Intelligent driving method and system for liquid crystal display

Abstract

The application relates to the technical field of liquid crystal display, and discloses an intelligent driving method and system of a liquid crystal display screen, which comprises the following steps: S1, collecting environment, content and user behavior information, and constructing a multi-modal fusion feature vector; S2, extracting features by using a physical enhanced neural network, and generating a visual importance mapping in combination with a liquid crystal response model; S3, dividing display areas based on visual importance and a content change rate, and calculating refresh priority and a refresh rate; S4, determining optimal driving parameters in combination with a multi-target optimization model; S5, performing driving control and collecting response data; and S6, updating a liquid crystal health atlas and network parameters, and realizing closed-loop optimization. The application adopts multi-modal perception and fusion feature vector construction technology, can comprehensively collect environment information, display content information and user behavior data, and generates a high-dimensional feature vector by reasonably fusing the information.

Description

Technical Field

[0001] This invention relates to the field of liquid crystal display technology, specifically to an intelligent driving method and system for a liquid crystal display screen. Background Technology

[0002] With the widespread use of electronic devices, LCD screens have become an indispensable part of daily life. Whether in mobile phones, televisions, computers, or other smart devices, LCD screens play a crucial role in user interaction. However, with the diversification of application needs, traditional LCD screens often struggle to meet the demands of different environments. Especially under the influence of factors such as varying lighting, temperature, viewing angles, and user behavior, the display effect is often unsatisfactory. Therefore, improving the adaptability and display effect of LCD screens in changing environments has become a pressing technological challenge.

[0003] Current LCD screen driving systems primarily rely on static display driving modes, typically adjusting display parameters based on the brightness and contrast of the displayed content to ensure display quality under certain conditions. For example, the adaptive brightness adjustment function in traditional technology can adjust screen brightness under different lighting conditions, ensuring users can clearly see the screen content in both indoor and outdoor environments. Furthermore, LCD screen response optimization technology usually adjusts the display panel voltage, refresh rate, and other parameters to achieve a certain degree of display quality improvement, meeting users' basic visual needs.

[0004] While existing technologies can provide reasonable display effects in some basic scenarios, their performance is significantly inadequate under dynamic changes in environment and user behavior. Firstly, traditional LCD driving methods ignore the impact of environmental information and user behavior on display effects, failing to achieve personalized adjustment. In existing technologies, most display systems fail to dynamically adjust display quality based on information such as the user's viewing area, viewing angle, or dwell time, resulting in inaccurate display effects or wasted resources. Secondly, existing response optimization methods often do not fully consider the physical response characteristics of liquid crystals, such as the impact of temperature and voltage changes on liquid crystal pixels. Therefore, under different usage conditions, changes in display response delay and brightness can lead to a poor visual experience. Thirdly, refresh rate control in existing technologies is mostly statically allocated, unable to dynamically adjust the refresh rate according to changes in displayed content, resulting in wasted power. Finally, existing feedback mechanisms often cannot adaptively optimize based on real-time response data, leading to a gradual decline in display effects over long-term use. Therefore, those skilled in the art propose an intelligent driving method and system for LCD screens to address these problems. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent driving method and system for liquid crystal displays, solving the problem that existing liquid crystal displays cannot dynamically optimize display effects based on environmental changes and user behavior.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent driving method for a liquid crystal display screen, comprising the following steps:

[0007] S1. Collect environmental information, display content information, and user behavior information through sensors to construct a multimodal fusion feature vector;

[0008] S2. Based on the fused feature vector of the multimodal model, spatial and temporal features are extracted using a physical information-enhanced neural network, and a visual importance mapping is generated by combining the liquid crystal physical response model.

[0009] S3. Based on the visual importance mapping and content change rate, divide the display area and calculate the refresh priority and refresh rate of each area;

[0010] S4. Determine the optimal driving parameters based on the refresh priority, computing resource status, and the multi-objective optimization model of display quality, power consumption, and response speed;

[0011] S5. Execute the optimal driving parameters to control the pixel driving circuit of the liquid crystal display to output driving signals and collect liquid crystal response data;

[0012] S6. Update the liquid crystal health status map and neural network parameters according to the response data to achieve closed-loop optimization.

[0013] Preferably, step S1 includes:

[0014] Collect environmental information, including ambient light intensity, display panel temperature, and device attitude angle collected by ambient light sensor, temperature sensor, and attitude sensor, respectively.

[0015] Obtain display content information, wherein the display content information is the image pixel matrix of the current display frame;

[0016] Collect user behavior information, including the user's gaze area, viewing angle, and dwell time;

[0017] Feature extraction and normalization are performed on the environmental information, displayed content information, and user behavior information.

[0018] The normalized features are combined to construct a multimodal fusion feature vector, which serves as the input data for intelligent driving.

[0019] Preferably, step S2 includes:

[0020] Spatial texture and edge features of the current frame image are extracted based on a convolutional neural network.

[0021] Temporal modeling of continuous frame sequences is performed based on recurrent neural networks or Transformer architecture to extract dynamic change trends;

[0022] Construct physical auxiliary features based on liquid crystal response speed, torsion angle, and the relationship between voltage and temperature;

[0023] Feature fusion is performed by combining spatial features, temporal features, and physical auxiliary features;

[0024] Visual importance mappings are computed using a fully connected mapping network to represent the contribution of each pixel to human visual perception.

[0025] Preferably, step S3 includes:

[0026] The display area will be divided into a fixed grid or an adaptive region based on clustering results;

[0027] Calculate the mean visual importance, historical frame change rate, and complexity of displayed content for each region.

[0028] The refresh priority of each region is determined based on a weighted calculation model;

[0029] Based on refresh priority and system resource limitations, a corresponding refresh rate is allocated to each region;

[0030] The output area refresh rate is used as one of the inputs for subsequent multi-objective optimization.

[0031] Preferably, step S4 includes:

[0032] Construct a multi-objective optimization function, with optimization objectives including maximizing display quality, minimizing power consumption, and optimizing response speed;

[0033] The area refresh rate, compensation voltage, and pixel state are treated as optimizable variables.

[0034] Set hardware load, voltage limit, and display stability as constraints;

[0035] Optimal parameter search is performed using either the weighted summation method or the Pareto front method.

[0036] Output the set of driving parameters that achieve the best overall performance under the given constraints.

[0037] Preferably, step S5 includes:

[0038] The optimal driving parameters are mapped to pixel-level driving signals;

[0039] The voltage change of the TFT liquid crystal pixel unit is controlled by the voltage drive module;

[0040] Synchronously collect liquid crystal response data during each driving cycle, including response delay, brightness change and pixel stability;

[0041] The response data is then passed to the subsequent feedback module for status updates.

[0042] Preferably, step S6 includes:

[0043] Based on the collected response data, the current health status of each pixel is modeled and corrected.

[0044] Update the LCD health status map to reflect pixel fatigue, electromigration, and response deviation;

[0045] The updated health map can be used as additional input to retrain or fine-tune the neural network.

[0046] Adjust the visual importance mapping parameters and physical modeling structure;

[0047] Achieve closed-loop optimization and dynamic adaptation of the drive system.

[0048] Preferably, the physical information augmentation neural network adopts a lightweight, variable architecture that supports dividing model depth and complexity by region, specifically including:

[0049] Deep network computation is used for visually important regions;

[0050] For areas of low visual attention, use pruned or quantized smaller networks;

[0051] The model inference process employs parallel processing using heterogeneous computing units;

[0052] Improve inference efficiency for duplicate regions through caching mechanisms.

[0053] Preferably, the physical response model process combines historical voltage-driven data with temperature response curves and is completed through the following processing steps:

[0054] A pixel-level dynamic model is constructed based on the torsional torque of liquid crystal and the applied electric field;

[0055] The actual response time of the fused acquisition is calibrated and corrected;

[0056] The modeling results are used to assist the neural network in predicting nonlinear responses;

[0057] A mapping relationship is established between the response compensation result and the driving signal.

[0058] An intelligent driving system for a liquid crystal display screen includes:

[0059] The multimodal information acquisition module is used to acquire environmental information, display content information, and user behavior information, and to construct a fused feature vector.

[0060] The neural network inference module is used to receive fused feature vectors and extract spatial and temporal features, while fusing physical modeling results to generate a visual importance map;

[0061] The area division and refresh control module is used to divide the display area according to the visual importance mapping and content change rate, and to calculate the refresh priority and refresh rate.

[0062] A multi-objective optimization module is used to construct an optimization function and determine the optimal driving parameters based on display quality, power consumption, and response speed.

[0063] The drive execution module is used to generate drive signals according to the optimal drive parameters to control the display behavior of the liquid crystal pixel units;

[0064] The closed-loop feedback module is used to collect LCD response data and update the LCD health status map and neural network model parameters.

[0065] This invention provides an intelligent driving method and system for a liquid crystal display screen. It has the following beneficial effects:

[0066] 1. This invention employs multimodal perception and fusion feature vector construction technology, which can comprehensively collect environmental information, display content information, and user behavior data, and generate high-dimensional feature vectors by reasonably fusing this information. This technical solution enables the LCD screen driver to accurately adapt to different environments and usage scenarios. Compared with the traditional single input mode in existing technologies, it solves the problem of slow response to changes in environment and user behavior, and significantly improves the personalization and intelligence of the display effect.

[0067] 2. This invention extracts spatial and temporal features through a Physical Information Enhancement Neural Network (PE-NN) and combines it with a liquid crystal physical response model to generate a Visual Importance Map (VIMM), thereby achieving a deep integration of image content and the physical characteristics of liquid crystal displays. This technical solution effectively improves the visual effect and response accuracy of liquid crystal displays, solves the problem of neglecting physical response and dynamic display characteristics in existing technologies, and ensures a high degree of matching between display content and device performance.

[0068] 3. This invention introduces a region division and refresh rate optimization mechanism. Based on visual importance mapping, content change rate and liquid crystal response characteristics, the refresh rate of the display area is reasonably allocated, which optimizes display performance and energy efficiency. This solution enables the liquid crystal display screen to still operate efficiently under high load and complex scenarios, solves the problems of inaccurate refresh rate control and unreasonable resource allocation in the prior art, and improves the smoothness of display effect and energy efficiency.

[0069] 4. This invention uses a closed-loop feedback mechanism to update the liquid crystal health status map based on real-time response data and fine-tune the neural network parameters, thereby achieving dynamic adaptation of the drive system. This technical solution enables the liquid crystal display to maintain a stable display effect during long-term operation and adjust the display parameters in a timely manner, avoiding the problems of display degradation or response lag in the prior art, and significantly improving the long-term stability and reliability of the device. Attached Figure Description

[0070] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0071] Figure 2 This is a schematic diagram of the system architecture of the present invention. Detailed Implementation

[0072] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] Please see the appendix Figure 1 This invention provides an intelligent driving method for a liquid crystal display screen, comprising the following steps:

[0074] S1. Collect environmental information, display content information, and user behavior information through sensors to construct a multimodal fusion feature vector;

[0075] Specifically, in this embodiment, step S1 involves collecting external environmental information, displayed content information, and user behavior information from the LCD screen using multiple sensors to construct a multimodal fusion feature vector. This feature vector will serve as the input for intelligent driving, and subsequent decisions and optimizations will be based on this data. The collection of this information helps to achieve personalized display driving based on environment and user behavior.

[0076] Generally, environmental information, displayed content information, and user behavior information are intertwined and correlated. Therefore, accurately acquiring, processing, and appropriately integrating this information is crucial for subsequent display optimization. In this step, by integrating this information, a more intelligent and dynamic display adjustment capability is provided for the LCD screen.

[0077] Environmental information acquisition: In one possible implementation, environmental information acquisition typically includes data from the following sensors:

[0078] Ambient light sensor: Used to collect the light intensity in the current environment to help adjust the brightness and contrast of the display screen to cope with different ambient light conditions.

[0079] Temperature sensor: Used to monitor temperature changes of the display panel. The performance of LCD screens varies under different temperature conditions, so the input of the temperature sensor is particularly important for optimizing display effects and saving energy.

[0080] Attitude sensor: Used to acquire the device's attitude angles, including pitch, roll, and yaw angles. This information provides more accurate visual results for adjusting the orientation of the screen display, especially in devices with automatically rotating screens.

[0081] Display content information acquisition: Display content information mainly refers to the image pixel matrix of the current display frame. Each frame of the image is formed by the arrangement of pixel arrays, with different brightness and color distributions. For intelligent driving, it is necessary to acquire the data of each frame of the image in real time and extract the feature information of each pixel through image processing algorithms.

[0082] In some embodiments, the acquired display content information may also include data such as content complexity and display change rate. This data will help determine which areas require a higher refresh rate or a stronger drive signal.

[0083] User behavior information collection: User behavior information collection includes the following aspects:

[0084] User gaze area: This feature uses eye-tracking technology to capture the screen area the user is currently looking at, which is crucial for dynamic optimization of the display area. This information helps the system identify the user's focus points, enabling it to provide better display quality in areas of high attention.

[0085] Viewing angle: The relative angle between the user and the screen has a significant impact on visual perception. By obtaining the user's viewing angle, the system can adjust the screen display to ensure the best viewing experience.

[0086] Dwell time: The length of time a user stays in a particular area also reflects their level of interest in that area. Longer dwell times mean that the area may require higher display quality and therefore should be prioritized.

[0087] Data processing and normalization: After collecting the various types of data mentioned above, feature extraction and normalization are required. Specifically, multi-dimensional data such as ambient light, temperature, posture, and gaze area can be standardized to unify data at different scales onto the same dimension.

[0088] For example, the outputs of ambient light intensity and temperature sensors may have different dimensions, so they need to be mapped to a unified scale according to a preset standard range in order to perform subsequent processing and analysis.

[0089] In one embodiment, the normalized features can be processed as follows:

[0090]

[0091] Where: X represents the original data; μ is the mean of the data; σ is the standard deviation; This is the normalized data.

[0092] The above method can eliminate the dimensional differences between different sensor data and convert them into a unified feature vector, which facilitates subsequent fusion and analysis.

[0093] Multimodal fusion feature vector construction: After normalization, all data from environmental information, displayed content information, and user behavior information are combined to form a multimodal fusion feature vector. This vector contains information from different data sources, providing a more comprehensive input for subsequent neural network inference.

[0094] Specifically, this information can be arranged according to certain rules to form a high-dimensional feature vector. For example, environmental information can occupy the first part of the feature vector, displayed content information can occupy the middle part, and user behavior information can occupy the last part. In this case, the fused feature vector can be represented as:

[0095] F = [F env ,F content ,F user ];

[0096] Wherein: F env F is an environmental information vector; content To display the content information vector; F user This is a vector of user behavior information.

[0097] The ultimate goal of this step is to provide a comprehensive input feature vector for subsequent steps. In this embodiment, the multimodal feature vector formed by processing various types of information collected by sensors will be used as input to the neural network and passed to the subsequent neural network inference module for further spatial and temporal feature extraction. The quality and accuracy of this feature vector will directly affect the accuracy of subsequent optimization algorithms and driving parameters, thereby improving the intelligent driving effect of the LCD screen.

[0098] S2. Based on the fused feature vector of the multimodal model, spatial and temporal features are extracted using a physical information-enhanced neural network, and a visual importance mapping is generated by combining the liquid crystal physical response model.

[0099] Specifically, in this embodiment, step S2 introduces a Physical Information Enhancement Neural Network (PE-NN) module, combining deep neural networks with liquid crystal physical modeling to extract spatial and temporal features of the image and the physical response characteristics of the liquid crystal, thereby generating a Visual Importance Map (VIMM). This mapping will provide effective support for subsequent refresh rate control and driving strategies, further improving the intelligent driving effect of the liquid crystal display.

[0100] First, based on the image frame data contained in the fused feature vector, a convolutional neural network (CNN) is used to extract the spatial structure features of the current frame.

[0101] Spatial feature extraction: Let the current frame image be... Where w and h represent the width and height of the image, respectively, and the channel dimension is RGB three channels.

[0102] In one possible implementation, the spatial feature extraction function can be expressed as:

[0103] F spatial =f CNN (C);

[0104] Where: f CNN (·) represents a deep feature extraction network consisting of multiple convolutional layers, activation layers, and pooling layers. Output features Where d represents the number of channels after extraction, and w′ and h′ are the spatial dimensions of the feature map after downsampling.

[0105] Temporal feature extraction: In order to obtain the dynamic change trend of the displayed content in the time dimension, the system also introduces a recurrent neural network structure to model continuous frames.

[0106] In some embodiments, the time series modeling can employ a standard RNN structure, or an LSTM or GRU network to enhance the ability to capture long-term dependencies.

[0107] Let the sequence of consecutive input frames be... Where T is the sequence length, the temporal feature extraction is expressed as:

[0108]

[0109] in: f represents the high-order feature representation of each frame of the image in the time dimension; RNN (·) represents a neural network function used to model time-dependent neural networks.

[0110] Modeling of physical response characteristics of liquid crystal: Based on the above, in order to further improve the model's adaptability to the response characteristics of liquid crystal, the system introduces liquid crystal torsional dynamics modeling and response time modeling as physical auxiliary modules.

[0111] The response behavior of liquid crystal pixels can typically be modeled using the following dynamic equations:

[0112]

[0113] Where: θ represents the torsion angle of the liquid crystal molecules; E is the applied electric field strength; φ is the pretilt angle; K1 and K2 are the elastic constant and electro-optic coupling constant of the liquid crystal material, respectively.

[0114] In addition, to model the effect of driving conditions (such as temperature, applied voltage, etc.) on the liquid crystal response time, the following empirical model is introduced:

[0115]

[0116] Where: τ represents the response time; T is the display panel temperature; V is the voltage value; p is the pixel position or state parameter; α, β, γ are fitting constants; and the function f(p) is used to describe the correction of spatial non-uniformity to the response.

[0117] The results of these physical models will be encoded as auxiliary features and fused with features extracted by neural networks to enhance the accuracy and physical consistency of perception.

[0118] Feature fusion: In one possible implementation, spatial features, temporal features, and response time features can be jointly input into a fusion network to construct the following fusion function:

[0119] F phys =Φ(F spatial ,F temporal ,τ);

[0120] Where: Φ(·) represents the fusion module, which is usually a set of fully connected layers or cross attention layers.

[0121] Visual Importance Map Generation: The final output physical enhancement features will be used to construct a Visual Importance Mapping (VIMM), which measures the importance of different display areas to human visual perception.

[0122] VIMM(x,y)=σ(W·F phys (x,y)+b);

[0123] Where: σ(·) represents the Sigmoid or Softmax function; W is the trainable weight matrix; b is the bias term; Fphys (x,y) represents the fusion feature of the (x,y)th pixel.

[0124] Improved Model Inference Efficiency: As an extension, in some embodiments, this module can employ a variable-structure neural network to improve model inference efficiency and computational resource utilization. In this structure, high visual importance regions use deep network paths, while low-interest regions use pruned or quantized lightweight networks, achieving a balance between heterogeneous computing and energy consumption.

[0125] In addition, to speed up the inference process, the model can be equipped with a caching mechanism to reuse intermediate feature representations in repeated regions, avoid redundant calculations, and improve processing efficiency.

[0126] S3. Based on the visual importance mapping and content change rate, divide the display area and calculate the refresh priority and refresh rate of each area;

[0127] Specifically, in this embodiment, step S3, based on the Visual Importance Map (VIMM) obtained in step S2, further introduces a region division mechanism to spatially reconstruct the entire frame image according to its importance level, in order to support differentiated refresh rate control. This region division not only needs to consider the pixel-level visual weight distribution, but also needs to be dynamically adjusted in combination with multiple factors such as liquid crystal response characteristics, the continuity of the user's gaze area, and the rate of change between frames, thereby providing a basic support for subsequent refresh rate resource allocation and driving parameter configuration.

[0128] This process takes over the output of step S2 and directly transmits the result to the refresh rate control unit, serving as an intermediate bridge link in the entire dynamic drive framework.

[0129] Visual importance mapping normalization and grading: First, based on the visual importance mapping VIMM(x,y) obtained in step S2, the entire frame image is graded for importance. Generally, the visual importance values ​​are normalized, and multiple threshold intervals are set for classification, for example:

[0130]

[0131] in: This represents the importance value after normalization; v min With v max These represent the minimum and maximum values ​​of VIMM in this frame, respectively.

[0132] Region segmentation: In one possible implementation, it is based on normalized visual importance. The entire frame of the image is divided into a High Attention Region (HAR), a Medium Attention Region (MAR), and a Low Attention Region (LAR).

[0133] The specific classification criteria are as follows:

[0134] like Then the pixel belongs to HAR;

[0135] like Then it belongs to MAR;

[0136] like It belongs to LAR.

[0137] Where: θ h With θ l This is an empirical threshold, generally taken between [0.6, 0.9] and [0.2, 0.4]. The specific value can be set according to the experiment or dynamically adjusted.

[0138] Region merging: To reduce the fragmentation of segmented regions, an aggregation-type spatial fusion algorithm is used to enhance the connectivity of the initial region division results. In some embodiments, a method based on connected component growth (CCG) is used to merge consecutive pixel blocks with the same importance level to generate several closed sub-regions.

[0139] Let each subregion be denoted as R. k Where k is the region number, the final frame partition set can be denoted as:

[0140] S = {R1,R2,…,R} K};

[0141] Where: K represents the total number of regions. Ω is the set of pixels in an image frame that satisfies:

[0142] Regional Change Rate Modeling and Refresh Priority: To further improve the spatial efficiency of refresh rate control, it is necessary to statistically model the inter-frame variation frequency between regions. In one implementation, a regional change rate function ρ is introduced. k The definition is as follows:

[0143]

[0144] Where: C t (x,y) represents the pixel value of the current frame, C t-1 (x,y) represents the corresponding pixel value in the previous frame, |Rk | represents the number of pixels within the region, ρ k It reflects the degree of dynamic change in the image content within the area.

[0145] Region refresh priority function: To optimize based on the visual importance, dynamic changes, and liquid crystal response characteristics of each region, a liquid crystal response time model (see step S2) can be embedded into the region evaluation function to weight the refresh priority of different regions. Let the average response time corresponding to each region be... Then refresh the priority function π k The definition is as follows:

[0146]

[0147] in: ω1, ω2, and ω3 are the average visual importance values ​​within the region; ω1, ω2, and ω3 are weighting coefficients that satisfy ω1 + ω2 + ω3 = 1; each parameter can be determined based on a pre-strategy or optimization algorithm.

[0148] Refresh rate allocation and resource scheduling: based on regional refresh priority π k A suitable refresh rate is allocated to each region, and the overall refresh rate scheduling is optimized based on system resource availability. In some embodiments, to improve real-time performance and stability, the region partitioning strategy is executed every N frames, and interpolation is used to map and predict intermediate frames to reduce computational burden. Simultaneously, region information is cached in the driver chip and transmitted to the control unit in a compressed format to ensure a balance between real-time performance and power consumption in refresh scheduling.

[0149] S4. Determine the optimal driving parameters based on the refresh priority, computing resource status, and the multi-objective optimization model of display quality, power consumption, and response speed;

[0150] Specifically, after completing the visual importance mapping and region segmentation processing in step S3, the system has obtained the visual weight and dynamic change information of each region. Step S4 further determines the optimal driving parameters by combining multiple optimization objectives such as display quality, power consumption, and response speed through a multi-objective optimization function. This step provides efficient and balanced parameters for subsequent driving execution, ensuring the best coordination between display effect, energy efficiency, and response speed.

[0151] In this embodiment, a multi-objective optimization function is first constructed, with objectives including maximizing display quality, minimizing power consumption, and optimizing response speed. To effectively optimize these objectives, the system uses refresh rate, compensation voltage, and pixel state of each region as optimizable variables. Simultaneously, the system sets constraints such as hardware load, voltage limits, and display stability to ensure that the optimized drive parameters meet actual hardware and performance requirements.

[0152] Multi-objective optimization model construction: Generally, multi-objective optimization problems involve multiple competing objectives, requiring a reasonable weighting strategy to balance them. In this embodiment, the constructed optimization function includes the following main objectives:

[0153] Display quality: Display quality Q(x) can be measured using image quality metrics, typically based on parameters such as visual appeal and color accuracy. The expression for the display quality objective function is:

[0154] Q(x) = Q VEM (x);

[0155] Among them: Q VEM (x) represents Visual Effect Measurement, used to evaluate the display quality of an image.

[0156] Power consumption: Power consumption P(x) is closely related to the display voltage and refresh rate of the LCD screen. To minimize power consumption, the objective function needs to consider the voltage compensation, refresh rate, and area of ​​each region. The expression for the power consumption objective function is:

[0157]

[0158] Where: V comp (x i ) represents region x i The compensation voltage, R(x) i ) represents region x i refresh rate, A(x) i ) represents region x i The display area.

[0159] The objective function reflects the region's electricity consumption.

[0160] Response speed: The response speed S(x) is closely related to the response time of the liquid crystal pixel. To improve the response speed, it is necessary to minimize the response delay of each region. The expression for the objective function of response speed is:

[0161] S(x) = max(τ(x));

[0162] Where: τ(x) is the response time of region x, representing the time it takes for a pixel to transition from one state to another.

[0163] Constraints: When performing multi-objective optimization, various hardware and physical constraints must also be considered, such as hardware load, voltage limits, and display stability. These constraints can typically be expressed as:

[0164] g j (x)≤0,hk (x) = 0;

[0165] Wherein: g j (x) denotes the inequality constraint, h k (x) represents an equality constraint, which ensures that the optimized solution meets hardware and physical constraints.

[0166] Weighted Summation and Pareto Front: To balance multiple objectives, the weighted summation method or the Pareto front method is generally used to search for optimal parameters. The weighted summation method combines all objective functions into a single objective function by assigning a weight coefficient to each objective. The optimization process can be solved using standard optimization algorithms. The objective function of the weighted summation method can be expressed as:

[0167] F(x)=w1·Q(x)+w2·P(x)+w3·S(x);

[0168] Where w1, w2, and w3 are the weighting coefficients for display quality, power consumption, and response speed, respectively, satisfying w1 + w2 + w3 = 1.

[0169] In another possible implementation, the Pareto front method is used, which obtains a set of balanced optimal solutions by optimizing the non-dominated solutions among multiple objectives.

[0170] Determining the optimal driving parameters: By solving the multi-objective optimization function described above, the system will obtain a set of optimal driving parameters. These parameters include the refresh rate, compensation voltage, and pixel state for each region, etc. The final optimization result can be expressed by the following formula:

[0171]

[0172] in: Let x represent the solution space that satisfies all constraints. * This represents the final optimal solution; the optimized set of driver parameters will be used for subsequent driver execution.

[0173] S5. Execute the optimal driving parameters to control the pixel driving circuit of the liquid crystal display to output driving signals and collect liquid crystal response data;

[0174] Specifically, after completing the multi-objective optimization and determining the optimal driving parameters in step S4, step S5 is mainly responsible for executing the obtained optimal driving parameters and outputting corresponding driving signals by controlling the pixel driving circuit of the liquid crystal display. In addition, the system will synchronously collect the response data of the liquid crystal display in each driving cycle, including response delay, brightness change, and pixel stability. This response data will be used for subsequent state updates and system optimization. Step S5 is a key link in the driving signal execution and feedback loop, providing necessary data support for the dynamic adjustment and optimization of the system.

[0175] In this embodiment, the optimal driving parameters obtained in step S4 are first mapped to specific pixel-level driving signals. These driving signals will be actually executed in the liquid crystal driving circuit to control the voltage change of each pixel in order to achieve the optimized display effect.

[0176] Optimal driving signal generation and mapping: Based on the optimal driving parameters, the system maps these parameters into specific pixel-level driving signals. Specifically, the driving signal D(x,y) for each pixel can be represented by the following formula:

[0177] D(x,y)=f driver (V comp (x,y),R(x,y),A(x,y));

[0178] Where: V comp (x,y) represents the compensation voltage of the (x,y)th pixel, R(x,y) represents the refresh rate corresponding to this pixel, and A(x,y) represents the area displayed by the pixel.

[0179] This drive signal is transmitted to each pixel unit through the liquid crystal driving circuit to control its display state.

[0180] LCD Response Data Acquisition: Simultaneously with the execution of the drive signal, the system also needs to acquire the response data of the LCD screen. The response data includes the following key indicators:

[0181] Response delay: Represents the time required from the input of the driving signal to the pixel reaching the target state.

[0182] Brightness variation: Monitors the brightness variation of each pixel to ensure display accuracy and consistency.

[0183] Pixel stability: Reflects the stability of each pixel under long-term driving, preventing distortion or color change due to long-term operation.

[0184] Response data R d (x,y,t) can be represented by the following formula:

[0185] R d(x,y,t)=[ΔL(x,y,t),Δτ(x,y,t),σ(x,y,t)];

[0186] Where: ΔL(x,y,t) is the brightness change of pixel (x,y) at time t, Δτ(x,y,t) is the response delay, and σ(x,y,t) is the standard deviation of pixel stability.

[0187] Data Feedback and Status Update: The collected LCD response data will be fed back to the subsequent feedback module to update the health status map of the LCD screen and the parameters of the neural network. This process will further enhance the system's adaptive adjustment capability.

[0188] The health status map H(x,y) will be updated using the following formula:

[0189] H(x,y)=H previous (x,y)+α·(ΔL(x,y)+Δτ(x,y));

[0190] Where α is the correction coefficient, and ΔL(x,y) and Δτ(x,y) represent the changes in brightness and response delay, respectively. By continuously updating the health status map, the system can correct the fatigue, electromigration, and response deviation of the liquid crystal pixels in real time.

[0191] Neural network parameter fine-tuning: As the health status map is updated, the system will fine-tune the model parameters in the neural network based on this data. Specifically, the feedback liquid crystal response data will be used to retrain or fine-tune the neural network to improve the network's accuracy in predicting liquid crystal response characteristics.

[0192] The fine-tuning process can be represented by the following formula:

[0193]

[0194] in: R represents the loss function; d (x,y) represents the actual response data collected; The response data predicted by the neural network; θ i Here are the neural network parameters; λ is the regularization coefficient; θ previous,i These are the previous network parameters.

[0195] By minimizing the loss function, the neural network gradually improves its prediction accuracy.

[0196] S6. Update the liquid crystal health status map and neural network parameters according to the response data to achieve closed-loop optimization.

[0197] Specifically, in step S5, the system executes the optimal driving signal and collects the response data of the LCD screen. To further optimize the LCD display effect and achieve closed-loop adjustment, step S6 mainly processes this response data, updates the LCD health status map, and fine-tunes the neural network parameters. In this way, the system can dynamically adapt and optimize the display effect based on real-time feedback, ensuring the stability and efficiency of the LCD screen during long-term operation.

[0198] In this embodiment, the response data will be used to establish and update the liquid crystal health status map. The health status map reflects the state of the display screen in different usage cycles, including pixel fatigue, electromigration problems, and response deviations. Updating this map will provide feedback for subsequent drive signal adjustments and optimizations.

[0199] Health status map modeling and updating: In step S5, the system has collected the response data R of each pixel. d (x,y,t), these data are used to reflect the display stability and dynamic changes of pixels. Based on this response data, the system models and updates the health status of each pixel. The health status map H(x,y) will be adjusted based on response latency, brightness changes, and pixel stability, with the specific update formula as follows:

[0200] H(x,y)=H previous (x,y)+α·(ΔL(x,y)+Δτ(x,y));

[0201] Wherein: H previous (x,y) represents the health status of the pixel at the previous time point, ΔL(x,y) and Δτ(x,y) represent the currently acquired brightness change and response delay change, respectively, and α is a correction coefficient used to control the update step size.

[0202] In some embodiments, the health status map can be further optimized by introducing correction terms from the liquid crystal response physical model to estimate the health status of each pixel. For example, considering the effects of temperature variations and long-term electromigration in the liquid crystal panel, the system can use these physical characteristics as additional input features to improve the accuracy of the health status map.

[0203] Neural network parameter fine-tuning: Based on the updated health status map, the system will continue to fine-tune the neural network to improve its adaptability to the liquid crystal response characteristics. By using the health status map as an additional input, the system can finely adjust the parameters of the neural network to better reflect the actual operating status of the current liquid crystal screen.

[0204] The fine-tuning of neural network parameters can be optimized using the following loss function:

[0205]

[0206] in: Let R be the loss function. d (x,y) represents the actual response data collected. For the response data predicted by the neural network, θ i Here are the parameters of the neural network, λ is the regularization coefficient, and θ is the... previous,i These are the parameters from the previous time step. By minimizing the loss function, the neural network can gradually adjust its parameters, making the predicted response more consistent with the actual liquid crystal response data.

[0207] Closed-loop optimization and dynamic adaptation: Through updating the health status map and fine-tuning the neural network parameters, the system achieves closed-loop optimization. Whenever new response data is collected, the system updates the health status map in real time and adjusts the driving parameters based on the updated map. This dynamic adaptation capability ensures that the LCD screen maintains optimal performance under different environmental and usage conditions.

[0208] In some embodiments, the system can further optimize the structure of the neural network based on long-term feedback data. For example, deeper network structures can be used to process visually important regions, or pruning and quantization techniques can be employed to reduce the computational burden on low-visual-interest regions.

[0209] The intelligent driving system for a liquid crystal display screen described below and the intelligent driving method for a liquid crystal display screen described above can be referred to in correspondence.

[0210] Please see the appendix Figure 2 An intelligent driving system for a liquid crystal display screen includes:

[0211] The multimodal information acquisition module is used to acquire environmental information, display content information, and user behavior information, and to construct a fused feature vector.

[0212] The neural network inference module is used to receive fused feature vectors and extract spatial and temporal features, while fusing physical modeling results to generate a visual importance map;

[0213] The area division and refresh control module is used to divide the display area according to the visual importance mapping and content change rate, and to calculate the refresh priority and refresh rate.

[0214] A multi-objective optimization module is used to construct an optimization function and determine the optimal driving parameters based on display quality, power consumption, and response speed.

[0215] The drive execution module is used to generate drive signals according to the optimal drive parameters to control the display behavior of the liquid crystal pixel units;

[0216] The closed-loop feedback module is used to collect liquid crystal response data and update the liquid crystal health status map and neural network model parameters.

[0217] The system in this embodiment can be used to execute the above method embodiments, and its principle and technical effect are similar, so they will not be described again here.

[0218] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart driving method for a liquid crystal display screen, characterized in that, Includes the following steps: S1. Collect environmental information, display content information, and user behavior information through sensors to construct a multimodal fusion feature vector; S2. Based on the fused feature vectors of the multimodal model, spatial and temporal features are extracted using a physical information-enhanced neural network. Specifically, a convolutional neural network is used to extract spatial texture and edge features of the current frame image; a recurrent neural network or Transformer architecture is used to perform temporal modeling on the continuous frame sequence to extract dynamic trends; simultaneously, combining historical voltage-driven data and temperature response curves, a pixel-level dynamic model is constructed based on the liquid crystal torsional torque and the applied electric field; this pixel-level dynamic model is derived from the liquid crystal torsional dynamics equation. Modeling, and through response time empirical models The modeling process incorporates the effects of temperature and voltage on the liquid crystal response time, and then integrates the acquired actual response time for calibration and correction. The modeling results are used as physical auxiliary features to assist the neural network in predicting nonlinear responses. The spatial features, temporal features, and physical auxiliary features are fused together, and a visual importance mapping is generated using the liquid crystal physical response model. Indicates the twist angle of liquid crystal molecules. For the applied electric field strength, Pre-tilt angle, These are the elastic constant and electro-optic coupling constant of the liquid crystal material, respectively. Indicates response time. To display the panel temperature, This is the voltage value. For pixel position or state parameters, For the fitting constant, the function Used to describe the correction of response to spatial non-uniformity; S3. Based on the visual importance mapping and content change rate, divide the display area and calculate the refresh priority and refresh rate of each area; S4. Based on the refresh priority, computing resource status, and the multi-objective optimization model of display quality, power consumption and response speed, the area refresh rate, compensation voltage and pixel status are taken as optimizable variables, and the hardware load, voltage limit and display stability are set as constraints. The optimal parameters are searched by weighted summation method or Pareto front method to determine the optimal driving parameters, and a mapping relationship is established between the response compensation result and the driving signal. S5. Execute the optimal driving parameters to control the pixel driving circuit of the liquid crystal display to output driving signals, control the voltage change of the TFT liquid crystal pixel unit through the voltage driving module, synchronously collect liquid crystal response data including response delay, brightness change and pixel stability in each driving cycle, and transmit the liquid crystal response data to the subsequent feedback module for status update. S6. Update the liquid crystal health status map and neural network parameters based on the response data to achieve closed-loop optimization. Specifically, based on the collected response data, model and correct the current health status of each pixel using the formula... Update the liquid crystal health status map to reflect pixel fatigue, electromigration, and response deviation; among which, This represents the health status of the pixel at the previous time point. and These represent the currently acquired changes in brightness and response delay, respectively. The correction coefficient is used to control the update step size; the updated liquid crystal health status map is used as an additional input to retrain or fine-tune the neural network, adjust the visual importance mapping parameters and physical modeling structure, and realize the closed-loop optimization and dynamic adaptation of the driving system.

2. The intelligent driving method for a liquid crystal display screen according to claim 1, characterized in that, Step S1 includes: Collect environmental information, including ambient light intensity, display panel temperature, and device attitude angle collected by ambient light sensor, temperature sensor, and attitude sensor, respectively. Obtain display content information, wherein the display content information is the image pixel matrix of the current display frame; Collect user behavior information, including the user's gaze area, viewing angle, and dwell time; Feature extraction and normalization are performed on the environmental information, displayed content information, and user behavior information. The normalized features are combined to construct a multimodal fusion feature vector, which serves as the input data for intelligent driving.

3. The intelligent driving method for a liquid crystal display screen according to claim 1, characterized in that, Step S3 includes: The display area will be divided into a fixed grid or an adaptive region based on clustering results; Calculate the mean visual importance, historical frame change rate, and complexity of displayed content for each region. The refresh priority of each region is determined based on a weighted calculation model; Based on refresh priority and system resource limitations, a corresponding refresh rate is allocated to each region; The output area refresh rate is used as one of the inputs for subsequent multi-objective optimization.

4. The intelligent driving method for a liquid crystal display screen according to claim 1, characterized in that, The physical information augmentation neural network adopts a lightweight and variable architecture, supporting the division of model depth and complexity by region, specifically including: Deep network computation is used for visually important regions; For areas of low visual attention, use pruned or quantized smaller networks; The model inference process employs parallel processing using heterogeneous computing units; Improve inference efficiency for duplicate regions through caching mechanisms.

5. An intelligent driving system for a liquid crystal display screen, applied to the intelligent driving method for a liquid crystal display screen according to any one of claims 1-4, characterized in that, include: The multimodal information acquisition module is used to acquire environmental information, display content information, and user behavior information, and to construct a fused feature vector. The neural network inference module is used to receive fused feature vectors and extract spatial and temporal features, while fusing physical modeling results to generate a visual importance map; The area division and refresh control module is used to divide the display area according to the visual importance mapping and content change rate, and to calculate the refresh priority and refresh rate. A multi-objective optimization module is used to construct an optimization function and determine the optimal driving parameters based on display quality, power consumption, and response speed. The drive execution module is used to generate drive signals according to the optimal drive parameters to control the display behavior of the liquid crystal pixel units; The closed-loop feedback module is used to collect LCD response data and update the LCD health status map and neural network model parameters.

Images