An ambient light based display screen adjustment system

Abstract

This invention relates to the field of display screen adjustment technology and discloses a display screen adjustment system based on ambient light. A data acquisition module collects initial ambient light data and visual feature data from the user's historical records. A classification module classifies ambient light intensity using a support vector machine (SVM) method. A clustering module performs cluster analysis on ambient light color temperature using a clustering algorithm to determine the color temperature type of the ambient light. An integration module integrates the results of the classification and clustering modules to obtain key feature information of the ambient light. A calculation module retrieves user visual feature data matching the current ambient light, calculates the optimal adjustment parameters for the display screen, and transmits these parameters to the display screen's control module. A control module adjusts the display screen in real time according to the optimal adjustment parameters and continuously monitors user interaction data after adjustment. This invention improves display performance and meets users' personalized needs.

Description

Technical Field

[0001] This invention relates to the field of display screen adjustment technology, and more specifically to a display screen adjustment system based on ambient light. Background Technology

[0002] With the widespread use of electronic devices, the display screen, as a crucial interface for human-computer interaction, has a vital impact on user experience. Under different ambient light conditions, users have varying needs regarding parameters such as brightness and contrast. Current technologies mostly employ fixed adjustment modes or rely solely on simple sensors to obtain ambient light intensity and then make linear adjustments. However, this approach has several problems. Firstly, simple linear adjustment cannot accurately match users' visual needs in different scenarios. For example, in bright light, even with increased brightness, the displayed content may still be difficult to see; in low light, excessive brightness is not only glaring but also increases device power consumption. Secondly, fixed adjustment modes lack consideration for users' personalized visual habits, making it difficult to meet the needs of different users. Summary of the Invention

[0003] The purpose of this invention is to solve the above-mentioned problems by designing a display screen adjustment system based on ambient light.

[0004] This invention provides a display screen adjustment system based on ambient light, the system comprising:

[0005] The data acquisition module is used to collect initial ambient light data in real time using an ambient light sensor and to acquire visual feature data from the user's historical records.

[0006] The classification module is used to classify ambient light intensity based on initial ambient light data using the support vector machine method.

[0007] The clustering module is used to perform cluster analysis on the ambient light color temperature based on the initial data of ambient light, and to determine the color temperature type of the ambient light.

[0008] The integration module is used to integrate the results of the classification module and the clustering module to obtain key feature information of ambient light;

[0009] The calculation module is used to retrieve user visual feature data that matches the current ambient light based on the key feature information of ambient light, construct a multimodal fusion model based on Transformer-GNN, optimize and improve the model with an adaptive attention mechanism, calculate the comprehensive feature representation under dynamic weight allocation through the model, calculate the optimal adjustment parameters of the display screen, and transmit the optimal adjustment parameters to the control module of the display screen.

[0010] The control module is used to adjust the display screen in real time according to the optimal adjustment parameters, and after the display screen is adjusted, it continuously monitors the user's interaction data with the display screen.

[0011] Optionally, in a first implementation of the present invention, the data acquisition module includes:

[0012] The module is used to build a federated learning network between multiple distributed devices and a central server to obtain visual feature data of user history collected locally by each distributed device.

[0013] The configuration submodule is used to set the initial sampling frequency, the initial state vector and covariance matrix of the Kalman filter. The ambient light sensors of each distributed device collect the initial data of the ambient light according to the initial sampling frequency. Using the Kalman filter algorithm, based on the historical ambient light data and the state estimate before the current moment, the change trend and predicted value of the ambient light parameters at the next moment are predicted through the state transition equation.

[0014] The comparison submodule is used to compare the currently collected ambient light data with the predicted value, calculate the actual change in ambient light, and obtain the rate of change of ambient light by combining the sampling time interval. The sampling frequency is then dynamically adjusted according to the rate of change of ambient light.

[0015] Optionally, in a second implementation of the present invention, the classification module includes:

[0016] The standardization submodule is used to extract features related to ambient light intensity classification from the initial ambient light data, combine them into a feature vector, use principal component analysis to reduce the dimensionality of the feature vector, and then standardize the dimensionality-reduced feature vector.

[0017] The mapping submodule is used to map the standardized data to a high-dimensional Hilbert space through a quantum kernel function, encode it into a quantum state, measure the quantum state, and output the quantum kernel matrix.

[0018] The solution submodule is used to construct an SVM model based on the quantum kernel matrix and use optimization algorithms such as gradient descent to solve the dual problem of SVM to obtain the optimal classification hyperplane.

[0019] The classification submodule is used to classify ambient light intensity based on the classification hyperplane, determine the category of ambient light intensity, and output the classification result.

[0020] Optionally, in a third implementation of the present invention, the clustering module includes:

[0021] The extraction submodule is used to extract color temperature data from the initial ambient light data and obtain the temporal information of color temperature according to the collection timestamp;

[0022] The initialization submodule is used to extract latent features and cluster them using the DEC algorithm. It initializes the parameters of GMM based on the results of DEC clustering, inputs the time series information into the LSTM model, uses the LSTM model to predict the color temperature data, and dynamically adjusts the cluster centers using GMM based on the prediction results and the current color temperature data.

[0023] The determination submodule is used to integrate the cluster labels obtained from DEC clustering and the cluster center information adjusted by time-aware clustering to determine the color temperature type of ambient light.

[0024] Optionally, in a fourth implementation of the present invention, the initialization submodule includes:

[0025] The design unit is used to design an autoencoder network, which includes an input layer, a hidden layer and an output layer. Color temperature data is input into the autoencoder, and the output of the hidden layer is extracted as the latent features of the color temperature data.

[0026] The iterative unit is used to cluster the extracted latent features using the DEC algorithm, initialize cluster centers, and iteratively update the cluster centers and cluster assignments to minimize the clustering loss function until convergence, thereby dividing the color temperature data into different categories.

[0027] Optionally, in a fifth implementation of the present invention, the computing module includes:

[0028] The data preparation submodule is used to extract user visual feature data that matches the current ambient light, and to transform the key feature information of the ambient light and the user visual feature data into the feature vector form of a graph neural network.

[0029] The fusion submodule is used to perform multimodal fusion network modeling on the output of the data preparation submodule to obtain the fused feature vector;

[0030] The training submodule is used to take the fused feature vector as the state of reinforcement learning, determine the adjustable parameters of the display screen to obtain the action space of reinforcement learning, construct the DQN network, and train the DQN network for reinforcement learning.

[0031] The optimization submodule is used to dynamically fine-tune the parameters of the reinforcement learning model output by the training submodule. It updates the Q-value mapping relationship of the model in real time through an online learning strategy and performs incremental optimization of the model by combining user interaction data feedback.

[0032] The selection submodule is used to input the fused feature vector into the trained DQN network. Based on the Q value output by the DQN network, the action with the largest Q value is selected, and the display parameters corresponding to this action are the calculated optimal adjustment parameters.

[0033] Optionally, in a sixth implementation of the present invention, the fusion submodule includes:

[0034] The building unit is used to construct the graph edges based on the relationship between the key feature information of ambient light and the user's visual feature data as nodes in the graph.

[0035] The adjustment unit is used to take the constructed graph as the input of the graph neural network, train the graph neural network with historical data, and continuously adjust the network parameters so that the graph neural network can learn the complex correlation patterns between ambient light features and user visual features.

[0036] The first computational unit is used to fuse key feature information of ambient light and user visual feature data through multi-layer propagation and computation of the graph neural network to obtain a comprehensive feature representation.

[0037] Optionally, in a seventh implementation of the present invention, the DQN network includes an input layer, a hidden layer, and an output layer, wherein the output layer outputs the Q value corresponding to each action.

[0038] Optionally, in an eighth implementation of the present invention, the training submodule includes:

[0039] The selection unit is used to select an action using the DQN network in the current state and apply the action to the display screen.

[0040] The continuous monitoring unit is used to continuously monitor the user's fatigue level, use changes in user fatigue level as reward signals, and store the current state, action, reward and next state in the experience replay buffer.

[0041] The second computational unit is used to randomly sample a batch of data from the experience replay buffer for training the DQN network. Based on the sampled data, it calculates the loss function of the DQN network and uses the Adam optimizer to update the parameters of the DQN network to complete the reinforcement learning training of the DQN network.

[0042] Optionally, in a ninth implementation of the present invention, the control module includes:

[0043] The collection submodule is used to collect user interaction data with the display screen through sensors on the local device after the display screen is adjusted. The types of interaction data include at least the accidental touch rate, the force and frequency of touching the screen, and the dwell time when reading text.

[0044] The storage submodule is used to collect user interaction data with the display screen in real time, and to organize and store the interaction data.

[0045] The technical solution provided by this invention utilizes an ambient light sensor to collect initial ambient light data in real time and acquires visual feature data from the user's historical records. Based on the initial ambient light data, a support vector machine method is used to classify the ambient light intensity. Based on the initial ambient light data, a clustering algorithm is used to perform cluster analysis on the ambient light color temperature to determine the color temperature type of the ambient light. The results of the classification module and the clustering module are integrated to obtain key feature information of the ambient light. According to the key feature information of the ambient light, user visual feature data matching the current ambient light is retrieved, and the optimal adjustment parameters of the display screen are calculated and transmitted to the control module of the display screen. The display screen is adjusted in real time according to the optimal adjustment parameters, and after the display screen is adjusted, the interaction data between the user and the display screen is continuously monitored. This invention achieves precise adjustment of the display screen under different ambient light conditions through multi-dimensional analysis of ambient light data and accurate matching of user visual features, effectively solving the problem of inaccurate adjustment in the prior art and improving the display effect. It fully considers the user's personalized visual habits and physiological characteristics, providing customized display screen adjustment solutions for different users, meeting the user's personalized needs, and improving the user experience. Attached Figure Description

[0046] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0047] Figure 1 A schematic diagram of the structure of a display screen adjustment system based on ambient light provided in an embodiment of the present invention;

[0048] Figure 2 This is a schematic diagram of the structure of the data acquisition module provided in an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram of the classification module provided in an embodiment of the present invention. Detailed Implementation

[0050] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0051] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 A schematic diagram of the structure of a display screen adjustment system based on ambient light provided in this embodiment of the invention. The method specifically includes the following steps:

[0052] The data acquisition module is used to collect initial ambient light data in real time using an ambient light sensor and to acquire visual feature data from the user's historical records.

[0053] The classification module is used to classify ambient light intensity based on initial ambient light data using the support vector machine method.

[0054] The clustering module is used to perform cluster analysis on the ambient light color temperature based on the initial data of ambient light, and to determine the color temperature type of the ambient light.

[0055] The integration module is used to integrate the results of the classification module and the clustering module to obtain key feature information of ambient light;

[0056] The calculation module is used to retrieve user visual feature data that matches the current ambient light based on the key feature information of ambient light, construct a multimodal fusion model based on Transformer-GNN, optimize and improve the model with an adaptive attention mechanism, calculate the comprehensive feature representation under dynamic weight allocation through the model, calculate the optimal adjustment parameters of the display screen, and transmit the optimal adjustment parameters to the control module of the display screen.

[0057] The control module is used to adjust the display screen in real time according to the optimal adjustment parameters, and after the display screen is adjusted, it continuously monitors the user's interaction data with the display screen.

[0058] In this embodiment, please refer to Figure 2 The data acquisition module includes:

[0059] The module is used to build a federated learning network between multiple distributed devices and a central server to obtain visual feature data of user history collected locally by each distributed device.

[0060] The configuration submodule is used to set the initial sampling frequency, the initial state vector and covariance matrix of the Kalman filter. The ambient light sensors of each distributed device collect the initial data of the ambient light according to the initial sampling frequency. Using the Kalman filter algorithm, based on the historical ambient light data and the state estimate before the current moment, the change trend and predicted value of the ambient light parameters at the next moment are predicted through the state transition equation.

[0061] The comparison submodule is used to compare the currently collected ambient light data with the predicted value, calculate the actual change in ambient light, and obtain the rate of change of ambient light by combining the sampling time interval. The sampling frequency is then dynamically adjusted according to the rate of change of ambient light.

[0062] In this embodiment, please refer to Figure 3 The classification module includes:

[0063] The standardization submodule is used to extract features related to ambient light intensity classification from the initial ambient light data, combine them into a feature vector, use principal component analysis to reduce the dimensionality of the feature vector, and then standardize the dimensionality-reduced feature vector.

[0064] The mapping submodule is used to map the standardized data to a high-dimensional Hilbert space through a quantum kernel function, encode it into a quantum state, measure the quantum state, and output the quantum kernel matrix.

[0065] The solution submodule is used to construct an SVM model based on the quantum kernel matrix and use optimization algorithms such as gradient descent to solve the dual problem of SVM to obtain the optimal classification hyperplane.

[0066] The classification submodule is used to classify ambient light intensity based on the classification hyperplane, determine the category of ambient light intensity, and output the classification result.

[0067] In this embodiment, the clustering module includes:

[0068] The extraction submodule is used to extract color temperature data from the initial ambient light data and obtain the temporal information of color temperature according to the collection timestamp;

[0069] The initialization submodule is used to extract latent features and cluster them using the DEC algorithm. It initializes the parameters of GMM based on the results of DEC clustering, inputs the time series information into the LSTM model, uses the LSTM model to predict the color temperature data, and dynamically adjusts the cluster centers using GMM based on the prediction results and the current color temperature data.

[0070] The determination submodule is used to integrate the cluster labels obtained from DEC clustering and the cluster center information adjusted by time-aware clustering to determine the color temperature type of ambient light.

[0071] In this embodiment, the initialization submodule includes:

[0072] The design unit is used to design an autoencoder network, which includes an input layer, a hidden layer and an output layer. Color temperature data is input into the autoencoder, and the output of the hidden layer is extracted as the latent features of the color temperature data.

[0073] The iterative unit is used to cluster the extracted latent features using the DEC algorithm, initialize cluster centers, and iteratively update the cluster centers and cluster assignments to minimize the clustering loss function until convergence, thereby dividing the color temperature data into different categories.

[0074] In this embodiment, the calculation module includes:

[0075] The data preparation submodule is used to extract user visual feature data that matches the current ambient light, and to transform the key feature information of the ambient light and the user visual feature data into the feature vector form of a graph neural network.

[0076] The fusion submodule is used to perform multimodal fusion network modeling on the output of the data preparation submodule to obtain the fused feature vector;

[0077] The training submodule is used to take the fused feature vector as the state of reinforcement learning, determine the adjustable parameters of the display screen to obtain the action space of reinforcement learning, construct the DQN network, and train the DQN network for reinforcement learning.

[0078] The optimization submodule is used to dynamically fine-tune the parameters of the reinforcement learning model output by the training submodule. It updates the Q-value mapping relationship of the model in real time through an online learning strategy and performs incremental optimization of the model by combining user interaction data feedback.

[0079] The selection submodule is used to input the fused feature vector into the trained DQN network. Based on the Q value output by the DQN network, the action with the largest Q value is selected, and the display parameters corresponding to this action are the calculated optimal adjustment parameters.

[0080] In this embodiment, the fusion submodule includes:

[0081] The building unit is used to construct the graph edges based on the relationship between the key feature information of ambient light and the user's visual feature data as nodes in the graph.

[0082] The adjustment unit is used to take the constructed graph as the input of the graph neural network, train the graph neural network with historical data, and continuously adjust the network parameters so that the graph neural network can learn the complex correlation patterns between ambient light features and user visual features.

[0083] The first computational unit is used to fuse key feature information of ambient light and user visual feature data through multi-layer propagation and computation of the graph neural network to obtain a comprehensive feature representation.

[0084] In this embodiment, the DQN network includes an input layer, a hidden layer, and an output layer, wherein the output layer outputs the Q value corresponding to each action.

[0085] In this embodiment, the training submodule includes:

[0086] The selection unit is used to select an action using the DQN network in the current state and apply the action to the display screen.

[0087] The continuous monitoring unit is used to continuously monitor the user's fatigue level, use changes in user fatigue level as reward signals, and store the current state, action, reward and next state in the experience replay buffer.

[0088] The second computational unit is used to randomly sample a batch of data from the experience replay buffer for training the DQN network. Based on the sampled data, it calculates the loss function of the DQN network and uses the Adam optimizer to update the parameters of the DQN network to complete the reinforcement learning training of the DQN network.

[0089] In this embodiment, the control module includes:

[0090] The collection submodule is used to collect user interaction data with the display screen through sensors on the local device after the display screen is adjusted. The types of interaction data include at least the accidental touch rate, the force and frequency of touching the screen, and the dwell time when reading text.

[0091] The storage submodule is used to collect user interaction data with the display screen in real time, and to organize and store the interaction data.

[0092] In this embodiment, an ambient light sensor is used to collect initial ambient light data in real time, including but not limited to parameters such as light intensity and color temperature. Simultaneously, visual characteristic data from the user's historical records is obtained from the user device's local database. This data covers basic information such as the user's age, gender, and vision, as well as past operational habits such as manually adjusting screen brightness and contrast, and physiological characteristic data of viewing the screen under different ambient light conditions (such as viewing duration and blink frequency). The collected initial ambient light data and user visual characteristic data are initially processed. An initial sampling frequency, an initial state vector for the Kalman filter, and a covariance matrix are set for the adaptive sampling algorithm. The initial sampling frequency is set to a suitable value. To ensure the initial acquisition of ambient light data, the Kalman filter's state vector contains initial estimates of parameters such as ambient light intensity and color temperature, while the covariance matrix measures the uncertainty of these initial estimates. A federated learning network is established between multiple distributed devices and the central server, creating a secure data communication channel. Each device and the central server agree on data interaction protocols and encryption methods to ensure secure data transmission. The local processing module uses the Kalman filter algorithm to predict the trends and predicted values ​​of ambient light intensity, color temperature, and other parameters at the next moment based on historically acquired ambient light data and state estimates prior to the current moment, using state transition equations. Simultaneously, the prediction is updated based on the state transition matrix and the process noise covariance matrix. Error covariance matrix; compare the currently collected ambient light data with the predicted values ​​to calculate the actual changes in ambient light intensity, color temperature, etc., and then combine this with the sampling time interval to obtain the rate of change of ambient light; based on the calculated rate of change of ambient light, dynamically adjust the sampling frequency according to pre-set rules; if the rate of change of ambient light is large, it indicates that the ambient light is changing drastically, so increase the sampling frequency to capture changes in ambient light more promptly; if the rate of change of ambient light is small, decrease the sampling frequency to reduce data processing load and equipment power consumption; then, the ambient light sensor continues to collect ambient light data according to the new sampling frequency, and repeats the above process of prediction, rate of change calculation, and sampling frequency adjustment; each distributed device collects ambient light data locally. The system collects historical data from ambient light sensors and visual feature data from user history records. For historical ambient light data, it extracts information such as the sensor's original measurement values ​​and acquisition time. For user visual feature data, it compiles data on user age, gender, vision status, manual adjustment of display parameters, and physiological characteristics of viewing the display. Each device performs preliminary processing on the historical ambient light data, extracting feature data related to sensor errors (such as deviations in multiple measurements under the same environment). This processed error calibration data is then encrypted and uploaded to the central server. At the same time, complete visual feature data containing user privacy information is not uploaded; only the processed features used for collaborative calibration are uploaded.The central server receives error calibration data from different devices and uses a federated learning aggregation algorithm (such as federated averaging) to perform aggregation calculations on this data, obtaining a global sensor error calibration model. This model reflects the comprehensive situation and patterns of sensor errors from multiple devices. The central server then distributes the global sensor error calibration model to each distributed device. After receiving the model, each device uses it to calibrate the current data collected by its local ambient light sensor, correcting sensor measurement errors and obtaining more accurate ambient light data. Simultaneously, the calibrated ambient light data is integrated with locally stored visual feature data from user historical records to form a valid dataset for subsequent display adjustment.

[0093] In this embodiment, initial ambient light data is acquired based on an ambient light sensor. This data may contain multi-dimensional information such as light intensity, color, and angle. Simultaneously, user operation records under different ambient light conditions are collected as auxiliary information. The data is checked for missing values ​​and outliers. Missing values ​​can be filled using methods such as mean and median. Outliers are corrected or removed based on the actual situation. Feature extraction: Features related to ambient light intensity classification are extracted from the initial ambient light data and combined into a feature vector. Principal component analysis (PCA) and other methods can be used to reduce the dimensionality of the features and decrease computational load. A suitable quantum circuit structure is selected, and the characteristics of quantum computing are utilized to calculate... The process involves: calculating the kernel function; efficiently calculating the similarity between samples in high-dimensional space using the superposition and entanglement properties of qubits; encoding the feature vectors of the initial ambient light data and mapping them to the quantum state space; executing quantum circuits to measure the quantum states and obtain the kernel function calculation results; model training: constructing an SVM model based on the kernel matrix calculated from the quantum kernel function; using the ambient light feature vectors and their corresponding ambient light intensity classification labels in the training dataset, determining the model parameters (such as Lagrange multipliers, bias terms, etc.) by solving the SVM optimization problem; using optimization algorithms such as gradient descent to solve the dual problem of the SVM to obtain the optimal classification hyperplane; and continuously monitoring new ambient light data. When a new lighting scenario appears (such as a sudden switch to indoor or outdoor light), the new data is labeled and collected; the new data is added to the existing training dataset; incremental learning is used to update the SVM model without retraining the entire model; online learning algorithms, such as incremental versions of Sequence Minimum Optimization (SMO), can be used to adjust the model parameters based on the new data, enabling the model to adapt to the new lighting scenario; adversarial training algorithms (such as Fast Gradient Signed Method, FGSM) are used to generate adversarial examples for the current SVM model; adversarial examples add small perturbations to the original samples, causing the model to misclassify these samples; the generated adversarial examples should... The process involves: 1) Simulating data characteristics under extreme lighting conditions (such as strong glare); 2) Adversarial training: Adding generated adversarial examples to the training dataset; 3) Retraining the SVM model to allow it to learn on both normal and adversarial samples; 4) Continuously adjusting the model's parameters to better resist the interference of adversarial samples and improve robustness under extreme lighting conditions; 5) Inputting ambient light intensity classification features: Inputting the feature vectors of the preprocessed initial ambient light data into the trained and enhanced SVM model; 6) The model calculates the input feature vectors based on the determined classification hyperplane and parameters, determines the ambient light intensity category (such as strong light, medium light, weak light, etc.), and outputs the classification result.

[0094] In this embodiment, an ambient light sensor continuously collects ambient light color temperature data and records the timestamp of the collection to obtain temporal information about the color temperature. Additionally, it collects historical user preference data for different color temperatures, such as records of users manually adjusting the display screen's color temperature. The collected color temperature data is checked for missing or outlier values. Missing values ​​can be filled using linear interpolation; outliers can be identified and corrected using statistical methods (such as based on standard deviation). Normalization is performed on the color temperature data, ensuring its values ​​fall within the range of [0,1] to eliminate the influence of different data scales on subsequent clustering analysis. User preference data is also normalized accordingly. An autoencoder network is designed, comprising an input layer, hidden layers, and an output layer. The input layer receives the preprocessed color temperature data, the hidden layer extracts latent features from the data, and the output layer attempts to reconstruct the input data. The collected color temperature data is used to train the autoencoder. The autoencoder parameters are adjusted by minimizing the error (such as mean square error) between the input data and the reconstructed data, enabling the autoencoder to learn the latent feature representation of the color temperature data. Preprocessed color temperature data is input into a trained autoencoder, and the output of the hidden layer is extracted as the latent features of the color temperature data. The DEC algorithm is used to cluster the extracted latent features. First, cluster centers are initialized, and then the cluster centers and sample cluster assignments are iteratively updated to minimize the clustering loss function until convergence, thus dividing the color temperature data into different categories. An LSTM network is designed, taking a color temperature data sequence containing timestamps as input and a prediction of future color temperature data as output. The LSTM model is trained using historical color temperature time series data, and the parameters of the LSTM model are adjusted by minimizing the error between the predicted and actual values, enabling it to learn the temporal variation patterns of color temperature. Based on the results of DEC clustering, the parameters of the GMM are initialized, including the mean, covariance, and weights of each Gaussian component. The trained LSTM model is used to predict future color temperature data, and the cluster centers are dynamically adjusted using the GMM based on the prediction results and current color temperature data. Specifically, the parameters of the GMM are updated according to the new data so that the cluster centers can adapt to the temporal changes in color temperature (such as color temperature changes caused by sunrise and sunset). Features related to color temperature are extracted from collected historical user preference data, such as the user's preferred color temperature range and color temperature variation patterns. A collaborative filtering algorithm is used to recommend potentially preferred color temperature categories to each user based on user similarity. Specifically, the similarity between users is calculated (e.g., based on cosine similarity), other users similar to the target user are identified, and color temperature categories are recommended based on the preferences of these similar users. Based on the results of the collaborative filtering recommendations, personalized color temperature preference labels are generated for each user, such as "warm color preference user" or "cool color preference user."The cluster labels obtained from DEC clustering, the cluster center information adjusted by time-aware clustering, and the preference labels obtained from user-personalized clustering are integrated to form key feature information of ambient light. Based on the integrated key feature information, the color temperature type of ambient light is determined. For example, combining the clustering results and user preferences, it can be determined whether the current ambient light color temperature belongs to a warm color type suitable for "warm color preference users" or a cool color type suitable for "cool color preference users," etc.

[0095] In this embodiment, the key ambient light feature information obtained from the previous clustering analysis is further organized, including the classification results of ambient light intensity and the clustering type of color temperature. This information is transformed into feature vectors suitable for input into a graph neural network (GNN). User visual feature data matching the current ambient light, such as pupil size and blink frequency, is extracted from the stored data. Simultaneously, user fatigue information is acquired through camera monitoring. The user visual feature data is also converted into a suitable vector representation, maintaining compatibility with the ambient light feature vector in terms of dimension and data format. Ambient light features and user visual features are respectively used as nodes in the graph, and edges are constructed based on their relationships. For example, ambient light intensity and color temperature... Temperature can affect a user's pupil size and blinking frequency, so edges can be established between corresponding nodes. Appropriate attributes are assigned to the nodes and edges in the graph; node attributes are feature vectors, and edge attributes represent the correlation strength between features. A suitable GNN architecture (such as GCN, GAT, etc.) is selected, and the constructed graph is used as input. The GNN is trained using historical data, and by continuously adjusting the network parameters, the GNN can learn the complex correlation patterns between ambient light features and user visual features. After multiple layers of propagation and computation by the GNN, the ambient light features and user visual features are fused to obtain a comprehensive feature representation. The fused feature vector, which contains both ambient light and user visual features, is obtained from the output layer of the GNN. The process involves: 1) Joint information of visual features; 2) Using the fused feature vectors as the state for reinforcement learning, defining the scope and dimensions of the state space; 3) Defining the action space by specifying adjustable parameters of the display screen, such as brightness, color temperature, and contrast, and defining different combinations of these parameters as the action space for reinforcement learning; 4) Constructing a Deep Q-Network (DQN), including an input layer, hidden layers, and an output layer; 5) The input layer receives state information, and the output layer outputs the Q-value corresponding to each action; 6) Randomly initializing the DQN network parameters and creating a target network with the same structure and initial parameters as the DQN; 7) Using the DQN network to select an action (i.e., a set of adjustment parameters for the display screen) in the current state and applying that action to the display screen; 8) Through imaging... The camera continuously monitors user fatigue and uses changes in user fatigue as reward signals. A positive reward is given if user fatigue decreases, and a negative reward is given if fatigue increases. The current state, action, reward, and next state are stored in an experience replay buffer. A batch of data is randomly sampled from the experience replay buffer to train the DQN network. Based on the sampled data, the loss function (e.g., mean squared error) of the DQN network is calculated, and the parameters of the DQN network are updated using an optimization algorithm (e.g., the Adam optimizer). The parameters of the target network are periodically updated to keep them consistent with the parameters of the DQN network. When a new user is encountered, a small amount of ambient light and visual feature data of that user, along with corresponding user fatigue feedback, is collected.Using a meta-learning algorithm (MAML), based on a pre-trained DQN network and limited data from new users, the network parameters are rapidly adjusted to adapt to the visual characteristics of the new users. Through multiple iterations of updating the network parameters, the DQN network can provide personalized display adjustment parameter recommendations for new users. The current ambient light and the user's visual characteristics are fused and input into the meta-learned adapted DQN network. Based on the Q-value output by the DQN network, the action with the largest Q-value is selected, and the display parameters corresponding to this action are the calculated optimal adjustment parameters.

[0096] In this embodiment, a virtual screen model is constructed in digital space based on the physical parameters of the display screen, key features of the current ambient light, and user visual feature data. This model has the same attributes and characteristics as the real display screen. The model can simulate the display effect under different screen parameters (brightness, color temperature, contrast, etc.). The optimal adjustment parameters are input into the virtual screen model to simulate the display effect after adjustment. A Generative Adversarial Network (GAN) is used, combined with historical user satisfaction feedback on different display effects, to predict user satisfaction under the current adjustment parameters. The GAN generator generates simulated user satisfaction data, and the discriminator judges the authenticity of the generated data. Through continuous adversarial training, the prediction accuracy is improved. The accuracy of the measurement is ensured by transmitting the optimal adjustment parameters, evaluated through digital twin simulation, from the local device or the cloud to the display's control module. Based on the received optimal adjustment parameters, the display's control module adjusts the brightness, color temperature, and contrast in real time to achieve the desired display effect. After the display adjustment is complete, the sensors and monitoring programs on the local device are activated to collect user interaction data. The types of interaction data to be monitored are determined, such as accidental touch rate, touch pressure and frequency, and dwell time on text. The local device collects user interaction data in real time and performs preliminary processing and storage. The collected interaction data is then cleaned to remove noise and outliers. The process involves: associating interaction data with current screen parameters to construct a dataset for causal inference analysis; using causal inference models to analyze the causal relationship between interaction data and screen parameters; for example, identifying which screen parameter adjustments led to changes in interaction data and the magnitude of the causal effect of these changes through methods such as propensity score matching and instrumental variable methods; determining whether there is over-adjustment of current screen parameters based on the results of the causal inference analysis; marking parameters that negatively impact interaction data, such as a significant increase in accidental touch rate, if found; performing preliminary analysis and processing of the collected interaction data on the local device to extract key features; and then applying the results of the causal inference model analysis. On the local device, screen parameters are fine-tuned within a small range to respond to user interaction feedback in real time and reduce adjustment latency. Data is then uploaded to the cloud. The local device uploads processed interaction data and key features to the cloud server. Simultaneously, relevant information from the local device during the adjustment process is uploaded, such as changes in adjustment parameters and predicted user satisfaction results. The cloud server receives data from multiple local devices, performs data aggregation and global analysis, and uses this global data to optimize the model, including digital twin simulation models and causal inference models, improving the model's generalization ability and accuracy. The cloud server then distributes the optimized model to each local device, which updates the model, providing more accurate guidance for subsequent display adjustments.

[0097] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A display screen adjustment system based on ambient light, characterized in that, The system includes: The data acquisition module is used to collect initial ambient light data in real time using an ambient light sensor and to acquire visual feature data from the user's historical records. The classification module is used to classify ambient light intensity based on initial ambient light data using the support vector machine method. The clustering module is used to perform cluster analysis on the ambient light color temperature based on the initial data of ambient light, and to determine the color temperature type of the ambient light. The integration module is used to integrate the results of the classification module and the clustering module to obtain key feature information of ambient light; The calculation module is used to retrieve user visual feature data that matches the current ambient light based on the key feature information of ambient light, construct a multimodal fusion model based on Transformer-GNN, optimize and improve the model with an adaptive attention mechanism, calculate the comprehensive feature representation under dynamic weight allocation through the model, calculate the optimal adjustment parameters of the display screen, and transmit the optimal adjustment parameters to the control module of the display screen. The control module is used to adjust the display screen in real time according to the optimal adjustment parameters, and after the display screen is adjusted, it continuously monitors the interaction data between the user and the display screen. The computing module includes: The data preparation submodule is used to extract user visual feature data that matches the current ambient light, and to transform the key feature information of the ambient light and the user visual feature data into the feature vector form of a graph neural network. The fusion submodule is used to perform multimodal fusion network modeling on the output of the data preparation submodule to obtain the fused feature vector; The training submodule is used to take the fused feature vector as the state of reinforcement learning, determine the adjustable parameters of the display screen to obtain the action space of reinforcement learning, construct the DQN network, and train the DQN network for reinforcement learning. The optimization submodule is used to dynamically fine-tune the parameters of the reinforcement learning model output by the training submodule. It updates the Q-value mapping relationship of the model in real time through an online learning strategy and performs incremental optimization of the model by combining user interaction data feedback. The selection submodule is used to input the fused feature vector into the trained DQN network. Based on the Q value output by the DQN network, the action with the largest Q value is selected, and the display parameters corresponding to this action are the calculated optimal adjustment parameters.

2. The display screen adjustment system based on ambient light as described in claim 1, characterized in that, The data acquisition module includes: The module is used to build a federated learning network between multiple distributed devices and a central server to obtain visual feature data of user history collected locally by each distributed device. The configuration submodule is used to set the initial sampling frequency, the initial state vector and covariance matrix of the Kalman filter. The ambient light sensors of each distributed device collect the initial data of the ambient light according to the initial sampling frequency. Using the Kalman filter algorithm, based on the historical ambient light data and the state estimate before the current moment, the change trend and predicted value of the ambient light parameters at the next moment are predicted through the state transition equation. The comparison submodule is used to compare the currently collected ambient light data with the predicted value, calculate the actual change in ambient light, and obtain the rate of change of ambient light by combining the sampling time interval. The sampling frequency is then dynamically adjusted according to the rate of change of ambient light.

3. The display screen adjustment system based on ambient light as described in claim 1, characterized in that, The classification module includes: The standardization submodule is used to extract features related to ambient light intensity classification from the initial ambient light data, combine them into a feature vector, use principal component analysis to reduce the dimensionality of the feature vector, and then standardize the dimensionality-reduced feature vector. The mapping submodule is used to map the standardized data to a high-dimensional Hilbert space through a quantum kernel function, encode it into a quantum state, measure the quantum state, and output the quantum kernel matrix. The solution submodule is used to construct an SVM model based on the quantum kernel matrix and use optimization algorithms such as gradient descent to solve the dual problem of SVM to obtain the optimal classification hyperplane. The classification submodule is used to classify ambient light intensity based on the classification hyperplane, determine the category of ambient light intensity, and output the classification result.

4. The display screen adjustment system based on ambient light as described in claim 1, characterized in that, The clustering module includes: The extraction submodule is used to extract color temperature data from the initial ambient light data and obtain the temporal information of color temperature according to the collection timestamp; The initialization submodule is used to extract latent features and cluster them using the DEC algorithm. It initializes the parameters of GMM based on the results of DEC clustering, inputs the time series information into the LSTM model, uses the LSTM model to predict the color temperature data, and dynamically adjusts the cluster centers using GMM based on the prediction results and the current color temperature data. The determination submodule is used to integrate the cluster labels obtained from DEC clustering and the cluster center information adjusted by time-aware clustering to determine the color temperature type of ambient light.

5. The display screen adjustment system based on ambient light as described in claim 4, characterized in that, The initialization submodule includes: The design unit is used to design an autoencoder network, which includes an input layer, a hidden layer and an output layer. Color temperature data is input into the autoencoder, and the output of the hidden layer is extracted as the latent features of the color temperature data. The iterative unit is used to cluster the extracted latent features using the DEC algorithm, initialize cluster centers, and iteratively update the cluster centers and cluster assignments to minimize the clustering loss function until convergence, thereby dividing the color temperature data into different categories.

6. The display screen adjustment system based on ambient light as described in claim 1, characterized in that, The fusion submodule includes: The building unit is used to construct the graph edges based on the relationship between the key feature information of ambient light and the user's visual feature data as nodes in the graph. The adjustment unit is used to take the constructed graph as the input of the graph neural network, train the graph neural network with historical data, and continuously adjust the network parameters so that the graph neural network can learn the complex correlation patterns between ambient light features and user visual features. The first computational unit is used to fuse key feature information of ambient light and user visual feature data through multi-layer propagation and computation of the graph neural network to obtain a comprehensive feature representation.

7. The display screen adjustment system based on ambient light as described in claim 1, characterized in that, The DQN network includes an input layer, a hidden layer, and an output layer, wherein the output layer outputs the Q value corresponding to each action.

8. The display screen adjustment system based on ambient light as described in claim 1, characterized in that, The training submodule includes: The selection unit is used to select an action using the DQN network in the current state and apply the action to the display screen. The continuous monitoring unit is used to continuously monitor the user's fatigue level, use changes in user fatigue level as reward signals, and store the current state, action, reward and next state in the experience replay buffer. The second computational unit is used to randomly sample a batch of data from the experience replay buffer for training the DQN network. Based on the sampled data, it calculates the loss function of the DQN network and uses the Adam optimizer to update the parameters of the DQN network to complete the reinforcement learning training of the DQN network.

9. The display screen adjustment system based on ambient light as described in claim 1, characterized in that, The control module includes: The collection submodule is used to collect user interaction data with the display screen through sensors on the local device after the display screen is adjusted. The types of interaction data include at least the accidental touch rate, the force and frequency of touching the screen, and the dwell time when reading text. The storage submodule is used to collect user interaction data with the display screen in real time, and to organize and store the interaction data.

Images