Visual fatigue detection method based on paper whiteness

Abstract

The invention discloses an asthenopia detection method based on paper whiteness, and the method comprises the steps: S101, obtaining an eye time series data set which is multi-modal data acquired based on a preset paper whiteness parameter; s102, performing asthenopia analysis on the eye time sequence data set based on a pre-trained machine learning model to obtain asthenopia data; the asthenopia data is used for quantifying the asthenopia degree under the stimulation of specific whiteness; s103, correlation analysis is conducted according to the multiple sets of asthenopia data and the corresponding paper whiteness parameters, a whiteness-asthenopia trend model is obtained, and the whiteness-asthenopia trend model is used for representing the nonlinear relation between the paper whiteness and the asthenopia degree. By adopting the method, the degree of association between the optical characteristics of the paper and the asthenopia can be quantified, and accurate asthenopia monitoring in a paper reading scene is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of visual fatigue detection, and particularly relates to a visual fatigue detection method based on paper whiteness. Background Art

[0002] Existing visual fatigue detection methods usually detect visual fatigue by judging rubbing-eye behavior through behavioral characteristics, which is easily interfered by the habitual actions of the tested users, resulting in a high misjudgment rate. Moreover, the existing technology usually ignores the influence of paper whiteness on the ciliary muscle regulation frequency, resulting in a large deviation in the detection results in actual applications and unable to meet the requirements of human eye comfort in the paper reading scenario.

[0003] Therefore, there is an urgent need for a detection method that can quantify the correlation between paper optical properties and visual fatigue to achieve accurate visual fatigue monitoring in the paper reading scenario. Summary of the Invention

[0004] Based on this, in view of the above technical problems, it is necessary to provide a visual fatigue detection method based on paper whiteness, which can quantify the correlation between paper whiteness and visual fatigue and improve the accuracy of visual fatigue detection.

[0005] In a first aspect, the present application provides a visual fatigue detection method based on paper whiteness, including:

[0006] S101: Obtain an eye temporal dataset, where the eye temporal dataset is multi-modal data collected based on preset paper whiteness parameters;

[0007] S102: Perform visual fatigue analysis on the eye temporal dataset based on a pre-trained machine learning model to obtain visual fatigue data; the visual fatigue data is used to quantify the degree of visual fatigue under specific whiteness stimulation;

[0008] S103: Perform correlation analysis based on multiple groups of visual fatigue data and corresponding paper whiteness parameters to obtain a whiteness-visual fatigue trend model, and the whiteness-visual fatigue trend model is used to characterize the non-linear relationship between paper whiteness and the degree of visual fatigue.

[0009] In one of the embodiments, performing visual fatigue analysis on the eye temporal dataset based on a pre-trained machine learning model to obtain visual fatigue data includes:

[0010] S201: Use the following formula to perform eye region image segmentation on the eye temporal dataset to obtain an eye mask, and the eye mask is used to locate the iris-sclera boundary coordinates:

[0011] M t = Softmax(W c *I t +b c )

[0012] Among them, M t is the eye mask of the t-th frame image I t , W c is the convolution kernel, and b c is the first bias term;

[0013] S202: Using the following formula, extract the dynamic visual feature vector through the spatio-temporal convolutional network based on the eye mask. The dynamic visual feature vector includes the eyelid movement rate component:

[0014]

[0015] Among them, V is the dynamic visual feature vector, ⊙ is the Hadamard product, and F t is the frequency domain feature matrix of the t-th frame, W d is the spatio-temporal convolution kernel, and b d is the second bias term;

[0016] S203: Perform iris oscillation frequency analysis and calculate the eyelid closing rate based on the iris-sclera boundary coordinates and the dynamic visual feature vector to obtain visual fatigue data; the visual fatigue data is the weighted quantization value of the iris oscillation data and the eyelid closing rate.

[0017] In one embodiment, the acquisition of the eye time series data set in S101 further includes:

[0018] S301: Obtain the data of the multi-modal sensor and align the data of each multi-modal sensor based on the dynamic time warping algorithm to obtain a cross-modal synchronous data set with eliminated time phase difference;

[0019] S302: Perform frequency comparison based on the cross-modal synchronous data set. Using the data with the highest frequency in the cross-modal synchronous data set as the reference frequency, perform linear interpolation on the data stream with a frequency lower than the reference frequency to align the frequencies of the cross-modal synchronous data set and obtain the eye time series data set.

[0020] In one embodiment, the association analysis in S103 is realized through the following steps:

[0021] S401: Normalize the whiteness parameter of the paper, map the whiteness values in different light source environments to the D65 standard light source reference system, and obtain the standardized whiteness parameter;

[0022] S402: Use principal component analysis to reduce the dimension of the visual fatigue data, extract the principal components with a correlation with whiteness higher than the preset threshold, and obtain the whiteness-principal component weight matrix;

[0023] S403: Use the stratified sampling method to divide the data into a training set and a validation set, and obtain the whiteness-visual fatigue trend model based on the training set using the cubic spline interpolation method.

[0024] In one embodiment, the weighted quantization value in S203 is obtained by the following method:

[0025] S501: Obtain the user's age and, based on the standard database of the ophthalmological society, obtain the physiological attenuation coefficient corresponding to the user's age;

[0026] S502: Based on a preset dynamic detection threshold, apply non-linear compensation to the physiological attenuation coefficient to obtain the weighted quantization values corresponding to the iris tremor data and the eyelid closure rate.

[0027] In one embodiment, the pre-trained machine learning model in S102 is optimized in the following manner:

[0028] S601: Obtain the user's multi-day eye movement data to construct an incremental data set;

[0029] S602: Based on the knowledge distillation technique, transfer the general model parameters to a lightweight sub-network;

[0030] S603: Fine-tune the sub-network based on the incremental data set to obtain the pre-trained machine learning model.

[0031] In one embodiment, the preset paper whiteness parameter is obtained by the following method:

[0032] S701: Obtain the light intensity distribution of each wavelength within the visible light band to generate an ambient light feature vector including color temperature and illuminance; the light intensity distribution is obtained by measuring the surface of the target paper in real time with a spectrophotometer;

[0033] S702: Input the original image data and the ambient light feature vector into a pre-trained inverse rendering model to solve the true spectral reflectance curve of the paper;

[0034] S703: According to the correlation analysis between color temperature and CIE whiteness value, obtain the whiteness compensation coefficient under the current environment by means of a look-up table method, and obtain the paper whiteness parameter based on the whiteness compensation coefficient.

[0035] In a second aspect, the present application also provides a visual fatigue detection device based on paper whiteness, including:

[0036] A data acquisition module for acquiring an eye movement time series data set, the eye movement time series data set being multi-modal data collected based on a preset paper whiteness parameter;

[0037] A visual fatigue analysis module for performing visual fatigue analysis on the eye movement time series data set based on a pre-trained machine learning model to obtain visual fatigue data; the visual fatigue data is used to quantify the degree of visual fatigue under a specific whiteness stimulus;

[0038] The correlation analysis module is used to perform correlation analysis based on multiple sets of visual fatigue data and corresponding paper whiteness parameters to obtain a whiteness-visual fatigue trend model, which is used to characterize the non-linear relationship between paper whiteness and visual fatigue degree.

[0039] In a third aspect, the present application also provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the above-mentioned visual fatigue detection method based on paper whiteness is implemented.

[0040] In a fourth aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned visual fatigue detection method based on paper whiteness is implemented.

[0041] The above-mentioned visual fatigue detection method based on paper whiteness collects multi-modal data based on preset paper whiteness parameters, constructs an eye time series data set, and uses a pre-trained machine learning model to analyze visual fatigue of this data set, so as to quantify the degree of visual fatigue under specific whiteness stimuli, generate visual fatigue data, perform correlation analysis on multiple sets of visual fatigue data and corresponding paper whiteness parameters, and construct a whiteness-visual fatigue trend model that can accurately characterize the non-linear relationship between paper whiteness and visual fatigue degree. The above method realizes the non-linear correlation quantification of the correlation between paper whiteness and visual fatigue, and improves the accuracy of visual fatigue detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0043] Figure 1 It is a schematic flowchart of a visual fatigue detection method based on paper whiteness provided by an embodiment of the present invention;

[0044] Figure 2 It is a schematic structural diagram of a visual fatigue detection device based on paper whiteness provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] In order to make the objectives, technical solutions and advantages of the present application clearer, the following further details the present application in conjunction with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0046] First, a brief introduction to the terms involved in the embodiments of the present application is given.

[0047] The whiteness of paper refers to the brightness of the paper, that is, the ability of the paper to reflect light. Whiteness is an important parameter of paper, which represents the light reflection ability of the paper within the full wavelength range. The larger the whiteness value, the higher the whiteness degree of the paper. The measurement of paper whiteness usually uses a whiteness meter, and the reflectance of the paper under blue light irradiation is measured to determine it. In China's national standard GB / T7974, brightness is measured by the reflectance factor of the specimen within the main wavelength blue light under the condition of D65 light source.

[0048] The Dynamic Time Warping (DTW) algorithm is used to measure the similarity between two time series, and can also be used to align multiple test sequences with a standard sequence, thereby realizing the normalization of sequence length.

[0049] According to the above noun explanations, the implementation environment of the visual fatigue detection method based on paper whiteness provided by the embodiments of this application is described. Schematically, this implementation environment includes: a sensor array, a processor, and a storage device. Among them, the processor is connected to the sensor array and the storage device through network signals; the sensor array can be an optical camera, an eye tracker, an electrooculogram sensor, or an electrooculogram sensor; the processor includes but is not limited to a central processing unit, a multi-core processor, or an artificial intelligence chip, etc.; the storage device can be a distributed storage system or a centralized storage system, which is not limited here.

[0050] Combined with the above noun explanations and implementation environment, the application scenarios of the embodiments of this application are described. The visual fatigue detection method based on paper whiteness provided in the embodiments of this application can be applied to the following scenarios including but not limited to:

[0051] When a paper production enterprise develops new products, by applying the visual fatigue detection technology based on paper whiteness, it can deeply understand consumers' acceptance of papers with different whitenesses and their visual fatigue feedback. By optimizing the production process, paper products that meet market demands and can effectively reduce visual fatigue can be produced, meeting the requirements for paper quality and eye protection functions in multiple fields such as office, education, and printing, and enhancing the competitiveness of the enterprise's products in the market.

[0052] In an office environment, employees daily process a large number of paper documents. Through this technical solution, an enterprise can analyze the impact of the paper whiteness read by employees in different departments on visual fatigue. For example, employees in the copywriting editing department read copywriting manuscripts for a long time. If high-whiteness paper is used, it may lead to increased visual fatigue, affecting work efficiency and the visual health of employees. The enterprise can customize papers with appropriate whiteness for employees according to the detection results, such as choosing papers that are slightly yellowish and have a whiteness of 80%-85%, which can not only ensure that the text is clearly readable but also effectively reduce visual fatigue and improve employees' comfort and work efficiency.

[0053] Schematically, the visual fatigue detection method based on paper whiteness provided by the embodiments of the present application can also be applied to other application scenarios. Only examples are given here, and the specific application scenarios are not limited.

[0054] In an exemplary embodiment, as Figure 1 shown, a visual fatigue detection method based on paper whiteness is provided, including the following S101 to S103:

[0055] S101: Obtain an eye temporal dataset, which is multi-modal data collected based on preset paper whiteness parameters.

[0056] Specifically, a series of preset paper whiteness parameter values covering different whiteness ranges can be determined. The tester conducts eye-using activities such as reading and writing in an environment of papers with different whiteness. Multiple sensor devices, such as an eye tracker and an eye muscle electrical sensor, are used to collect multi-modal data of the eyes, including the eye movement trajectory, blink frequency, eye muscle electrical signals, etc., and record them in time sequence to form an eye temporal dataset, so as to comprehensively reflect the real-time state of the eyes under different paper whiteness stimuli.

[0057] S102: Perform visual fatigue analysis on the eye temporal dataset based on a pre-trained machine learning model to obtain visual fatigue data; the visual fatigue data is used to quantify the degree of visual fatigue under specific whiteness stimuli.

[0058] Specifically, a machine learning model trained with visual fatigue-related data, such as a deep neural network model, can be used to input the eye temporal dataset into the pre-trained model. Based on the features and patterns it has learned, the model deeply analyzes the data and outputs visual fatigue data that can quantify the degree of visual fatigue. For example, the model may comprehensively judge the degree of visual fatigue according to the strength of the eye muscle electrical signal, the change range of the blink frequency, etc., and present it in numerical form.

[0059] S103: Conduct correlation analysis based on multiple groups of visual fatigue data and the corresponding paper whiteness parameters to obtain a whiteness-visual fatigue trend model, which is used to characterize the non-linear relationship between paper whiteness and the degree of visual fatigue.

[0060] Specifically, by collecting multiple groups of visual fatigue data obtained under different paper whiteness parameters and the actual parameter values of the corresponding paper whiteness, data mining and statistical analysis methods are used for correlation analysis to obtain a whiteness-visual fatigue trend model.

[0061] The above-mentioned visual fatigue detection method based on paper whiteness collects multi-modal data based on preset paper whiteness parameters, constructs an eye time-series data set, and uses a pre-trained machine learning model to analyze the visual fatigue of the data set, thereby quantifying the degree of visual fatigue under specific whiteness stimuli, generating visual fatigue data, and conducting correlation analysis on multiple groups of visual fatigue data and corresponding paper whiteness parameters to construct a whiteness-visual fatigue trend model that can accurately represent the non-linear relationship between paper whiteness and visual fatigue degree. The above method realizes the non-linear correlation quantification of the correlation between paper whiteness and visual fatigue, and improves the accuracy of visual fatigue detection.

[0062] In one embodiment, the visual fatigue analysis of the eye time-series data set based on the pre-trained machine learning model to obtain visual fatigue data may include the following steps:

[0063] S201: Use the following formula to perform eye region image segmentation on the eye time-series data set to obtain an eye mask, and the eye mask is used to locate the iris-sclera boundary coordinates:

[0064] M t = Softmax(W c * I t + b c )

[0065] where M t is the eye mask of the t-th frame image I t , W c is the convolution kernel, and b c is the first bias term.

[0066] Specifically, perform pixel-level segmentation on the eye time-series images through a convolutional neural network to generate a high-precision eye mask, which can accurately locate the boundary contour of the iris and sclera and provide a spatial reference for subsequent feature extraction.

[0067] S202: Use the following formula to extract dynamic visual feature vectors based on the eye mask through a spatio-temporal convolutional network, and the dynamic visual feature vectors include eyelid movement rate components:

[0068]

[0069] where V is the dynamic visual feature vector, ⊙ is the Hadamard product, F t is the frequency domain feature matrix of the t-th frame, W d is the spatio-temporal convolution kernel, and b d is the second bias term.

[0070] Specifically, use the spatio-temporal convolutional network to fuse the spatio-temporal information of multiple frames of eye masks, extract the eyelid movement trajectory features in the frequency domain space, and construct dynamic feature vectors representing visual fatigue.

[0071] S203: Analyze the iris oscillation frequency based on the iris-sclera boundary coordinates and dynamic vision feature vectors, calculate the eyelid closing rate, and obtain visual fatigue data; the visual fatigue data is a weighted quantization value of the iris oscillation data and the eyelid closing rate.

[0072] Specifically, by combining the geometric features of the iris boundary coordinates with the dynamic vision feature vectors, calculate the main frequency of iris oscillation through the frequency domain decomposition algorithm, synchronously count the eyelid closing rate, and finally fuse and generate a weighted fatigue degree quantization value. Through the above technical methods, it is possible to break through the limitations of single indicators, establish a multi-dimensional fatigue assessment system, and improve the accuracy of visual fatigue data detection.

[0073] In one embodiment, the acquisition of the eye temporal dataset in S101 further includes:

[0074] S301: Obtain the data of the multi-modal sensors and align the data of each multi-modal sensor based on the dynamic time warping algorithm to obtain a cross-modal synchronous dataset with the time phase difference eliminated.

[0075] Specifically, analyze the time offset of the data streams of each sensor through the dynamic time warping algorithm, solve the signal misalignment problem caused by the sampling start time difference of devices such as eye trackers and electroencephalographs, and at the same time retain the original data time series correlation, enhancing the physical meaning consistency of multi-modal feature fusion.

[0076] S302: Perform frequency comparison based on the cross-modal synchronous dataset, use the data with the highest frequency in the cross-modal synchronous dataset as the reference frequency, linearly interpolate the data streams with frequencies lower than the reference frequency, align the frequencies of the cross-modal synchronous dataset, and obtain the eye temporal dataset.

[0077] Exemplarily, taking the data of the high-sampling rate sensor (such as a 1200Hz eye tracker) as the reference, interpolate and reconstruct the low-frequency data stream (such as a 30Hz ambient light sensor) to generate an equally spaced dataset with a unified timestamp. The above technical solution can eliminate the loss of high-frequency signal details caused by sampling rate differences, ensure the comparability of different frequency features in time domain analysis, avoid training biases caused by input dimension mismatches in machine learning models, and improve the reliability of the models.

[0078] In one embodiment, the correlation analysis in S103 can be achieved through the following steps:

[0079] S401: Normalize the paper whiteness parameter, map the whiteness values in different light source environments to the D65 standard light source reference system, and obtain the standardized whiteness parameter.

[0080] Specifically, through the light source spectral matching algorithm, the whiteness measurement values in different color temperature environments are converted to the chromaticity space of the D65 standard light source, eliminating the interference of ambient light color deviation on the paper whiteness evaluation, so as to convert the whiteness measurement values in different color temperature environments to the chromaticity space of the D65 standard light source, eliminating the interference of ambient light color deviation on the paper whiteness evaluation.

[0081] S402: Use principal component analysis to reduce the dimension of the visual fatigue data, extract the principal components with a correlation with whiteness higher than the preset threshold, and obtain the whiteness - principal component weight matrix.

[0082] Specifically, through principal component analysis, the physiological feature components significantly correlated with the change in whiteness are screened, the noise interference terms are removed, a low - dimensional and highly interpretable fatigue feature space is constructed, and the model calculation efficiency is improved.

[0083] S403: Use the stratified sampling method to divide the data into a training set and a validation set, and based on the training set, use the cubic spline interpolation method to obtain the whiteness - visual fatigue trend model.

[0084] Specifically, the stratified sampling method ensures that the training set and the validation set can reasonably represent the characteristics of the original data set, improves the reliability and generalization ability of model training, and avoids the influence of sampling bias on the model performance; the model constructed by the cubic spline interpolation method can well fit the change trend of the data, generate a smooth and continuous curve, more accurately reflect the complex non - linear relationship between whiteness and visual fatigue, provides an intuitive and effective tool for visual fatigue detection and prediction, and improves the practicality of the technical solution.

[0085] Furthermore, the weighted quantization value in S203 is obtained by the following method:

[0086] S501: Obtain the user's age and based on the standard database of the ophthalmological society, obtain the physiological attenuation coefficient corresponding to the user's age.

[0087] Specifically, by querying the standard database of the ophthalmological society, the user's age can be mapped to the corresponding physiological attenuation coefficient, and an age - related visual function decline benchmark model can be established to introduce the age parameter to achieve personalized fatigue assessment calibration, effectively distinguishing the physiological signal differences between natural aging and abnormal fatigue.

[0088] S502: Based on the preset dynamic detection threshold, apply non - linear compensation to the physiological attenuation coefficient to obtain the weighted quantization values corresponding to the iris oscillation data and the eyelid closure rate.

[0089] Specifically, adjust the non - linear compensation function according to the dynamic detection threshold, associate biometric features such as iris oscillation frequency and eyelid closure rate to the standardized weight space to enhance the horizontal comparability of fatigue magnitude values of users of different age groups.

[0090] Further, the pre-trained machine learning model in S102 is optimized in the following manner:

[0091] S601: Obtain the eye movement dynamic data of the user for multiple consecutive days to construct an incremental data set.

[0092] Specifically, by continuously collecting the eye movement, pupil, and ambient light data in the user's daily reading scenario, a personalized training sample library evolving over time is constructed to capture the long-term change pattern of the user's eye-using habits and enhance the model's adaptability to individual physiological feature drift.

[0093] S602: Transfer the general model parameters to a lightweight sub-network based on the knowledge distillation technique.

[0094] Specifically, transfer the deep feature representation ability of the general large model to the lightweight sub-network through a soft label matching mechanism, and retain the core visual fatigue detection logic to reduce the model's computational complexity, adapt to the deployment of embedded devices, and improve the model's generalization ability.

[0095] S603: Fine-tune the sub-network based on the incremental data set to obtain the pre-trained machine learning model.

[0096] Specifically, dynamically adjust the attention weight distribution of the sub-network based on the incremental data, optimize its response sensitivity to the user-specific fatigue signals, improve the real-time performance and accuracy of personalized fatigue detection, and prevent the performance degradation of the model caused by local data overfitting.

[0097] In one embodiment, the preset paper whiteness parameter is obtained by the following method:

[0098] S701: Obtain the light intensity distribution of each wavelength within the visible light band to obtain an environmental light feature vector containing color temperature and illuminance; the light intensity distribution is obtained by measuring the surface of the target paper in real time with a spectrophotometer.

[0099] Specifically, scan the surface reflection spectrum of the paper with a spectrophotometer, and convert the light intensity distribution in the visible light band (380 - 780nm) into a feature vector containing color temperature, illuminance, and spectral power distribution to achieve synchronous and accurate measurement of multi-dimensional optical parameters.

[0100] S702: Input the original image data and the environmental light feature vector into the pre-trained inverse rendering model to solve the true spectral reflectance curve of the paper.

[0101] Specifically, input the original image data and the environmental light feature into the pre-trained inverse rendering model, and inversely deduce the intrinsic spectral reflectance of the paper surface through the physical rendering equation to break through the limitation that traditional RGB cameras cannot capture full-spectrum information.

[0102] S703: According to the correlation analysis between color temperature and CIE whiteness value, obtain the whiteness compensation coefficient in the current environment through the look-up table method, and obtain the paper whiteness parameter based on the whiteness compensation coefficient.

[0103] Specifically, query the preset whiteness compensation mapping table according to the ambient light color temperature, correct the weight of the blue light band of the reflectivity curve through linear weighting, and output the environment-adaptive standardized whiteness parameter to dynamically compensate for the influence of ambient color temperature offset on whiteness perception, ensuring the consistency of whiteness evaluation results across light source scenarios.

[0104] In summary, the visual fatigue detection method based on paper whiteness provided by the embodiments of the present application collects high-precision optical parameters and eye biometric data in real time through a multi-modal sensor array, realizes spatio-temporal synchronization of cross-source heterogeneous data through the dynamic time warping algorithm, and uses machine learning algorithms for visual fatigue feature decoupling and non-linear modeling; based on the inverse rendering technology, calculate the intrinsic spectral characteristics of the paper, construct a dose-response relationship model between the whiteness parameter and the physiological fatigue index through multi-dimensional regression analysis, and introduce the knowledge distillation method and the incremental learning mechanism to realize model lightweight and personalized adaptation. The above technical solutions can improve the interpretability of the physiological mechanism of visual fatigue detection through multi-modal data fusion, enhance cross-group adaptability based on the personalized weight assignment of the age attenuation model, realize the non-linear correlation quantification of the correlation between paper whiteness and visual fatigue, and improve the accuracy of visual fatigue detection.

[0105] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are shown sequentially according to the arrows, these steps do not necessarily have to be executed in the order indicated by the arrows. Unless otherwise clearly stated in this article, there is no strict order limit for the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages do not necessarily have to be executed at the same time, but can be executed at different times. The execution order of these steps or stages does not necessarily have to be sequential, but can be executed alternately or alternately with at least a part of the steps or stages in other steps or other steps.

[0106] Based on the same inventive concept, the embodiments of the present application also provide a visual fatigue detection device based on paper whiteness for implementing the above-mentioned visual fatigue detection method based on paper whiteness. The implementation solutions provided by the device to solve problems are similar to the implementation solutions described in the above method. Therefore, the specific limitations in one or more embodiments of the visual fatigue detection device based on paper whiteness provided below can refer to the limitations on the visual fatigue detection method based on paper whiteness in the above text, and will not be repeated here.

[0107] In an exemplary embodiment, as Figure 2 shown, a visual fatigue detection device 20 based on paper whiteness is provided, including:

[0108] A data acquisition module 21, configured to acquire an eye time-series data set, where the eye time-series data set is multimodal data collected based on preset paper whiteness parameters.

[0109] A visual fatigue analysis module 22, configured to perform visual fatigue analysis on the eye time-series data set based on a pre-trained machine learning model to obtain visual fatigue data; the visual fatigue data is used to quantify the degree of visual fatigue under specific whiteness stimuli.

[0110] A correlation analysis module 23, configured to perform correlation analysis according to multiple groups of visual fatigue data and corresponding paper whiteness parameters to obtain a whiteness-visual fatigue trend model, and the whiteness-visual fatigue trend model is used to characterize the non-linear relationship between paper whiteness and the degree of visual fatigue.

[0111] In one embodiment, the visual fatigue analysis module 22 may include:

[0112] A mask segmentation unit 221, configured to use the following formula to perform eye region image segmentation on the eye time-series data set to obtain an eye mask, and the eye mask is used to locate the iris-sclera boundary coordinates:

[0113] M t = Softmax(W c *I t +b c )

[0114] where M t is the eye mask of the t-th frame image I t , W c is the convolution kernel, and b c is the first bias term.

[0115] A dynamic feature extraction unit 222, configured to use the following formula to extract a dynamic visual feature vector based on the eye mask through a spatio-temporal convolutional network, and the dynamic visual feature vector includes an eyelid movement rate component:

[0116]

[0117] where V is the dynamic visual feature vector, ⊙ is the Hadamard product, F t is the frequency domain feature matrix of the t-th frame, W d is the spatio-temporal convolution kernel, and b d is the second bias term.

[0118] A visual fatigue data analysis unit 223, configured to analyze the iris oscillation frequency based on the iris-sclera boundary coordinates and the dynamic vision feature vector, calculate the eyelid closing rate, and obtain visual fatigue data; the visual fatigue data is a weighted quantization value of the iris oscillation data and the eyelid closing rate.

[0119] In one embodiment, the acquisition of the eye temporal data set in the data acquisition module 21 may further include:

[0120] A hardware synchronization unit 211, configured to acquire the data of the multi-modal sensors and align the data of the multi-modal sensors based on the dynamic time warping algorithm to obtain a cross-modal synchronization data set with the time phase difference eliminated.

[0121] An interpolation synchronization unit 212, configured to perform frequency comparison based on the cross-modal synchronization data set, use the data with the highest frequency in the cross-modal synchronization data set as the reference frequency, perform linear interpolation on the data stream with a frequency lower than the reference frequency, align the frequencies of the cross-modal synchronization data set, and obtain an eye temporal data set.

[0122] In one embodiment, the correlation analysis in the correlation analysis module 23 can be realized through the following steps:

[0123] S401: Normalize the paper whiteness parameter, map the whiteness values under different light source environments to the D65 standard light source reference system, and obtain a standardized whiteness parameter.

[0124] S402: Use principal component analysis to reduce the dimension of the visual fatigue data, extract the principal components with a correlation with whiteness higher than a preset threshold, and obtain a whiteness-principal component weight matrix.

[0125] S403: Use the stratified sampling method to divide the data into a training set and a validation set, and obtain a whiteness-visual fatigue trend model based on the training set using the cubic spline interpolation method.

[0126] In one embodiment, the weighted quantization value in the visual fatigue data analysis unit 223 is obtained by the following method:

[0127] S501: Obtain the user's age and, based on the standard database of the ophthalmological society, obtain the physiological attenuation coefficient corresponding to the user's age;

[0128] S502: Apply non-linear compensation to the physiological attenuation coefficient based on a preset dynamic detection threshold to obtain the weighted quantization value corresponding to the iris oscillation data and the eyelid closing rate.

[0129] In one embodiment, the pre-trained machine learning model in the data acquisition module 21 is optimized in the following manner:

[0130] S601: Obtain the eye dynamic data of the user for multiple consecutive days to construct an incremental data set;

[0131] S602: Transfer the general model parameters to the lightweight sub-network based on the knowledge distillation technique;

[0132] S603: Fine-tune the sub-network based on the incremental dataset to obtain a pre-trained machine learning model.

[0133] In one embodiment, the paper whiteness parameter preset in the data acquisition module 21 is obtained by the following method:

[0134] S701: Obtain the light intensity distribution of each wavelength within the visible light band to generate an ambient light feature vector containing color temperature and illuminance; the light intensity distribution is obtained by measuring the surface of the target paper in real time with a spectrophotometer;

[0135] S702: Input the original image data and the ambient light feature vector into the pre-trained inverse rendering model to solve the true spectral reflectance curve of the paper;

[0136] S703: According to the correlation analysis between the color temperature and the CIE whiteness value, obtain the whiteness compensation coefficient in the current environment by the look-up table method, and obtain the paper whiteness parameter based on the whiteness compensation coefficient.

[0137] In one embodiment, a computer device is provided, including a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the steps of a visual fatigue detection method based on paper whiteness as described above are implemented.

[0138] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0139] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can be referred to the partial description of the method embodiment. The device embodiments described above are only illustrative, where the components described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present disclosure solution. Those of ordinary skill in the art can understand and implement it without creative work.

[0140] The above-described embodiments merely represent several implementation manners of the embodiments of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the embodiments of the application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the embodiments of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the embodiments of the present application.

Claims

1. A visual fatigue detection method based on paper whiteness, characterized in that, The method includes: S101: Obtain an eye time-series data set, where the eye time-series data set is multi-modal data collected based on a preset paper whiteness parameter; S102: Perform visual fatigue analysis on the eye time-series data set based on a pre-trained machine learning model to obtain visual fatigue data; the visual fatigue data is used to quantify the degree of visual fatigue under a specific whiteness stimulus; S103: Conduct correlation analysis based on multiple groups of the visual fatigue data and the corresponding paper whiteness parameters to obtain a whiteness-visual fatigue trend model, where the whiteness-visual fatigue trend model is used to characterize the non-linear relationship between the paper whiteness and the degree of visual fatigue.

2. The method according to claim 1, wherein The performing visual fatigue analysis on the eye time-series data set based on a pre-trained machine learning model to obtain visual fatigue data includes: S201: Use the following formula to perform eye region image segmentation on the eye time-series data set to obtain an eye mask, where the eye mask is used to locate the iris-sclera boundary coordinates: M t = Softmax(W c * I t + b c ) Among them, M t is the eye mask of the t-th frame image I t , W c is the convolutional kernel, b c is the first bias term; S202: Use the following formula to extract dynamic visual feature vectors based on the eye mask through a spatio-temporal convolutional network, where the dynamic visual feature vectors include eyelid movement rate components: Wherein, V is the dynamic visual feature vector, ⊙ is the Hadamard product, F t is the frequency domain feature matrix of the t-th frame, W d is the spatio-temporal convolution kernel, b d is the second bias term; S203: Conduct iris oscillation frequency analysis based on the iris-sclera boundary coordinates and the dynamic visual feature vectors and calculate the eyelid closure rate to obtain visual fatigue data; the visual fatigue data is a weighted quantization value of the iris oscillation data and the eyelid closure rate.

3. The method according to claim 2, wherein The acquisition of the eye time-series data set in S101 further includes: S301: Obtain data from multi-modal sensors and align the data of each multi-modal sensor based on the dynamic time warping algorithm to obtain a cross-modal synchronous data set with eliminated time phase differences; S302: Conduct frequency comparison based on the cross-modal synchronous data set. Taking the data with the highest frequency in the cross-modal synchronous data set as the reference frequency, perform linear interpolation on the data stream with a frequency lower than the reference frequency to align the frequencies of the cross-modal synchronous data set to obtain the eye time-series data set.

4. The method according to claim 1, wherein The correlation analysis in S103 is achieved through the following sub-steps: S401: Perform normalization processing on the paper whiteness parameters, map the whiteness values under different light source environments to the D65 standard light source reference system to obtain standardized whiteness parameters; S403: Use the principal component analysis to reduce the dimension of the visual fatigue data, extract the principal components with a correlation with whiteness higher than a preset threshold to obtain a whiteness-principal component weight matrix; S403: Use the stratified sampling method to divide the data into a training set and a validation set, and obtain a whiteness-visual fatigue trend model based on the training set using the cubic spline interpolation method.

5. The method according to claim 2, characterized in that The weighted quantization value in S203 is obtained through the following method: S501: Obtain the user's age and based on the standard database of the ophthalmological society, obtain the physiological attenuation coefficient corresponding to the user's age; S502: Apply non-linear compensation to the physiological attenuation coefficient based on a preset dynamic detection threshold to obtain the weighted quantization value corresponding to the iris oscillation data and the eyelid closure rate.

6. The method according to claim 2, characterized in that, The pre-trained machine learning model in S102 is optimized through the following method: S601: Obtain the eye movement data of the user for multiple consecutive days to construct an incremental data set; S602: Transfer the general model parameters to a lightweight sub-network based on the knowledge distillation technique; S603: Fine-tune the sub-network based on the incremental data set to obtain the pre-trained machine learning model.

7. The method according to claim 1, wherein The preset paper whiteness parameter is obtained by the following method: S701: Obtain the light intensity distribution at each wavelength within the visible light band to generate an ambient light feature vector including color temperature and illuminance; the light intensity distribution is obtained by measuring the surface of the target paper in real time with a spectrophotometer; S702: Input the original image data and the ambient light feature vector into a pre-trained inverse rendering model to solve the true spectral reflectance curve of the paper; S703: According to the correlation analysis between the color temperature and the CIE whiteness value, obtain the whiteness compensation coefficient in the current environment by means of a look-up table method, and obtain the paper whiteness parameter based on the whiteness compensation coefficient.

8. An apparatus for detecting visual fatigue based on the whiteness of paper, characterized in that, The device includes: A data acquisition module, configured to acquire an eye movement time series data set, where the eye movement time series data set is multi-modal data collected based on a preset paper whiteness parameter; A visual fatigue analysis module, configured to perform visual fatigue analysis on the eye movement time series data set based on a pre-trained machine learning model to obtain visual fatigue data; the visual fatigue data is used to quantify the degree of visual fatigue under a specific whiteness stimulus; A correlation analysis module, configured to perform correlation analysis on multiple groups of the visual fatigue data and the corresponding paper whiteness parameters to obtain a whiteness-visual fatigue trend model, where the whiteness-visual fatigue trend model is used to characterize the non-linear relationship between the paper whiteness and the degree of visual fatigue.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the method according to any one of claims 1 to 7 is implemented.

Images