A psychological stress recognition method and device, computer equipment and storage medium

Abstract

This invention discloses a method, device, computer equipment, and storage medium for identifying psychological stress, belonging to the field of computer technology. Addressing the problem in existing technologies that extract psychological stress information from a single indicator cannot effectively achieve stable psychological stress assessment, this invention provides a method, device, computer equipment, and storage medium for identifying psychological stress. It extracts pulse rate variability signal features, fuses these features with facial expression features to obtain fused features, and uses a forward-backward feature selection algorithm to collaboratively select the optimal subset of fused features. A psychological stress identification model is constructed using this optimal subset, and the model is then used to classify psychological stress, thereby fully acquiring psychological stress information and effectively improving the classification performance of the psychological stress identification model.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and more specifically, to a method, apparatus, computer device, and storage medium for identifying psychological stress. Background Technology

[0002] With the continuous development of society and the economy, people's pace of life is becoming increasingly fast, and they are facing increasing pressure from life and work, such as workload, academic pressure, family pressure, and social pressure. This increasing pressure has led to increasingly serious mental health problems, including diseases such as depression, which has become a serious social issue. We need to pay attention to mental health issues and take proactive measures to promptly detect people's psychological stress levels in order to cope with the increasingly fast pace of life and stressful environment. In this context, timely identification of psychological stress and intervention becomes particularly important. Psychological intervention includes psychological counseling, behavioral therapy, and medication; these methods can help people relieve stress and restore a normal life.

[0003] Under increased psychological stress, the autonomic nervous system (ANS) plays a crucial regulatory role. Heart rate variability (HRV) is one of the important indicators for measuring ANS activity. Since the ANS is not subject to the conscious control of the brain, the physiological signal changes caused by the body's own regulation are objective, and its effectiveness has been proven by numerous studies.

[0004] Currently, methods for measuring HRV mainly include electrocardiography (ECG), photoplethysmography (PPG), and video photoplethysmography (VPPG). While ECG offers high accuracy, it limits participant comfort and freedom of movement. ECG measurement also requires specialized equipment and skills, including an ECG machine and skilled operators, which restricts the availability and cost of HRV measurement. PPG is simpler and more comfortable than ECG, but it still has a degree of invasiveness. VPPG offers higher temporal and spatial resolution than traditional PPG, providing more detailed HRV measurement results, and it is more comfortable and convenient as it does not require direct skin contact. However, there is still room for improvement in VPPG-based psychological stress assessment; therefore, no effective solution has yet been proposed for achieving stable psychological stress assessment using video photoplethysmography.

[0005] A search revealed Chinese patent application No. 201911307960.1, published on April 24, 2020, which discloses a method and apparatus for identifying psychological stress based on a neural network. This method acquires photoplethysmography (PPG) data of a user whose psychological stress needs to be identified; segments the PPG data to obtain segmented PPG data; performs equal-length processing on the segmented PPG data to obtain equal-length PPG data; and inputs the equal-length PPG data into a psychological stress identification model to obtain the user's psychological stress information. The psychological stress identification model is obtained by pre-training a preset neural network model using PPG training data labeled with psychological stress information, thereby achieving psychological stress identification based on a neural network. However, this solution does not consider the low temporal and spatial resolution of the PPG data, resulting in low accuracy in identifying psychological stress. Summary of the Invention

[0006] 1. Technical problems to be solved

[0007] To address the problem that existing technologies cannot effectively achieve stable psychological stress assessment by extracting psychological stress information from a single indicator, this invention provides a psychological stress recognition method, device, computer equipment, and storage medium. It constructs a psychological stress recognition model by fusing pulse rate variability signal features and facial expression features at the feature layer, and then comprehensively analyzes these two features through the psychological stress recognition model, which can fully obtain psychological stress information and effectively improve the classification performance of the psychological stress recognition model.

[0008] 2. Technical Solution

[0009] The objective of this invention is achieved through the following technical solutions.

[0010] A method for identifying psychological stress includes the following steps:

[0011] Acquire video footage, identify individuals using the video information, and extract pulse rate variability signals and facial expression features from the individuals' faces;

[0012] Pulse rate variability signal features are extracted, and pulse rate variability signal features and facial expression features are fused to obtain fused features;

[0013] The optimal fusion feature subset is selected collaboratively using a sequence forward-backward feature selection algorithm, and a psychological stress recognition model is constructed using the optimal fusion feature subset.

[0014] Psychological stress is classified using a psychological stress identification model.

[0015] Furthermore, a sequence forward-backward feature selection algorithm is used to collaboratively select the optimal fused feature subset, including:

[0016] The sequential forward feature selection algorithm starts from an empty fusion feature subset and selects a fusion feature that optimizes the evaluation function value to add to the fusion feature subset at each step.

[0017] The sequential backward feature selection algorithm starts from the entire set of fused features and removes one fused feature at a time. After removing the fused feature, the subset of fused features makes the evaluation function value optimal.

[0018] The optimal fused feature subset is obtained by fusing the fused feature subset obtained by the forward sequence feature selection algorithm and the fused feature subset obtained by the backward sequence feature selection algorithm.

[0019] Furthermore, pulse rate variability signals are obtained from facial videos of individuals, including:

[0020] Identify people's faces and track the facial regions in each frame, separating the mixed signals of the facial region video into independent signals;

[0021] Spatially average all pixels in the region of interest of a person's face in the video to form an original signal, and then de-trendize and normalize the original signal;

[0022] After processing the detrended and normalized raw signal, the pulse-pulse intervals were recorded to construct a time interval sequence of facial pulse rate characteristics, thereby obtaining the pulse rate variability signal.

[0023] Furthermore, the pulse rate variability signal features include one or more of time-domain features, frequency-domain features, and nonlinear features; for time-domain features, the time-domain features of the pulse rate variability signal are extracted from the time interval sequence between two adjacent peaks.

[0024] Furthermore, for frequency domain features, Fourier transform is used to convert the pulse rate variability signal from time domain features to frequency domain features, and information from different frequency bands is extracted.

[0025] Furthermore, for nonlinear characteristics, the nonlinear characteristics of the pulse rate variability signal are obtained through the Poincaré diagram analysis method.

[0026] Furthermore, after classifying psychological stress using a psychological stress recognition model, the classification performance of the model is evaluated using statistical measures. The calculation formula is as follows:

[0027]

[0028] Where Acc represents the accuracy of the psychological stress recognition model output, Specificity represents the probability that the psychological stress recognition model displays a "non-stress label" in a stressful state, Sensitivity represents the probability that the psychological stress recognition model displays a "stress label" in a stress-free state, F1 represents the overall evaluation of the psychological stress model, TP represents correct prediction of the stress state, TN represents correct prediction of the non-stress state, FP represents incorrect prediction of the non-stress state, and FN represents incorrect prediction of the stress state.

[0029] A psychological stress recognition device, comprising:

[0030] The input module acquires video, identifies people through video information, and extracts pulse rate variability signals and facial expression features from the person's face.

[0031] The feature acquisition module extracts pulse rate variability signal features and fuses pulse rate variability signal features and facial expression features to obtain fused features;

[0032] The module constructs a psychological stress recognition model using a forward-backward feature selection algorithm to collaboratively select the optimal fusion feature subset. The algorithm involves: starting with an empty fusion feature subset and adding it to the subset at each step; then removing one fusion feature from the entire fusion feature set until the subset optimizes the evaluation function; finally, fusing the fusion feature subsets obtained through the forward and backward feature selection algorithms to obtain the optimal fusion feature subset.

[0033] The classification module uses a psychological stress recognition model to classify psychological stress.

[0034] A computer device includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the method described above.

[0035] A computer-readable storage medium storing a computer program that, when executed by a processor, performs the method described above.

[0036] 3. Beneficial effects

[0037] Compared with the prior art, the advantages of this invention are:

[0038] The present invention discloses a psychological stress recognition method, device, computer equipment, and storage medium. By extracting pulse rate variability signal features, fusing pulse rate variability signal features and facial expression features to obtain fused features, using a sequence forward-backward feature selection algorithm to collaboratively select the optimal fused feature subset, constructing a psychological stress recognition model using the optimal fused feature subset, and using the psychological stress recognition model to classify psychological stress, thereby fully acquiring psychological stress information and effectively improving the classification performance of the psychological stress recognition model. Attached Figure Description

[0039] Figure 1 This is a flowchart illustrating the psychological stress recognition process based on pulse rate variability, as described in an embodiment of the present invention.

[0040] Figure 2 This is a diagram illustrating the signal preprocessing framework of an embodiment of the present invention.

[0041] Figure 3 This is a diagram of the ICA-BVP signal before waveform preprocessing in an embodiment of the present invention;

[0042] Figure 4 This is a waveform preprocessing diagram of the ICA-BVP signal in an embodiment of the present invention;

[0043] Figure 5 This is a comparison diagram of the ICA-BVP signal before and after waveform preprocessing in an embodiment of the present invention. Detailed Implementation

[0044] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0045] Example

[0046] like Figure 1 The illustration shows a method for identifying psychological stress provided in this embodiment. The method involves acquiring video data, identifying individuals through the video information, and obtaining pulse rate variability (PRV) signals and facial expression features from the individuals' faces. PRV features are extracted, and the PRV and facial expression features are fused to obtain fused features. A forward-backward feature selection algorithm is used to collaboratively select the optimal subset of fused features, and a psychological stress identification model is constructed using this optimal subset. Finally, the psychological stress identification model is used to classify psychological stress.

[0047] Specifically, in this embodiment, video is acquired, and personnel are identified through the video information. Pulse rate variability signals and facial expression features are obtained from the personnel's faces. First, a psychological induction experiment is designed to induce stress in the participants. In this embodiment, 42 participants aged 21-29 years old participated in the psychological induction experiment, with 27 male and 15 female participants. All 42 participants were confirmed to be healthy based on their medical history and physical examinations. In this embodiment, the psychological induction experiment includes a mental arithmetic task and a color word test. The specific steps of the psychological stress induction experiment are as follows: the 42 participants were required to complete the mental arithmetic task and color word test within a specified time, and fill out a subjective task load questionnaire after the psychological induction experiment. The subjective task load questionnaire was used to assess the actual psychological stress level of the participants after the psychological induction experiment to determine whether the psychological induction experiment was successful. Simultaneously, facial video of the participants was captured during the psychological induction experiment. Specifically, artificial light sources were used in the experiment to ensure that the facial video of the 42 participants was captured under the same lighting conditions, thereby ensuring the reliability of the data. In this embodiment, a low-cost webcam is used to keep the participant's face centered in the video feed. The webcam's automatic exposure and white balance functions are disabled to avoid signal artifacts caused by continuous color adjustments. It should be noted that during the data acquisition process, participants are required to minimize unnecessary rigid head movements to ensure the stability and consistency of the facial area.

[0048] Therefore, by acquiring video and using the video information to identify people, pulse rate variability (PRV) signals and facial features can be further extracted from the people's faces. In this embodiment, video photoplethysmograph (VPPG) is used to extract Pulse Rate Variability (PRV) signals from the people's faces, such as... Figure 2 As shown, the specific steps include:

[0049] The system identifies and tracks facial regions in each frame, separating the mixed signal from the facial video into independent signals. Specifically, the Kanade-Lucas-Tomasi (KLT) algorithm is used to automatically identify and track facial regions in each frame. Tracking key feature points improves tracking stability, thereby reducing the impact of head movement on pulse rate variability signal extraction. Furthermore, Independent Component Analysis (ICA) is used to separate the mixed signal from the facial video into independent component signals. Figures 3-5As shown, in this embodiment, the signals observed from the RGB sensor of the video of the person's face area are represented by y1(t), y2(t), and y3(t), where y1(t), y2(t), and y3(t) represent the amplitudes of the recorded signal at time t, and the source signals are represented by x1(t), x2(t), and x3(t). Then:

[0050]

[0051] Wherein, the column vector is represented as , The original mixing matrix A contains mixing coefficients a ij T represents the transpose of the original mixture matrix A, and by using independent component analysis, a solution matrix W is found such that the inverse of the output original mixture matrix A is approximately represented as: ,in, This represents an approximation of the column vector x(t).

[0052] Furthermore, all pixels in the region of interest (ROI) of the facial video are spatially averaged to form a raw signal. This raw signal is then de-trended and normalized. Specifically, after separating the mixed signal of the facial video into independent signals using independent component analysis (ICA), the RGB signal source with the highest signal-to-noise ratio (SNR) is selected as the source signal. Spatially averaging is then performed on all pixels in the ROI of the facial video to generate an RGB measurement point for each frame, forming a raw signal. This raw signal is then de-trended and normalized using the following formula:

[0053]

[0054] Where x represents the original signal, and μ represents the mean of the original signal. This represents the standard deviation of the initial signal. This represents the original signal after detrending and normalization. In this embodiment, by detrending and normalizing the original signal, the interference of noise on the effective value of the original signal can be reduced, the peak signal-to-noise ratio of the original signal can be enhanced, the accuracy of the algorithm can be improved, the convergence of the effective part of the original signal can be enhanced, and the computational complexity can be reduced.

[0055] Finally, after processing the detrended and normalized original signal, a time interval sequence of pulse rate characteristics of the human face is constructed by recording pulse-pulse intervals to obtain the pulse rate variability signal. Specifically, a Butterworth filter is used to eliminate noise caused by small facial movements, breathing, blinking, and lighting. The detrended and normalized original signal is smoothed to remove high-frequency noise components, thereby improving the signal-to-noise ratio of the blood volume pulse signal. Furthermore, there may be a slight time difference between the video ground truth and the heart rate during sampling and processing. A difference operation is performed on the detrended and normalized original signal to align their time axes, thus eliminating the mismatch between the video ground truth and the heart rate. A sliding window peak identification algorithm is used to segment the detrended and normalized original signal. The location of the peak is determined by finding local maxima within each window. A time interval sequence of pulse rate characteristics of the human face is constructed by recording pulse-pulse intervals to obtain the pulse rate variability signal.

[0056] It is worth noting that in existing technologies, extracting psychological stress information solely from the pulse rate variability signal suffers from insufficient psychological stress information and fails to fully utilize facial video information, resulting in inadequate classification performance. Therefore, this embodiment introduces facial expression features to supplement psychological stress information. Specifically, the OpenFace face recognition framework is used to extract expression features from facial information, while outliers in the expression features are removed to obtain stable expression features.

[0057] Furthermore, pulse rate variability (PRV) signal features are extracted, and these features are fused with facial expression features to obtain fused features. In this embodiment, the PRV signal features include one or more of time-domain features, frequency-domain features, and nonlinear features. For time-domain features, the PRV signal time-domain features are extracted from the time interval sequence between two adjacent peaks. These time-domain features describe the variation and statistical characteristics of the PRV signal in the time domain. The time-domain features include:

[0058]

[0059] Where RMSSD represents the root mean square, SDNN represents the standard deviation, SDSD represents the standard deviation bias, N represents the number of heartbeat intervals in the time interval sequence, pNN50 represents the proportion of NN50 divided by the total number of NN intervals, NN50 represents the number of consecutive heartbeat intervals exceeding 50ms, i represents the variable, and RR i RR represents the i-th heartbeat interval between two wave peaks. i-1 This represents the (i-1)th heartbeat interval between two wave peaks. Represents the mean of the heartbeat interval sequence, RR i ＇This represents the difference between the i-th heartbeat interval and the (i+1)-th heartbeat interval. Let P represent the square of the difference between the mean heartbeat interval and the (i+1)th heartbeat interval, and let RR represent the probability. i+1 This represents the (i+1)th heartbeat interval between two wave peaks.

[0060] For frequency domain features, the pulse rate variability signal is transformed from time domain features to frequency domain features through Fourier transform, and information of different frequency bands is extracted. The frequency domain features are used to describe the energy distribution and frequency characteristics of the pulse rate variability signal in different frequency bands.

[0061] For nonlinear characteristics, the nonlinear features of the pulse rate variability signal are obtained through Poincaré diagram analysis. Specifically, for a continuous heartbeat interval, a rectangular coordinate system is established, with the i-th heartbeat interval as the abscissa and the (i+1)-th heartbeat interval as the ordinate. The coordinate points reflect the instantaneous changes in heart rate. The distribution of the coordinate points can be approximated as an ellipse, with the center of the ellipse located at the coordinate point determined by the average heartbeat interval. Under stress, it can exhibit a trend from dispersion to concentration. In this embodiment, the nonlinear features include:

[0062]

[0063] Where SD1 represents the semi-major axis of the ellipse, SD2 represents the semi-minor axis of the ellipse, j represents the number of heartbeat intervals, and RR j RR represents the j-th heartbeat interval between two peaks. j+1 Let S represent the (j+1)th heartbeat interval between two wave peaks, and let S represent the area of ​​the ellipse.

[0064] Furthermore, a sliding window calculation method is used to obtain the short-time approximate entropy of the pulse rate variability (PRV) signal, and the complexity and randomness indices of the PRV signal are analyzed to reflect its dynamic characteristics. Detrended fluctuation analysis is then used to obtain the time-series complexity and self-similarity information of the PRV signal, revealing its long-term dynamic characteristics and time-series properties.

[0065] Therefore, in this embodiment, fusing the extracted pulse rate variability signal features and facial expression features can more accurately capture and understand the characteristics and patterns of psychological stress.

[0066] Furthermore, a sequential forward-backward feature selection algorithm is used to collaboratively select the optimal fusion feature subset, and a psychological stress recognition model is constructed using the optimal fusion feature subset. In this embodiment, a combination of sequential forward selection (SFS) and sequential backward selection (SBS) algorithms is used for feature selection. The specific steps include: starting from an empty fusion feature subset using the sequential forward selection algorithm, selecting a fusion feature that optimizes the evaluation function value at each step and adding it to the fusion feature subset; starting from the complete set of fusion features using the sequential backward selection algorithm, removing one fusion feature at a time, until the fusion feature subset after removing the removed fusion feature optimizes the evaluation function value; and fusing the fusion feature subsets obtained by the sequential forward selection algorithm and the sequential backward selection algorithm to obtain the optimal fusion feature subset. It should be noted that the obtained fusion feature subset also undergoes a redundant feature subset removal operation to ensure a balance between the number and accuracy of the selected fusion feature subsets, thus ensuring that the final fusion feature subset is the most representative and relevant to the pulse rate variability signal. In this embodiment, the fusion feature subset that can be finally determined using existing software Kubius HRV Standard includes 4 time-domain features (Mean RR, SDSD, SDNN, RMSSD), 2 frequency-domain features (LF / HF, LF), and 5 nonlinear features (SD1, SD2, SD2 / SD1, ApEn, DFA_α2). Furthermore, the evaluation function used in this embodiment is an existing technology. The fusion feature subset is then divided into a training set and a validation set. The training set is used to train and optimize the psychological stress recognition model, while the validation set is used to evaluate the performance of the psychological stress recognition model. Thus, the psychological stress recognition model comprehensively analyzes the pulse rate variability signal features and facial expression features.

[0067] Therefore, after constructing a psychological stress recognition model by fusing the optimal feature subset, the model is used to classify psychological stress. After classifying psychological stress using the model, its classification performance is evaluated using statistical measures, calculated as follows:

[0068]

[0069] Where Acc represents the accuracy of the psychological stress recognition model output, Specificity represents the probability that the psychological stress recognition model displays a "non-stress label" in a stressful state, Sensitivity represents the probability that the psychological stress recognition model displays a "stress label" in a stress-free state, F1 represents the overall evaluation of the psychological stress model, TP represents correct prediction of the stress state, TN represents correct prediction of the non-stress state, FP represents incorrect prediction of the non-stress state, and FN represents incorrect prediction of the stress state.

[0070] Therefore, the psychological stress recognition method provided in this embodiment is a psychological stress recognition model based on the fusion of pulse rate variability signal features and facial expression features of a person's face. This psychological stress recognition model introduces facial expression features as a supplement to make up for the limitations of relying solely on pulse rate variability signal features, and comprehensively utilizes information from multiple feature sources, effectively improving the classification performance of the psychological stress recognition model.

[0071] This embodiment also provides a psychological stress recognition device. The psychological stress recognition device includes an input module, a feature acquisition module, a construction module, and a classification module. The input module acquires video, identifies personnel through video information, and obtains pulse rate variability signals and facial expression features from the personnel's faces. The feature acquisition module extracts pulse rate variability signal features and fuses the pulse rate variability signal features and facial expression features to obtain fused features. The construction module uses a forward-backward feature selection algorithm to collaboratively select the optimal fused feature subset, and constructs a psychological stress recognition model using the optimal fused feature subset. The collaborative selection of the optimal fused feature subset using the forward-backward feature selection algorithm includes: starting from an empty fused feature subset using the forward-backward feature selection algorithm, selecting a fused feature that optimizes the evaluation function value at each step and adding it to the fused feature subset; starting from the entire fused feature set using the backward-backward feature selection algorithm, removing one fused feature at a time, until the fused feature subset after removing the fused feature optimizes the evaluation function value; and fusing the fused feature subset obtained by the forward-backward feature selection algorithm and the fused feature subset obtained by the backward-backward feature selection algorithm to obtain the optimal fused feature subset. The classification module uses a psychological stress recognition model to classify psychological stress. The psychological stress recognition device provided in this embodiment can implement any of the methods described in the psychological stress recognition method, and the specific working process of the psychological stress recognition device can be referred to the corresponding process in the embodiment of the psychological stress recognition method. The method and device provided in this embodiment can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for example, the division of a certain module is merely a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Furthermore, the connections or communication connections shown or discussed can be indirect couplings or communication connections through some interfaces, devices, or units, or they can be electrical, mechanical, or other forms of connection.

[0072] This embodiment also provides a computer device. A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the aforementioned psychological stress recognition method.

[0073] This embodiment also provides a computer-readable storage medium. A computer-readable storage medium stores a computer program, which, when executed by a processor, performs the psychological stress recognition method described in this embodiment. The computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0074] The invention and its embodiments have been described above illustratively. This description is not restrictive, and the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. The accompanying drawings are only one embodiment of the invention, and the actual structure is not limited thereto. No reference numerals in the claims should limit the scope of the claims. Therefore, if a person skilled in the art is inspired by this description and designs a similar structure and embodiment without departing from the spirit of the invention, such design should fall within the scope of protection of this patent. Furthermore, the word "comprising" does not exclude other elements or steps, and the word "a" preceding an element does not exclude the inclusion of "a plurality" of that element. Multiple elements stated in the product claims may also be implemented by a single element through software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any specific order.

Claims

1. A method for identifying psychological stress, comprising the following steps: Acquire video footage, identify individuals using the video information, and extract pulse rate variability signals and facial expression features from the individuals' faces; Pulse rate variability signal features are extracted, and pulse rate variability signal features and facial expression features are fused to obtain fused features; The optimal fusion feature subset is selected collaboratively using a forward-backward feature selection algorithm. A psychological stress recognition model is then constructed using this optimal fusion feature subset. The selection process involves: starting with an empty fusion feature subset using the forward-backward feature selection algorithm, adding each fusion feature that optimizes the evaluation function to the subset; starting with the entire fusion feature set using the backward-backward feature selection algorithm, removing each fusion feature until the subset optimizes the evaluation function; and finally, fusing the fusion feature subsets obtained using the forward-backward feature selection algorithm to obtain the optimal fusion feature subset. Psychological stress is classified using a psychological stress identification model.

2. The method of claim 1, wherein, Extract pulse rate variability signals from facial videos, including: Identify people's faces and track the facial regions in each frame, separating the mixed signals of the facial region video into independent signals; Spatially average all pixels in the region of interest of a person's face in the video to form an original signal, and then de-trendize and normalize the original signal; After processing the detrended and normalized raw signal, the pulse-pulse intervals were recorded to construct a time interval sequence of facial pulse rate characteristics, thereby obtaining the pulse rate variability signal.

3. The psychological stress identification method according to claim 2, characterized in that, The pulse rate variability signal features include one or more of time-domain features, frequency-domain features, and nonlinear features; for time-domain features, the time-domain features of the pulse rate variability signal are extracted from the time interval sequence of two adjacent peaks.

4. The psychological stress identification method according to claim 3, characterized in that, For frequency domain features, Fourier transform is used to convert the pulse rate variability signal from time domain features to frequency domain features, and information of different frequency bands is extracted.

5. The psychological stress identification method according to claim 3, characterized in that, For nonlinear characteristics, the nonlinear characteristics of the pulse rate variability signal are obtained through the Poincaré diagram analysis method.

6. The psychological stress identification method according to claim 1, characterized in that, After classifying psychological stress using a psychological stress recognition model, the classification performance of the model is evaluated using statistical measures. The calculation formula is as follows: Where Acc represents the accuracy of the psychological stress recognition model output, Specificity represents the probability that the psychological stress recognition model displays a "non-stress label" in a stressful state, Sensitivity represents the probability that the psychological stress recognition model displays a "stress label" in a stress-free state, F1 represents the overall evaluation of the psychological stress model, TP represents correct prediction of the stress state, TN represents correct prediction of the non-stress state, FP represents incorrect prediction of the non-stress state, and FN represents incorrect prediction of the stress state.

7. A psychological stress recognition device, characterized in that, include: The input module acquires video, identifies people through video information, and extracts pulse rate variability signals and facial expression features from the person's face. The feature acquisition module extracts pulse rate variability signal features and fuses pulse rate variability signal features and facial expression features to obtain fused features; The module constructs a psychological stress recognition model using a forward-backward feature selection algorithm to collaboratively select the optimal fusion feature subset. The algorithm involves: starting with an empty fusion feature subset and adding it to the subset at each step; then removing one fusion feature from the entire fusion feature set until the subset optimizes the evaluation function; finally, fusing the fusion feature subsets obtained through the forward and backward feature selection algorithms to obtain the optimal fusion feature subset. The classification module uses a psychological stress recognition model to classify psychological stress.

8. A computer device, comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor executes the computer program, it implements the method described in any one of claims 1-6.

9. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program is executed by the processor to perform the method described in any one of claims 1-6.

Images