism is connected to the pupil dilation mechanism, and a pupil photosensitive mechanism is also connected inside the pupil dilation mechanism. The eyeball is connected to the base via the support plate. This invention integrates blinking and pupil dilation functions into a single unit, with a simple and compact structure and reliable transmission. By adding a control system to control the motor rotation in the subsequent part, it is possible to simulate the effects of blinking at different frequencies and pupil size changes caused by internal and external factors. The above technology is applied to the correspondence between the display brightness of the large screen and the changes in the viewer's pupils. By using the above technology to study the relationship between the changes in the human eye's pupils and the intensity of external light (i.e., the display brightness of the large screen in this case), the display brightness of the large screen is adjusted according to the changes in the pupil diameter. Experimental statistical calculations are performed to obtain the data collected by the data collection unit.

[0079] In one embodiment, when training a pre-defined deep learning model on a large amount of sample data based on pupil diameter and large screen display brightness, the validation step requires calculating the model's diagnostic accuracy, recall, and other metrics on the validation set. Therefore, the model training submodule also includes an optimization unit. This unit optimizes the parameters and structure of the pre-defined machine learning model based on the evaluation and feedback results. Specifically, the optimization unit first optimizes the model parameters using the Adagrad optimization algorithm, adaptively adjusting the learning rate based on the gradient changes of each parameter during training, enabling the model to converge to a better solution more quickly.

[0080] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A large-screen display system based on machine learning, characterized in that: It includes an environmental perception module, a user perception module, a computing module, a compensation module, and a display control module; The environmental sensing module is used to acquire environmental parameters, including ambient light intensity. The user perception module is used to obtain the viewer's pupil diameter parameters and positioning coordinate parameters; The environmental perception module and the user perception module respectively transmit the acquired data to the calculation module, and the calculation module calculates the initial brightness adjustment coefficient based on the environmental parameters; The compensation module calculates the distance and viewing angle based on the positioning coordinate parameters. When the distance and viewing angle exceed the preset distance threshold and viewing angle threshold, the module compensates the initial brightness adjustment coefficient based on the distance, viewing angle and pupil diameter parameters to generate a corrected brightness adjustment coefficient. The display control module dynamically adjusts the screen brightness according to the corrected brightness adjustment coefficient; The compensation module compensates for the initial brightness adjustment coefficient by including the following steps: Based on the pupil diameter parameter, a pre-trained machine learning model is used to predict the viewer's current visual comfort level quantification value V. Based on the positioning coordinate parameters, the viewer's viewing angle is calculated, the mapping relationship between the viewing angle, ambient light intensity and R value is established, and the ambient light influence factor R is evaluated. Based on the viewer's current visual comfort level quantification value and viewing angle, the initial brightness adjustment coefficient is compensated to generate a corrected brightness adjustment coefficient; Here, compensating for the initial brightness adjustment coefficient also includes the following steps: C_correction = C_initial ; Here, Ccorrected is the corrected brightness adjustment coefficient, Cinitial is the initial brightness adjustment coefficient, V is the quantified value of visual comfort level, and R is the ambient light intensity influence factor. These are the weighting coefficients.

2. The large-screen display system based on machine learning according to claim 1, characterized in that: The compensation module includes a model training submodule, which includes a data collection unit, a data preprocessing unit, a data feature unit, and a data training unit. The data collection unit is used to obtain historical pupil diameter parameters and ambient light intensity parameters; The data preprocessing unit is used to perform initial filtering on the historical pupil diameter parameters and ambient light intensity parameters using the wavelet transform method through the machine learning model, then perform secondary filtering on the data after the initial filtering using the Kalman filter, then perform noise reduction on the data after the secondary filtering, and supplement the missing data using mean padding and KNN interpolation, and then perform normalization processing on the supplemented complete data using the linear transformation method. The data feature unit is used to determine the original features, use principal component analysis to reduce the dimensionality of the original features, and then classify and encode the original features to convert them into numerical features. The data training unit is used to train the neural network model to obtain the preset machine learning model, wherein the neural network model includes one of convolutional neural networks, recurrent neural networks, and graph neural networks.

3. The large-screen display system based on machine learning according to claim 2, characterized in that: The model training submodule further includes a data evaluation unit, which is used to evaluate the preset machine learning model using test data. The evaluation metrics include at least one of mean absolute variance, mean square error, root mean square error, and R-variance.

4. A large-screen display system based on machine learning according to claim 3, characterized in that: The test data is generated iteratively using a generative adversarial network (GAN), whose training set consists of historical pupil diameter parameters and ambient light intensity parameters.

5. A large-screen display system based on machine learning according to claim 4, characterized in that: It also includes a display feedback adjustment module, which is used to receive feedback results from the current user after the brightness of the large screen is adjusted.

6. A large-screen display system based on machine learning according to claim 3, characterized in that: The model training submodule also includes an optimization unit, which is used to optimize the parameters and structure of the preset machine learning model based on the evaluation results and feedback results.

7. A large-screen display system based on machine learning according to claim 4, characterized in that: It also includes a display recognition module, which is electrically connected to the display control module. The display recognition module transmits the recognized content to the display control module. The display content includes both dynamic and non-dynamic content. The display control module compensates for the brightness adjustment coefficient based on the display content.

8. A large-screen display system based on machine learning according to claim 7, characterized in that: The compensation is that when the displayed content is determined to be in a dynamic state, the display control module increases the display brightness of the screen.

Images