Construction method for automatically reasoning illumination parameters based on task scene

Abstract

The invention discloses a construction method for automatically reasoning illumination parameters based on a task scene, and relates to the field of man-machine interaction and display design, and the method comprises the steps: S1, collecting task parameters, S2, carrying out feature extraction and evaluation after the S1, collecting the task parameters, S3, outputting illumination scene parameters after the S2, carrying out feature extraction and evaluation, and S4, outputting the illumination scene parameters. And S4, after the illumination scene parameters are output in the step S3, factor variables are set. According to the man-machine interaction and display design color coding optimization method based on the ambient light condition, a confidence evaluation and double-path decision mechanism is innovatively introduced, the problem that a single model is insufficient in reliability in a complex and changeable environment is effectively solved, the system can intelligently judge the credibility of sensor data, and the reliability of the sensor data is improved. When the confidence is high, the precision is improved by adopting a fine fusion result, and when the confidence is low, the reliability is guaranteed by starting a preset safety rule, so that the environment adaptability and the operation safety are both realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of human-computer interaction and display design, and particularly relates to a construction method for automatically inferring illumination parameters based on a task scene. BACKGROUND

[0002] In complex application environments such as aviation, aerospace, and vehicle driving, interface color coding is an important way to convey key information.

[0003] In the prior art, different environmental illumination conditions (such as day-night changes, color temperature differences, and brightness fluctuations) can significantly affect the recognizability and recognition efficiency of interface color coding.

[0004] In order to overcome the above-mentioned deficiencies, the prior art (publication number CN107622502B) discloses a path extraction and recognition method for a visual guidance system under complex illumination conditions. First, the relationship between illumination and image brightness components is analyzed to establish an illumination color model representing the color distribution of the image. Second, an image illumination classifier is designed to distinguish between high-light areas, normal-illumination areas, and dark-shadow areas in the complex-illumination path image. Image enhancement is performed on the low-illumination area in the RGB color space to restore the path color information. In the high-light area, the chroma components Cb and Cr are subjected to difference operation to suppress common-mode illumination interference, and then adaptive threshold segmentation is performed. Finally, the particle swarm optimization method is used to identify the optimal parameter model of the guidance path, which can significantly improve the accuracy, reliability, and intelligence of the visual guidance system in extracting and recognizing the guidance path under complex illumination.

[0005] In order to overcome the above-mentioned deficiencies, the prior art (publication number CN119624840A) discloses a gamma correction low-illumination image enhancement method based on convex optimization. First, the RGB color model of the original low-illumination image is converted to the YCbCr color model. Then, the brightness component Y of the YCbCr color model is normalized using a logarithmic function. After dividing the brightness image into dark and bright regions, the optimal gamma correction parameters for each region are estimated independently based on the convex optimization method. The enhanced dark and bright regions are combined with the optimal gamma correction parameters to obtain the enhanced image. Finally, the YCbCr color model after image enhancement is converted back to the RGB color model, solving the problem of detection performance being affected by methods such as target detection and target tracking in low-illumination environments such as night, overcast, and storm.

[0006] Although the prior art can overcome the above-mentioned deficiencies, there are still other problems in its operation process, such as: the existing color coding optimization methods are mostly based on color visibility experiments under static conditions, lacking adaptive adjustment mechanisms for dynamic illumination changes, lacking dynamic illumination adaptive adjustment mechanisms, and being unable to adjust in real time in combination with individual visual differences. SUMMARY

[0007] The present application aims to provide a construction method for automatically inferring lighting parameters based on task scenarios, to solve the problems of the prior art color coding optimization method based on static conditions for color visibility experiments, lack of adaptive adjustment mechanism for dynamic lighting changes, lack of dynamic lighting adaptive adjustment mechanism, and inability to combine individual visual differences for real-time adjustment.

[0008] To achieve the above-mentioned purpose, the present application provides the following technical scheme: a construction method for automatically inferring lighting parameters based on task scenarios, specifically comprising the following steps: S1, collecting task parameters; S2, feature extraction and evaluation; S3, outputting lighting scene parameters; S4, setting factor variables; S5, dynamic task simulation process; S6, associated storage; S7, multi-modal feature fusion; S8, prediction sub-model; S9, performance replay simulation and evaluation.

[0009] First, S1, collect task parameters, after S1, collect task parameters, S2, feature extraction and evaluation, after S2, feature extraction and evaluation, S3, output lighting scene parameters, after S3, output lighting scene parameters, S4, set factor variables, after S4, set factor variables, S5, dynamic task simulation process, after S5, dynamic task simulation process, S6, associated storage, after S6, associated storage, S7, multi-modal feature fusion, then S8, prediction sub-model, after S8, prediction sub-model, finally S9, performance replay simulation and evaluation.

[0010] Further, the S1, collecting task parameters, is to analyze the flight task scenario, obtain the task characteristic parameter set including flight altitude, UTC time, weather code and geographic location information, and call real-time environmental information based on the geographic location information, including satellite remote sensing data, weather radar data and task scene images collected by onboard cameras.

[0011] Further, the S2, feature extraction and evaluation, is to extract features from the task characteristic parameters, real-time environmental information and task scene images to obtain a multi-source feature vector, the multi-source feature vector is , the formula is: , wherein is the normalized numerical feature vector, is the image feature vector, the feature extraction of the task characteristic parameters, real-time environmental information and task scene images specifically includes: using a pre-trained convolutional neural network to extract deep visual features related to lighting from the task scene images, including image overall brightness distribution, shadow area proportion, highlight area saturation, to generate an image feature vector and minimum to maximum normalization is performed: ; at the same time, numerical features such as cloud optical thickness and cloud top height are extracted from satellite remote sensing data, and precipitation particle intensity distribution features are extracted from weather radar data to generate a numerical feature vector , and Z-Score normalization is performed: ; the multi-source feature vector is a comprehensive vector obtained by splicing the above visual features and numerical features after standardization , , and the confidence of the multi-source feature vector is evaluated according to the task scene, and the comprehensive confidence C is calculated, and the comprehensive confidence , wherein is a weight coefficient, is an illumination confidence calculated based on the standard deviation of cloud optical thickness and the image definition score, is a color temperature confidence calculated based on the estimated value of the weather radar color temperature and the average of the image color temperature distribution, and the confidence of the multi-source feature vector includes: calculating the standard deviation of the cloud optical thickness feature value from the satellite remote sensing data ; calculating the definition score of the spectral feature vector from the onboard camera image , based on the standard deviation and the definition score, the illumination confidence and the color temperature confidence are calculated by weighted fusion, and the comprehensive confidence C is obtained by .

[0012] Further, the S3, the output illumination scene parameter is operated according to the data, the comprehensive confidence C does not exceed the preset threshold, then the multi-source feature vector is input to the feature fusion network, the feature fusion network includes an image feature extraction subnetwork and a numerical feature extraction subnetwork, and the illumination scene parameter is output after fusion, including the illumination level and the color temperature level, and the steps are as follows:

[0013] First, the multi-head attention mechanism is fused to generate a fusion feature vector , Q, K, and V are query, key, and value matrices, which are obtained by transforming V final , second, the fusion feature vector is mapped through a fully connected layer, and the continuous illumination scene parameter X=W·F fused +b is output, including illumination and color temperature, wherein W∈R 1×512 , b∈R is the parameter of the fully connected layer, finally, in order to avoid numerical mutation, the output is smoothed and scaled by the Sigmoid function, and the final parameter is generated, and the preset output range of the illumination and color temperature is limited,

[0014] ;

[0015] ;

[0016] where β ∈ (0, 1) is a smoothing factor, according to the cloud thickness dynamically adjust,

[0017]

[0018] [L min , L max ] and [T min , T max ] are the preset output ranges of illuminance and color temperature, respectively, and when the comprehensive confidence C exceeds the threshold, the target illumination parameter set is generated by comparing with the preset task type-illumination parameter mapping table. The mapping table is established by analyzing historical task data and measured illumination data, and the high-altitude cruising, low-altitude penetration, and night landing task types are associated with typical illumination parameters.

[0019] Further, the S4, the setting factor variable is the setting environment illumination factor variable, including the environment illuminance level and the environment color temperature, the environment light condition includes five typical combinations of morning cloudy, noon cloudless, noon cloudy, dusk, and night, and the color coding factor variable is set according to the flight task characteristics, including the color type and the color brightness, the color type is red, green, and yellow, and the brightness level is set to 30, 60, and 90.

[0020] Further, the S5, the dynamic task simulation process is to construct a dynamic task simulation process, which includes a chain process composed of a target identification stage, a confirmation stage, and a decision stage executed in sequence. The target identification stage displays the aviation target icons and interference symbols to be identified. The confirmation stage requires determining whether the number of target symbols reaches a threshold. The decision stage requires responding under time pressure, and a timeout is considered an error. During the running of the dynamic task simulation process, the behavior indicators and physiological indicators of the subject are collected in real time. The behavior indicators include task accuracy and reaction time, and the physiological indicators include pupil diameter and blink frequency.

[0021] Further, the S6, the correlation storage is the behavior indicators or physiological indicators deviate from the preset performance threshold , i.e., continuously , the experimental system dynamically adjusts the color coding parameters or the task interference level according to the current state of the subject to guide the subject's state back to the target interval. The adjustment formula is: where P t is the current color coding parameter, Δ is the step size, E t , E target are the current and target performance indicators, respectively, and the preset performance threshold is calculated from the dynamic baseline values of reaction time and pupil diameter: where the dynamic baseline value is calculated by a preset number of times​ average performance indicator of each trial, is a proportionality coefficient, and the color coding parameters adjusted dynamically under each lighting condition are stored in association with the corresponding performance indicators for subsequent generation of color coding optimization suggestions. The dynamic adjustment of the color coding parameters follows the principle of step method, specifically: if the pupil diameter of the subject continuously exceeds the first threshold value or the reaction time continuously exceeds the second threshold value, the system increases the color brightness by a fixed step, and immediately retests after adjustment until the physiological indicators return to the pre-set target performance interval. The adjustment formula is: wherein P t is the current color coding parameter, Δ is the step, E t , E target are the current and target performance indicators, respectively.

[0022] Further, the S7, multi-modal feature fusion is a multi-modal feature fusion of lighting scene parameters, flight task characteristics, behavior indicators and physiological indicators collected in experiments, to obtain a comprehensive feature vector wherein Q com , K com , V com is calculated based on lighting, task and behavior / physiological feature vectors. wherein Q com , K com , V com is calculated based on lighting, task and behavior / physiological feature vectors.

[0023] Further, the S8, prediction sub-model is to input the comprehensive feature vector into an adaptive optimization module, which includes a prediction sub-model and a verification sub-model. The prediction sub-model generates a candidate color coding parameter set under the lighting condition based on a deep neural network. The deep neural network used in the prediction sub-model is an encoder-decoder structure based on Transformer. The encoder is used to encode the input comprehensive feature vector, and the decoder generates a candidate color coding parameter sequence in a self-recursive manner. The adaptive optimization module divides the experimental data set into a training set, a validation set and a test set to ensure the generalization ability of the model and prevent overfitting. The prediction sub-model and the verification sub-model learn on the training set, optimize the hyperparameters through the validation set and control early stopping, and finally evaluate the performance on a completely independent test set to ensure the reliability of the obtained dynamic optimization mapping table.

[0024] Further, the S9, performance replay simulation and evaluation is to verify the sub-model in the simulation environment to perform performance replay simulation on the candidate color coding parameters, evaluate the task accuracy A, reaction time RT and physiological load D, and screen out the target color coding parameters with optimal performance and high stability , wherein P i is a candidate color coding parameter, P opt is a target parameter, and the comprehensive performance score is obtained by normalized weighted fusion of performance indicators, = , the comprehensive stability is obtained by normalized inverse transformation of the minimum value of the performance indicator variance in repeated simulation, , the target color coding parameter with optimal performance and high stability is obtained in a high-fidelity virtual reality flight simulation environment, different light conditions are accurately reproduced, and candidate color parameters are applied to render interface symbols, through built-in computer vision models and human performance prediction algorithms, the target recognition efficiency, misreading risk and visual comfort are quantitatively evaluated, the target color coding parameter with optimal performance and high stability is screened out, first, the comprehensive performance score is calculated for each candidate color parameter scheme j , wherein A, RT and D are performance indicators predicted by the verification sub-model, A is the accuracy, RT is the reaction time, D is the physiological load, and α, β and γ are weight coefficients, and α+β+γ=1, the values can be adjusted according to the type of task; secondly, for the scheme with a comprehensive score S greater than the qualified line S_threshold, random disturbance (such as simulation of user differences) is introduced in the simulation environment to perform N times of performance replay simulation, and the comprehensive stability score is calculated = = =1

[0025] ,

[0026] , wherein is the accuracy standard deviation, is the reaction time variance, is the physiological load fluctuation, and finally is the final output parameter, wherein P i is a candidate color coding parameter, and P opt is a target parameter

[0027] The target color coding parameter and corresponding light condition are recorded in a dynamic optimization mapping table, which can be used for color coding design of a flight display interface or real-time adaptive color adjustment, the dynamic optimization mapping table has a multi-dimensional lookup table as a data structure, a combination of environment illumination value and color temperature value is used as a primary key, and each record stores the following contents: optimized color coding parameter, and corresponding performance indicators such as predicted task accuracy and average reaction time, and the design enables the system to obtain color adjustment instructions through fast lookup instead of complex calculation in real-time application, and meets the high real-time requirement.

[0028] Compared with the prior art, the present application has the following beneficial effects:

[0029] 1. The confidence evaluation and double-path decision mechanism are innovatively introduced, the reliability problem of a single model in a complex and variable environment is effectively solved, the system can intelligently judge the reliability of sensor data, the fine fusion result is used in high confidence to improve the precision, and the preset safety rule is used in low confidence to ensure the reliability, and the environmental adaptability and operation safety are combined.

[0030] 2. A complete technical closed loop from environment perception to physiological feedback is constructed, objective environment light measurement and subjective user physiological indicators are combined, the optimal color parameters under different illuminations are quantified through dynamic ladder experiments, so that the color optimization strategy has both theoretical basis and individual adaptability.

[0031] 3. The information perception efficiency and safety of a flight display interface are greatly improved, through the present application, the display symbol can still maintain the best visibility and recognition under extreme light conditions such as day-night conversion and meteorological mutation, and the cognitive load of a pilot is maintained to be low, and it is of great significance to aviation safety. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 It is a whole process schematic diagram of the present application.

[0033] Figure 2 It is an experimental system architecture diagram of the present application.

[0034] Figure 3 It is a verification experiment task flowchart of the present application.

[0035] Figure 4 It is a three-node brightness statistical schematic diagram of experimental accuracy data of the present application.

[0036] Figure 5 It is a three-node brightness statistical schematic diagram of experimental reaction time data of the present application.

[0037] Figure 6 It is a three-node brightness statistical schematic diagram of experimental pupil diameter data of the present application.

[0038] Figure 7 The schematic diagram of the selected aircraft symbol material for the experiment of the present application. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work belong to the scope of protection of the present application.

[0040] Embodiment one: as shown in the technical solutions in the technical solutions shown in Figure 1 and Figure 2 , a construction method for automatically inferring illumination parameters based on task scenarios discloses: S1, collecting task parameters; S2, feature extraction and evaluation; S3, outputting illumination scene parameters.

[0041] Firstly, S1, collecting task parameters is performed, and after S1, collecting task parameters is completed, S2, feature extraction and evaluation is performed, and after S2, feature extraction and evaluation is completed, S3, outputting illumination scene parameters is performed.

[0042] S1, collecting task parameters is to analyze the flight task scene, obtain a task feature parameter set including flight altitude, UTC time, weather code and geographic location information, and call real-time environment information based on the geographic location information, including satellite remote sensing data, weather radar data and task scene images collected by an onboard camera.

[0043] S2, feature extraction and evaluation is to perform feature extraction on the task feature parameters, real-time environment information and task scene images to obtain a multi-source feature vector, and the multi-source feature vector is , and the formula is: , wherein is a normalized numerical feature vector, is an image feature vector, and the feature extraction on the task feature parameters, real-time environment information and task scene images specifically includes: using a pre-trained convolutional neural network to extract deep visual features related to illumination from the task scene images, including image overall brightness distribution, shadow area proportion and highlight area saturation, to generate an image feature vector , and performing minimum-to-maximum normalization: ; at the same time, extracting cloud optical thickness and cloud top height from the satellite remote sensing data, and extracting precipitation particle intensity distribution characteristics from the weather radar data to generate a numerical feature vector , and performing Z-Score normalization: ; the multi-source feature vector is a comprehensive vector obtained by splicing the above-mentioned visual features and numerical features after standardization , , and the confidence of the multi-source feature vector is evaluated according to the task scene, and a comprehensive confidence C is calculated, the comprehensive confidence , wherein is a weight coefficient, is an illumination confidence calculated based on the standard deviation of the cloud optical thickness and the image definition score, is a color temperature confidence calculated based on the meteorological radar color temperature estimation value and the image color temperature distribution mean value, the evaluation of the confidence of the multi-source feature vector includes: calculating the standard deviation of the cloud optical thickness feature value from the satellite remote sensing data ; calculating the definition score of the spectral feature vector from the airborne camera image , based on the standard deviation and the definition score, the illumination confidence and the color temperature confidence are calculated by weighted fusion, and the comprehensive confidence C is obtained by .

[0044] S3, the output light scene parameter is operated according to the data, and the comprehensive confidence C does not exceed the preset threshold, then the multi-source feature vector is input to the feature fusion network, the feature fusion network includes an image feature extraction subnetwork and a numerical feature extraction subnetwork, and the light scene parameter is output after fusion, including the illumination level and the color temperature level, the steps are as follows:

[0045] Firstly, the multi-head attention mechanism is fused to generate a fusion feature vector , Q, K, and V are query, key, and value matrices, which are obtained by transforming V final , secondly, the fusion feature vector is mapped through a full connection layer, and the continuous light scene parameter X=W·F fused +b is output, including illumination and color temperature, wherein W∈R 1×512 , b∈R is the full connection layer parameter, finally, in order to avoid numerical mutation, the output is smoothed and scaled by a Sigmoid function, the final parameter is generated, and the preset output range of the illumination and the color temperature is limited,

[0046] ;

[0047] ;

[0048] , wherein β∈(0, 1) is a smoothing factor, which is dynamically adjusted according to the cloud thickness ,

[0049] ),

[0050] [L minL max ] and [T min T max The preset output ranges for illuminance and color temperature are respectively. When the comprehensive confidence level C exceeds the threshold, it is compared with the preset task type-illuminance parameter mapping table to generate the target illuminance parameter set. The mapping table is established by analyzing historical task data and measured illuminance data, and associates task types such as high-altitude cruise, low-altitude penetration, and night landing with typical illuminance parameters.

[0051] Example 2: Figures 1-6 The technical solution shown, based on Embodiment 1, discloses: S4, setting factor variables; S5, dynamic task simulation process; S6, associated storage;

[0052] S3. After outputting the lighting scene parameters, proceed to S4. Set factor variables. After S4. Set factor variables, proceed to S5. Dynamic task simulation process. After S5. Dynamic task simulation process, proceed to S6. Associated storage.

[0053] S4. Setting factor variables involves setting ambient light factor variables, including ambient illuminance level and ambient color temperature. Ambient light conditions include five typical combinations: cloudy in the early morning, cloudless at noon, cloudy at noon, dusk, and night. Color coding factor variables are also set according to the characteristics of the flight mission, including color type and color brightness. The color types are red, green, and yellow, and the brightness levels are set to 30, 60, and 90.

[0054] S5. The dynamic task simulation process is a chained process consisting of a target identification stage, a confirmation stage, and a decision-making stage executed sequentially. The target identification stage displays the aviation target icon to be identified and interference symbols. The confirmation stage requires determining whether the number of target symbols has reached a threshold. The decision-making stage requires a response under time pressure, and exceeding the time limit is considered an error. During the operation of the dynamic task simulation process, the subject's behavioral and physiological indicators are collected in real time. The behavioral indicators include task accuracy and reaction time, and the physiological indicators include pupil diameter and blink frequency.

[0055] S6. Associative storage is a behavioral or physiological indicator. Deviation from preset performance threshold Time, i.e., continuous The experimental system dynamically adjusts the color coding parameters or task interference level based on the subject's current state to guide the subject's state back to the target range. The adjustment formula is as follows: Among them, P t E represents the current color encoding parameters, Δ is the step size, and E is the value of E. t E target These are the current and target performance indicators, respectively. The preset performance threshold is calculated from the dynamic baseline value of the pupil diameter during reaction time. Among them, dynamic baseline value By preset number of times Calculation of average performance indicators for each trial. The proportional coefficient, and the dynamically adjusted color coding parameters under various lighting conditions, are associated with and stored with the corresponding performance indicators for subsequent generation of color coding optimization suggestions. The dynamic adjustment of color coding parameters follows the stepwise method principle, specifically: if the subject's pupil diameter is consistently greater than the first threshold or reaction time is consistently greater than the second threshold, the system increases the color brightness by a fixed step size and immediately retests after adjustment until the physiological indicators return to the preset target performance range. The adjustment formula is: Among them, P t E represents the current color encoding parameters, Δ is the step size, and E is the value of E. t E target These are the current and target performance metrics, respectively.

[0056] Example 3: Figures 1-6 The technical solution shown, based on Embodiment 2, discloses: S7, multimodal feature fusion; S8, prediction sub-model; S9, performance replay simulation and evaluation;

[0057] After S6 and associated storage are completed, proceed to S7 and multimodal feature fusion. Then proceed to S8 and prediction sub-model. After S8 and prediction sub-model are completed, finally proceed to S9 and performance replay simulation and evaluation.

[0058] S7. Multimodal feature fusion involves fusing lighting scene parameters, flight mission characteristics, and experimentally collected behavioral and physiological indicators to obtain a comprehensive feature vector. Q com K com V com Based on the computation of illumination, task, and behavioral / physiological feature vectors, multimodal feature fusion is performed. An attention-weighted feature fusion method is employed, which dynamically calculates and assigns the contribution weights of features from different sources to the final decision. This highlights key features, suppresses noise interference, and generates a more comprehensive feature vector with stronger representational capabilities during the fusion process. Q com K com V com Calculated based on illumination, task, and behavioral / physiological feature vectors.

[0059] S8, the prediction sub-model is input into the adaptive optimization module, the module includes the prediction sub-model and the verification sub-model, the prediction sub-model generates the candidate color coding parameter set under the light condition based on the deep neural network, the prediction sub-model adopts the encoder-decoder structure based on the Transformer, the encoder is used to encode the input comprehensive feature vector, and the decoder generates the candidate color coding parameter sequence in an autoregressive manner, the adaptive optimization module is divided into a training set, a validation set and a test set to ensure the generalization ability of the model and prevent overfitting, the prediction sub-model and the verification sub-model learn on the training set, the hyperparameters are optimized and early stopping control is performed through the validation set, and finally the performance is evaluated on a completely independent test set to ensure the reliability of the obtained dynamic optimization mapping table.

[0060] S9, the performance replay simulation and evaluation is that the verification sub-model performs performance replay simulation on the candidate color coding parameters in a simulation environment, evaluates the task accuracy A, reaction time RT and physiological load D, and selects the target color coding parameter with optimal performance and high stability , wherein P i is the candidate color coding parameter, P opt is the target parameter, and the comprehensive performance score is obtained by normalizing and weighting the performance indicators, = , and the comprehensive stability is obtained by normalizing and inversely transforming the minimum value of the performance indicator variance in repeated simulation, , the target color coding parameter with optimal performance and high stability is obtained in a high-fidelity virtual reality flight simulation environment, different light conditions are accurately reproduced, candidate color parameters are applied to render interface symbols, the built-in computer vision model and human performance prediction algorithm are used to quantitatively evaluate target recognition efficiency, misreading risk and visual comfort, and the target color coding parameter with optimal performance and high stability is selected, first, the comprehensive performance score of each candidate color parameter scheme j is calculated , wherein A, RT and D are the performance indicators predicted by the verification sub-model, A is the accuracy, RT is the reaction time, D is the physiological load, α, β and γ are weight coefficients, and α+β+γ=1, the values of which can be adjusted according to the type of task; secondly, for the scheme with a comprehensive score S greater than the threshold S_threshold, random disturbance (such as simulating user differences) is introduced in the simulation environment to repeatedly perform N times of performance replay simulation, and the comprehensive stability score of the scheme is calculated = = =1.

[0061] ,

[0062] wherein is the standard deviation of accuracy, is the variance of reaction time, is the fluctuation of physiological load, and finally screening out is the final output parameter, wherein P i is the candidate color encoding parameter, P opt is the target parameter;

[0063] record the target color encoding parameter and the corresponding lighting condition in a dynamic optimization mapping table, which can be used for color encoding design of the flight display interface or real-time adaptive color adjustment, the dynamic optimization mapping table has a multi-dimensional lookup table data structure, the primary key is the combination of the environmental illuminance value and the color temperature value, and each record stores the following contents: the optimized color encoding parameter, and the corresponding performance indicators such as the predicted task accuracy and the average reaction time, and the design enables the system to obtain the color adjustment instruction through fast lookup table instead of complex calculation in real-time application, thereby meeting the high real-time requirement.

[0064] Although the embodiments of the present application have been shown and described, it is to be understood that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for automatically inferring lighting parameters based on a task scenario, characterized in that, Specifically, the following steps are included: S1. Collect task parameters; S2. Feature extraction and evaluation; S3. Output lighting scene parameters; S4. Set factor variables; S5. Dynamic task simulation process; S6, Associated Storage; S7, Multimodal Feature Fusion; S8, Predictive Sub-model; S9, Performance Replay Simulation and Evaluation; First, S1 is performed to collect task parameters. After S1 is completed, S2 is performed to extract and evaluate features. After S2 is completed, S3 is performed to output lighting scene parameters. After S3 is completed, S4 is performed to set factor variables. After S4 is completed, S5 is performed to simulate the dynamic task process. After S5 is completed, S6 is performed to associate and store data. After S6 is completed, S7 is performed to fuse multimodal features. Then, S8 is performed to predict the sub-model. After S8 is completed, S9 is performed to simulate and evaluate the performance.

2. The method for constructing lighting parameters based on automatic inference of task scenarios according to claim 1, characterized in that: S1, collecting task parameters, involves parsing the flight mission scenario to obtain a set of task feature parameters, including flight altitude, UTC time, meteorological code, and geographical location information. Based on the geographical location information, real-time environmental information is retrieved, including satellite remote sensing data, meteorological radar data, and mission scenario images collected by airborne cameras.

3. The method for constructing lighting parameters based on automatic inference of task scenarios according to claim 2, characterized in that: S2, feature extraction and evaluation, involves extracting features from task feature parameters, real-time environmental information, and task scene images to obtain multi-source feature vectors. These multi-source feature vectors are... The formula is: ,in The normalized numerical eigenvectors To generate an image feature vector, feature extraction is performed on task feature parameters, real-time environmental information, and task scene images. Specifically, this includes using a pre-trained convolutional neural network to extract illumination-related deep visual features from the task scene images, including overall image brightness distribution, shadow area proportion, and highlight area saturation, to generate an image feature vector. And perform minimum-to-maximum normalization: Simultaneously, numerical features of cloud optical thickness and cloud top height are extracted from satellite remote sensing data, and precipitation particle intensity distribution features are extracted from meteorological radar data to generate numerical feature vectors. And perform Z-Score normalization: The multi-source feature vector is a composite vector formed by concatenating the aforementioned visual and numerical features after standardization. , Furthermore, based on the task scenario, the confidence level of the multi-source feature vectors is evaluated, and its comprehensive confidence level C is calculated. ,in These are the weighting coefficients. The illuminance confidence level is calculated based on the standard deviation of cloud optical thickness and image sharpness score. To assess the confidence level of the color temperature calculated based on meteorological radar color temperature estimates and the mean of image color temperature distribution, the evaluation of the multi-source feature vector confidence level includes: calculating the standard deviation of cloud optical thickness feature values ​​from satellite remote sensing data. ; Calculate the sharpness score from the spectral feature vectors of images from the airborne camera. Based on the standard deviation and sharpness score, the illuminance confidence level is calculated through weighted fusion. With color temperature confidence ,pass The overall confidence level C is obtained.

4. The method for constructing lighting parameters based on automatic inference of task scenarios according to claim 3, characterized in that: S3, outputting the lighting scene parameters, is based on data operations. If the overall confidence level C does not exceed a preset threshold, the multi-source feature vector is input into the feature fusion network. The feature fusion network includes an image feature extraction subnetwork and a numerical feature extraction subnetwork. After fusion, the lighting scene parameters are output, including illuminance level and color temperature level. The steps are as follows: First, a multi-head attention mechanism is used to fuse the features and generate a fused feature vector. Q, K, and V are query, key, and value matrices, respectively, derived from V. final The transformed feature vector is then mapped through a fully connected layer to output continuous lighting scene parameters X = W·F. fused +b, including illuminance and color temperature, where W∈R 1×512 Here, b∈R represents the parameters of the fully connected layer. Finally, to avoid abrupt changes in values, the output is smoothed and scaled using the sigmoid function to generate the final parameters, and the preset output ranges for illuminance and color temperature are defined. ； ； Where β∈(0,1) is the smoothing factor, which is determined based on cloud thickness. Dynamic adjustment based on changes ））， [L min L max ] and [T min T max The preset output ranges for illuminance and color temperature are respectively. When the comprehensive confidence level C exceeds the threshold, it is compared with the preset task type-illuminance parameter mapping table to generate the target illuminance parameter set. The mapping table is established by analyzing historical task data and measured illuminance data, and associates high-altitude cruise, low-altitude penetration, and night landing task types with typical illuminance parameters.

5. The method for constructing lighting parameters based on automatic inference of task scenarios according to claim 4, characterized in that: The S4 setting factor variables is to set ambient light factor variables, including ambient illuminance level and ambient color temperature. Ambient light conditions include five typical combinations: morning with clouds, noon without clouds, noon with clouds, dusk, and night. In addition, according to the characteristics of the flight mission, color coding factor variables are set, including color type and color brightness. The color types are red, green, and yellow, and the brightness levels are set to 30, 60, and 90.

6. The method for constructing lighting parameters based on automatic inference of task scenarios according to claim 5, characterized in that: The S5 dynamic task simulation process is a chain process consisting of a target identification stage, a confirmation stage, and a decision stage executed sequentially. The target identification stage displays the aviation target icon to be identified and interference symbols; the confirmation stage requires determining whether the number of target symbols has reached a threshold. The decision-making phase requires a response under time pressure, and exceeding the time limit is considered an error. During the dynamic task simulation process, the subjects' behavioral and physiological indicators are collected in real time. Behavioral indicators include task accuracy and reaction time, while physiological indicators include pupil diameter and blink frequency.

7. The method for constructing lighting parameters based on automatic inference of task scenarios according to claim 6, characterized in that: S6, the associated storage, is a behavioral indicator or physiological indicator. Deviation from preset performance threshold Time, i.e., continuous The experimental system dynamically adjusts the color coding parameters or task interference level based on the subject's current state to guide the subject's state back to the target range. The adjustment formula is as follows: Among them, P t E represents the current color encoding parameters, Δ is the step size, and E is the value of E. t E target These are the current and target performance indicators, respectively. The preset performance threshold is calculated from the dynamic baseline value of the pupil diameter during reaction time. Among them, dynamic baseline value By preset number of times Calculation of average performance indicators for each trial. The proportional coefficient, and the dynamically adjusted color coding parameters under various lighting conditions, are associated with and stored with the corresponding performance indicators for subsequent generation of color coding optimization suggestions. The dynamic adjustment of color coding parameters follows the stepwise method principle, specifically: if the subject's pupil diameter is consistently greater than the first threshold or reaction time is consistently greater than the second threshold, the system increases the color brightness by a fixed step size and immediately retests after adjustment until the physiological indicators return to the preset target performance range. The adjustment formula is: Among them, P t E represents the current color encoding parameters, Δ is the step size, and E is the value of E. t E target These are the current and target performance metrics, respectively.

8. The method for constructing lighting parameters based on automatic inference of task scenarios according to claim 7, characterized in that: The S7, multimodal feature fusion, involves fusing lighting scene parameters, flight mission characteristics, and experimentally collected behavioral and physiological indicators to obtain a comprehensive feature vector. Q com K com V com Based on the computation of illumination, task, and behavioral / physiological feature vectors, multimodal feature fusion is performed. An attention-weighted feature fusion method is employed, which dynamically calculates and assigns the contribution weights of features from different sources to the final decision. This highlights key features, suppresses noise interference, and generates a more powerful comprehensive feature vector during the fusion process. Q com K com V com Calculated based on illumination, task, and behavioral / physiological feature vectors.

9. The method for constructing lighting parameters based on automatic inference of task scenarios according to claim 8, characterized in that: The S8 prediction sub-model inputs the comprehensive feature vector into the adaptive optimization module. This module includes a prediction sub-model and a validation sub-model. The prediction sub-model generates a set of candidate color coding parameters under the given lighting conditions based on a deep neural network. The prediction sub-model uses a Transformer-based encoder-decoder structure. The encoder encodes the input comprehensive feature vector, while the decoder generates a sequence of candidate color coding parameters in an autoregressive manner. To ensure the model's generalization ability and prevent overfitting, the adaptive optimization module divides the experimental dataset into a training set, a validation set, and a test set. The prediction sub-model and the validation sub-model learn on the training set, perform hyperparameter tuning and early stopping control on the validation set, and finally evaluate performance on a completely independent test set to ensure the reliability of the resulting dynamic optimization mapping table.

10. The method for constructing lighting parameters based on automatic inference of task scenarios according to claim 9, characterized in that: S9, performance replay simulation and evaluation, verifies the sub-model's performance replay simulation in a simulation environment for the candidate color coding parameters, evaluates task accuracy (A), reaction time (RT), and physiological load (D), and selects the target color coding parameters with optimal performance and high stability. , where P i P is the candidate color coding parameter. opt The target parameter is the overall performance score. It is derived from the normalized weighted fusion of performance indicators. = Overall stability The value is obtained by minimizing the normalized inverse transform of the variance of the performance index in repeated simulations. The optimal and stable target color coding parameters were determined in a high-fidelity virtual reality flight simulation environment, accurately reproducing different lighting conditions. Candidate color parameters were applied to render interface symbols. Through a built-in computer vision model and human performance prediction algorithm, target recognition efficiency, misreading risk, and visual comfort were quantitatively evaluated to select the optimal and stable target color coding parameters. First, a comprehensive performance score was calculated for each candidate color parameter scheme j. Where A, RT, and D are the performance indicators predicted by the validation sub-model, A is the accuracy rate, RT is the reaction time, D is the physiological load, and α, β, and γ are weighting coefficients, with α+β+γ=1, whose values ​​can be adjusted according to the task type; secondly, for schemes with a comprehensive score S greater than the passing line S_threshold, random disturbances (such as simulating user differences) are introduced into the simulation environment to repeat the performance replay simulation N times, setting... = = =1 Calculate its overall stability score ,in The standard deviation of the accuracy rate This represents the variance in reaction time. To account for fluctuations in physiological load, the final selection was made... For the final output parameters, where P i P is the candidate color coding parameter. opt For target parameters; The target color coding parameters and corresponding lighting conditions are recorded in a dynamic optimization mapping table. This mapping table can be used for color coding design and real-time adaptive color adjustment of the flight display interface. The dynamic optimization mapping table has a multi-dimensional lookup table as its data structure. The primary key is a combination of ambient illuminance and color temperature values. Each record stores the following: the optimized color coding parameters, the corresponding prediction task accuracy, and the average reaction time performance index. This design allows the system to obtain color adjustment instructions through fast table lookup rather than complex calculations during real-time applications, meeting high real-time requirements.

Images