Emotion recognition method and system based on pulse signals, terminal equipment and storage medium

Abstract

The invention discloses an emotion recognition method and system based on a pulse signal, terminal equipment and a storage medium, and the method comprises the steps: carrying out the feature extraction of pulse signal data to be subjected to emotion recognition, and combining a preset feature weight matrix obtained through the iterative calculation of the pulse signal data with an emotion label, selecting an optimized feature data set, and finally utilizing a trained emotion recognition model to perform emotion recognition classification according to the optimized feature data set to obtain an emotion recognition result recognized by the pulse signal data. By utilizing the characteristics that the personal pulse signal cannot be controlled by the subjective consciousness of the human body and the personal pulse signal can reflect the personal emotion, the emotion recognition accuracy is improved, and the problem that the personal emotion cannot be accurately recognized due to a false facial image by utilizing a facial image recognition technology in the prior art can be solved.

Description

Technical Field

[0001] The present invention relates to the field of emotion recognition, and in particular to an emotion recognition method, system, terminal equipment and storage medium based on pulse signals. Background Art

[0002] In the daily life of modern society, human-computer interaction is becoming more and more frequent. Whether it is smart home devices, smart customer service systems or in-car interactions in smart driving, the system needs to be able to better understand the emotional state of humans. For example, if smart customer service can identify the user's dissatisfaction, it can adjust the answer strategy in time to improve user satisfaction. In addition, there is an urgent need for emotion recognition in the field of mental health. By identifying individual emotional changes, early warning and auxiliary diagnosis of mental illness can be carried out. For example, monitoring the emotional fluctuations of patients with depression can help to intervene in treatment in a timely manner. Therefore, the demand for emotion recognition is increasing.

[0003] At present, the commonly used emotion recognition technology is to perform facial emotion recognition on face images by building a neural network model. However, since an individual's facial image can be controlled by the individual's subjective consciousness, the facial image can make facial movements that are opposite to the actual emotions. As a result, the emotion recognition technology based on facial recognition is prone to being unable to accurately identify an individual's actual emotions due to false facial images. Summary of the invention

[0004] The present invention provides a pulse signal-based emotion recognition method, system, terminal device and storage medium, which can solve the problem that the existing technology using facial image recognition technology is prone to being unable to accurately recognize personal emotions due to false facial images.

[0005] In order to solve the above technical problems, an embodiment of the present invention provides an emotion recognition method based on a pulse signal, comprising:

[0006] Obtaining the pulse signal data to be identified of the tested person, and dividing the pulse signal data to be identified into time windows to obtain pulse signal data of several time windows;

[0007] Perform time domain feature extraction, frequency domain feature extraction and nonlinear feature extraction on the pulse signal data of each time window to obtain the features of each time window;

[0008] According to each feature of each time window, a feature data set of the pulse signal data is constructed;

[0009] Calculating the product of a preset feature weight matrix and each integrated feature in the feature data set to obtain an optimized feature data set;

[0010] The optimized feature data set is input into the trained emotion recognition model, so that the trained emotion recognition model performs emotion recognition classification according to the optimized feature data set to obtain the emotion recognition result.

[0011] Furthermore, the pulse signal data of each time window is subjected to time domain feature extraction, frequency domain feature extraction and nonlinear feature extraction, respectively, including:

[0012] Extracting time domain features from the pulse signal data of each time window to obtain time domain features of the pulse signal data of each time window; wherein the time domain features include the average value and standard deviation of all adjacent PP intervals, the root mean square and standard deviation of the differences between all adjacent PP intervals, and the number and percentage of the differences between all adjacent PP intervals that are greater than a preset threshold;

[0013] Performing frequency domain transformation on the pulse signal data of each time window by fast Fourier transform to obtain the power spectrum density of the pulse signal data of each time window;

[0014] According to the power spectrum density of the pulse signal data in each time window, the peak frequencies of different frequency bands are extracted for the pulse signal data in each time window, and the frequency domain features of the pulse signal data in each time window are generated according to the peak frequencies of different frequency bands; wherein the frequency domain features include the peak frequencies in the ultra-low frequency range, the low frequency range and the high frequency range, the total peak frequency, the peak frequency ratio of the low frequency to the high frequency, the normalized value of the low frequency peak frequency, the normalized value of the high frequency peak frequency, the peak frequency percentage of the ultra-low frequency, the peak frequency percentage of the low frequency and the peak frequency percentage of the high frequency;

[0015] The nonlinear features of the pulse signal data in each time window are extracted by Poincare diagram analysis, approximate entropy algorithm and sample entropy algorithm to obtain the Poincare diagram features, approximate entropy features and sample entropy features of the pulse signal data in each time window.

[0016] Furthermore, the calculation formula for the average value of all adjacent PP intervals is:

[0017]

[0018] Among them, MEAN PP is the average value of all adjacent PP intervals; PP i is the time interval between the i-th two adjacent pulse wave peaks in each time window; n is the total number of PP intervals in each time window;

[0019] The calculation formula for the standard deviation of all adjacent PP intervals is:

[0020]

[0021] Among them, SDNN PP is the standard deviation of all adjacent PP intervals; MEAN PP is the average value of all adjacent PP intervals; PP i is the time interval between the i-th two adjacent pulse wave peaks in each time window; n is the total number of PP intervals in each time window;

[0022] The calculation formula for the root mean square of the differences between all adjacent PP intervals is:

[0023]

[0024] Among them, RMSSD PP is the root mean square of the differences between all adjacent PP intervals; PP i is the time interval between the i-th two adjacent pulse wave peaks in each time window; PP i-1 is the time interval between the i-1th two adjacent pulse wave peaks in each time window; n is the total number of PP intervals in each time window;

[0025] The calculation formula for the standard deviation of the differences between all adjacent PP intervals is:

[0026]

[0027] Among them, SDSD PP RMSSD is the standard deviation of the differences between all adjacent PP intervals. PP is the root mean square of the differences between all adjacent PP intervals; PP i is the time interval between the i-th two adjacent pulse wave peaks in each time window; PP i-1 is the time interval between the i-1th two adjacent pulse wave peaks in each time window; n is the total number of PP intervals in each time window;

[0028] The calculation formula for the number of differences between all adjacent PP intervals that are greater than a preset threshold is:

[0029]

[0030] Among them, NN(A) is the number of all adjacent PP intervals with a difference greater than 50 ms; PP i is the time interval between the i-th two adjacent pulse wave peaks in each time window; PP i+1 is the time interval between the i+1th two adjacent pulse wave peaks in each time window; N is the total number of PP intervals in each time window; A is the preset threshold;

[0031] The calculation formula for the percentage of the number of differences between all adjacent PP intervals that is greater than the preset threshold is:

[0032]

[0033] Among them, pNN(A) is the percentage of all adjacent PP intervals with a difference greater than 50 ms; PP i is the time interval between the i-th two adjacent pulse wave peaks in each time window; PP i+1 is the time interval between the i+1th two adjacent pulse wave peaks in each time window; N is the total number of PP intervals in each time window; A is the preset threshold.

[0034] Furthermore, the calculation formula of the total peak frequency is:

[0035] TP=VLF+LF+HF;

[0036] Among them, TP is the total peak frequency; VLF is the peak frequency in the very low frequency range; LF is the peak frequency in the low frequency range; HF is the peak frequency in the high frequency range;

[0037] The calculation formula of the peak frequency ratio of the low frequency to the high frequency is:

[0038] Ratio_LH = LF / HF;

[0039] Among them, Ratio_LH is the peak frequency ratio of low frequency to high frequency; LF is the peak frequency in the low frequency range; HF is the peak frequency in the high frequency range;

[0040] The calculation formula of the normalized value of the low-frequency peak frequency is:

[0041] LF norm =LF / (LF+HF);

[0042] Among them, LF norm is the normalized value of the low-frequency peak frequency; LF is the peak frequency in the low-frequency range; HF is the peak frequency in the high-frequency range;

[0043] The calculation formula of the normalized value of the high frequency peak frequency is:

[0044] HF norm =HF / (LF+HF);

[0045] Among them, HF norm is the normalized value of the high-frequency peak frequency; LF is the peak frequency in the low-frequency range; HF is the peak frequency in the high-frequency range;

[0046] The calculation formula of the peak frequency percentage of the ultra-low frequency is:

[0047] pVLF=(VLF / TP)×100%;

[0048] Wherein, pVLF is the peak frequency percentage of very low frequency; TP is the total peak frequency; VLF is the peak frequency in the very low frequency range;

[0049] The calculation formula of the peak frequency percentage of the low frequency is:

[0050] pLF=(LF / TP)×100%;

[0051] Where pLF is the peak frequency percentage of ultra-low frequency; TP is the total peak frequency; LF is the peak frequency in the low frequency range;

[0052] The calculation formula of the peak frequency percentage of the high frequency is:

[0053] pHF = (HF / TP) × 100%;

[0054] Wherein, pHF is the peak frequency percentage of the ultra-low frequency; TP is the total peak frequency; HF is the peak frequency in the high frequency range.

[0055] Furthermore, the step of calculating the preset feature weight matrix includes:

[0056] Obtain pulse signal data with real emotion labels of different test subjects, and perform time window division and feature extraction on the pulse signal data with real emotion labels to obtain the features of each time window of the pulse signal data with real emotion labels;

[0057] According to each feature of each time window of the pulse signal data with real emotion labels, a feature data set of the pulse signal with real emotion labels is constructed;

[0058] According to the feature data set of pulse signals with real emotion labels, the preset feature weight matrix objective function is iteratively solved to obtain the feature weight matrix when the objective function is minimized.

[0059] Furthermore, the calculation formula of the objective function of the preset feature weight matrix is:

[0060]

[0061] Where W is the feature weight matrix; x i is the feature set of the i-th time window in the feature dataset of the pulse signal with the real emotion label; y i is the true emotion label corresponding to the feature set of the i-th time window in the feature data set of the pulse signal with the true emotion label; s iis the similarity between the feature set of the i-th time window and the feature sets of all time windows; is the square of the 2-norm; || || 2，1 For L 2,1 norm regularization; μ is the control parameter.

[0062] Furthermore, the model training of the emotion recognition model includes:

[0063] Acquire historical pulse signal data with real emotion labels of the tested person, divide the historical pulse signal data with real emotion labels into time windows and extract features to obtain the features of each time window of the historical pulse signal data with real emotion labels;

[0064] According to each feature of each time window of the historical pulse signal data with the real emotion label, a feature data set of the historical pulse signal data with the real emotion label is constructed;

[0065] Calculate the product of a preset feature weight matrix and each integrated feature in the feature data set of the historical pulse signal data to obtain an optimized feature data set of the historical pulse signal data with a true emotion label;

[0066] Inputting the optimized feature data set of historical pulse signal data with real emotion labels into the emotion recognition model to be trained, so that the emotion recognition model performs emotion recognition classification and obtains predicted emotion recognition results;

[0067] According to the predicted emotion recognition results and the historical real emotion labels, the loss value is calculated through the loss function, and the model parameters of the emotion recognition model are optimized according to the loss value until the loss value converges to obtain a trained emotion recognition model.

[0068] Based on the above method embodiment, the present invention provides a corresponding system embodiment;

[0069] An embodiment of the present invention provides an emotion recognition system based on pulse signals, comprising: a data acquisition module, a data division module, a feature extraction module, a feature data set construction module, a feature data set optimization module and an emotion recognition module;

[0070] The data acquisition module is used to obtain the pulse signal data to be identified of the tested person;

[0071] The data division module is used to divide the pulse signal data to be identified into time windows to obtain pulse signal data of several time windows;

[0072] The feature extraction module is used to perform time domain feature extraction, frequency domain feature extraction and nonlinear feature extraction on the pulse signal data of each time window to obtain the features of each time window;

[0073] The feature data set construction module is used to construct a feature data set of the pulse signal according to each feature of each time window;

[0074] The feature data set optimization module is used to calculate the product of a preset feature weight matrix and each integrated feature in the feature data set to obtain an optimized feature data set;

[0075] The emotion recognition module is used to input the optimized feature data set into the trained emotion recognition model, so that the trained emotion recognition model performs emotion recognition classification according to the optimized feature data set to obtain the emotion recognition result.

[0076] Furthermore, the feature extraction module includes: a time domain feature extraction unit, a frequency domain feature extraction unit and a nonlinear feature extraction unit;

[0077] The time domain feature extraction unit is used to extract the time domain features of the pulse signal data of each time window to obtain the time domain features of the pulse signal data of each time window; wherein the time domain features include the mean value and standard deviation of all adjacent PP intervals, the root mean square and standard deviation of the differences between all adjacent PP intervals, and the number and percentage of the differences between all adjacent PP intervals greater than 50ms;

[0078] The frequency domain feature extraction unit is used to perform frequency domain transformation on the pulse signal data of each time window by fast Fourier transform to obtain the power spectrum density of the pulse signal data of each time window, extract the peak frequency of different frequency bands for the pulse signal data of each time window according to the power spectrum density of the pulse signal data of each time window, and generate the frequency domain features of the pulse signal data of each time window according to the peak frequencies of different frequency bands; wherein the frequency domain features include the peak frequencies in the ultra-low frequency range, the low frequency range and the high frequency range, the total peak frequency, the peak frequency ratio of the low frequency to the high frequency, the normalized value of the low frequency peak frequency, the normalized value of the high frequency peak frequency, the ultra-low frequency peak frequency percentage, the low frequency peak frequency percentage and the high frequency peak frequency percentage;

[0079] The nonlinear feature extraction unit is used to perform nonlinear feature extraction on the pulse signal data of each time window through Poincare map analysis, approximate entropy algorithm and sample entropy algorithm to obtain the Poincare map features, approximate entropy features and sample entropy features of the pulse signal data of each time window.

[0080] Based on the above-mentioned method embodiment, the present invention provides a corresponding terminal device embodiment, including: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements an emotion recognition method based on pulse signals as described in the present invention.

[0081] Based on the above-mentioned method embodiment, the present invention provides a corresponding computer-readable storage medium embodiment, including: a stored computer program, which controls the device where the computer-readable storage medium is located to execute an emotion recognition method based on pulse signals as described in the present invention when the computer program is running.

[0082] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0083] The present invention divides the acquired pulse signal data into time windows to obtain pulse signal data of several time windows, then extracts time domain features, frequency domain features, Poincare map features and nonlinear dynamic features from the pulse signal data of each time window, integrates the extracted pulse feature data to obtain an integrated feature data set, multiplies the integrated feature data set by a preset feature weight matrix to obtain an optimized feature data set, and then uses a trained emotion recognition model to perform emotion recognition classification according to the optimized feature data set to obtain an emotion recognition result, that is, the present invention extracts features from the pulse signal data for emotion recognition, combines the preset feature weight matrix obtained by iterative calculation of the pulse signal data with emotion labels, selects the optimized feature data set, and finally uses the trained emotion recognition model to perform emotion recognition classification according to the optimized feature data set to obtain an emotion recognition result identified by the pulse signal data, utilizes the characteristics that the subjective consciousness of the human body cannot control the pulse signal of an individual and that the pulse signal of an individual can reflect the individual's emotions, improves the accuracy of emotion recognition, and solves the problem that the existing technology using facial image recognition technology is prone to fail to accurately recognize personal emotions due to false facial images. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 : A flow chart of the steps of an emotion recognition method based on pulse signals provided by an embodiment of the present invention;

[0085] Figure 2 : A system structure diagram of an emotion recognition system based on pulse signals provided by an embodiment of the present invention;

[0086] Figure 3 , which is a Poincare scatter plot of an emotion recognition method based on pulse signals provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0087] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0088] In the description of the present invention, it should be understood that the terms “first” and “second” are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features.

[0089] Embodiment 1:

[0090] Reference Figure 1 , is a flowchart of a method for emotion recognition based on pulse signals provided by an embodiment of the present invention, the method comprising at least the following steps:

[0091] Step S1: obtaining the pulse signal data to be identified of the person under test, and dividing the pulse signal data to be identified into time windows to obtain pulse signal data of several time windows;

[0092] In this embodiment, smart heart rate monitoring devices such as smart bracelets and smart watches can be used, and pulse data can be monitored and recorded in real time through technologies such as photoplethysmography (PPG).

[0093] In this embodiment, before dividing the pulse signal data to be identified into time windows, it also includes: performing data preprocessing on the acquired pulse signal data to be identified; wherein the data preprocessing includes but is not limited to filtering processing, denoising processing and normalization processing; the filtering processing can be but is not limited to using a 0.6Hz-30Hz Butterworth filter for filtering processing.

[0094] In this embodiment, the pulse signal data to be identified is divided into time windows to obtain pulse signal data of several time windows, including: the pulse signal data to be identified can be divided into time windows according to a preset time window size, number of time windows and degree of overlap to obtain pulse signal data of several non-overlapping time windows; wherein the time window size and the number of time windows can be set by a technician according to actual conditions. In this embodiment, the time window size is set to 10s and the number of time windows is set to 3.

[0095] Step S2: performing time domain feature extraction, frequency domain feature extraction and nonlinear feature extraction on the pulse signal data of each time window to obtain the features of each time window;

[0096] In this embodiment, the pulse signal data of each time window is subjected to time domain feature extraction, frequency domain feature extraction and nonlinear feature extraction respectively, including:

[0097] Extracting time domain features from the pulse signal data of each time window to obtain time domain features of the pulse signal data of each time window; wherein the time domain features include the average value and standard deviation of all adjacent PP intervals, the root mean square and standard deviation of the differences between all adjacent PP intervals, and the number and percentage of the differences between all adjacent PP intervals that are greater than a preset threshold;

[0098] In this embodiment, time domain features are extracted for the pulse signal data of each time window, and six time domain features of the pulse signal data of each time window are calculated; wherein the six time domain features are respectively the mean value and standard deviation of all adjacent PP intervals, the root mean square and standard deviation of the differences between all adjacent PP intervals, and the number and percentage of the differences between all adjacent PP intervals that are greater than a preset threshold.

[0099] In this embodiment, the calculation formulas of the six time domain features are as follows:

[0100] The calculation formula for the average value of all adjacent PP intervals is:

[0101]

[0102] Among them, MEAN PP is the average value of all adjacent PP intervals; PP i is the time interval between the i-th two adjacent pulse wave peaks in each time window; n is the total number of PP intervals in each time window;

[0103] The calculation formula for the standard deviation of all adjacent PP intervals is:

[0104]

[0105] Among them, SDNN PP is the standard deviation of all adjacent PP intervals; MEABN PP is the average value of all adjacent PP intervals; PP i is the time interval between the i-th two adjacent pulse wave peaks in each time window; n is the total number of PP intervals in each time window;

[0106] The calculation formula for the root mean square of the differences between all adjacent PP intervals is:

[0107]

[0108] Among them, RMSSD PPis the root mean square of the differences between all adjacent PP intervals; PP i is the time interval between the i-th two adjacent pulse wave peaks in each time window; PP i-1 is the time interval between the i-1th two adjacent pulse wave peaks in each time window; n is the total number of PP intervals in each time window;

[0109] The calculation formula for the standard deviation of the differences between all adjacent PP intervals is:

[0110]

[0111] Among them, SDSD PP RMSSD is the standard deviation of the differences between all adjacent PP intervals. PP is the root mean square of the differences between all adjacent PP intervals; PP i is the time interval between the i-th two adjacent pulse wave peaks in each time window; PP i-1 is the time interval between the i-1th two adjacent pulse wave peaks in each time window; n is the total number of PP intervals in each time window;

[0112] The calculation formula for the number of differences between all adjacent PP intervals that are greater than a preset threshold is:

[0113]

[0114] Among them, NN(A) is the number of all adjacent PP intervals with a difference greater than 50 ms; PP i is the time interval between the i-th two adjacent pulse wave peaks in each time window; PP i+1 is the time interval between the i+1th two adjacent pulse wave peaks in each time window; N is the total number of PP intervals in each time window; A is a preset threshold value, which can be set by a technician according to actual conditions. In this embodiment, A can be set to 50;

[0115] The calculation formula for the percentage of the number of differences between all adjacent PP intervals that is greater than the preset threshold is:

[0116]

[0117] Among them, pNN(A) is the percentage of all adjacent PP intervals with a difference greater than 50 ms; PP i is the time interval between the i-th two adjacent pulse wave peaks in each time window; PP i+1is the time interval between the i+1th two adjacent pulse wave peaks in each time window; N is the total number of PP intervals in each time window; A is a preset threshold, which can be set by technicians according to actual conditions. In this embodiment, A can be set to 50.

[0118] Performing frequency domain transformation on the pulse signal data of each time window by fast Fourier transform to obtain the power spectrum density of the pulse signal data of each time window;

[0119] According to the power spectrum density of the pulse signal data in each time window, the peak frequencies of different frequency bands are extracted for the pulse signal data in each time window, and the frequency domain features of the pulse signal data in each time window are generated according to the peak frequencies of different frequency bands; wherein the frequency domain features include the peak frequencies in the ultra-low frequency range, the low frequency range and the high frequency range, the total peak frequency, the peak frequency ratio of the low frequency to the high frequency, the normalized value of the low frequency peak frequency, the normalized value of the high frequency peak frequency, the peak frequency percentage of the ultra-low frequency, the peak frequency percentage of the low frequency and the peak frequency percentage of the high frequency;

[0120] In this embodiment, the range thresholds of the ultra-low frequency range, low frequency range and high frequency range can be set by technical personnel according to actual conditions. In this embodiment, the range threshold of the ultra-low frequency range is 0.003Hz-0.04Hz; the range threshold of the low frequency range is 0.04Hz-0.15Hz; the range threshold of the high frequency range is 0.15Hz-0.4Hz.

[0121] In this embodiment, 10 frequency domain features of the pulse signal data in each time window are generated according to the peak frequencies of different frequency bands; wherein the 10 frequency domain features are respectively the peak frequency in the ultra-low frequency range, the peak frequency in the low frequency range, the peak frequency in the high frequency range, the total peak frequency, the peak frequency ratio of the low frequency to the high frequency, the normalized value of the low frequency peak frequency, the normalized value of the high frequency peak frequency, the peak frequency percentage of the ultra-low frequency, the peak frequency percentage of the low frequency and the peak frequency percentage of the high frequency.

[0122] In this embodiment, the calculation formula of the total peak frequency is:

[0123] TP=VLF+LF+HF;

[0124] Wherein, TP is the total peak frequency; VLF is the peak frequency in the ultra-low frequency range; :F is the peak frequency in the low frequency range; HF is the peak frequency in the high frequency range;

[0125] The calculation formula of the peak frequency ratio of the low frequency to the high frequency is:

[0126] Ratio_LH = LF / HF;

[0127] Among them, Ratio_LH is the peak frequency ratio of low frequency to high frequency; LF is the peak frequency in the low frequency range; HF is the peak frequency in the high frequency range;

[0128] The calculation formula of the normalized value of the low-frequency peak frequency is:

[0129] LF norm =LF / (LF+HF);

[0130] Among them, LF norm is the normalized value of the low-frequency peak frequency; LF is the peak frequency in the low-frequency range; HF is the peak frequency in the high-frequency range;

[0131] The calculation formula of the normalized value of the high frequency peak frequency is:

[0132] HF norm =HF / (LF+HF);

[0133] Among them, HF norm is the normalized value of the high-frequency peak frequency; LF is the peak frequency in the low-frequency range; HF is the peak frequency in the high-frequency range;

[0134] The calculation formula of the peak frequency percentage of the ultra-low frequency is:

[0135] pVLF=(VLF / TP)×100%;

[0136] Wherein, pVLF is the peak frequency percentage of very low frequency; TP is the total peak frequency; VLF is the peak frequency in the very low frequency range;

[0137] The calculation formula of the peak frequency percentage of the low frequency is:

[0138] pLF=(LF / TP)×100%;

[0139] Where pLF is the peak frequency percentage of ultra-low frequency; TP is the total peak frequency; LF is the peak frequency in the low frequency range;

[0140] The calculation formula of the peak frequency percentage of the high frequency is:

[0141] pHF = (HF / TP) × 100%;

[0142] Wherein, pHF is the peak frequency percentage of the ultra-low frequency; TP is the total peak frequency; HF is the peak frequency in the high frequency range.

[0143] The nonlinear features of the pulse signal data in each time window are extracted by Poincare diagram analysis, approximate entropy algorithm and sample entropy algorithm to obtain the Poincare diagram features, approximate entropy features and sample entropy features of the pulse signal data in each time window.

[0144] In this embodiment, refer to Figure 3 , which is a Poincare scatter plot of an emotion recognition method based on a pulse signal provided by an embodiment of the present invention, and nonlinear feature extraction is performed on the pulse signal data of each time window by using a Poincare plot analysis method, an approximate entropy algorithm, and a sample entropy algorithm, including:

[0145] A PP interval sequence is constructed for the pulse signal data of each time window, and a scatter plot is constructed by a Poincare plot analysis method, and the scatter plot is analyzed and calculated to obtain 12 Poincare plot features; wherein the 12 Poincare plot features include the standard deviation of the short axis of the scatter plot, the standard deviation of the long axis of the scatter plot, the ratio of the standard deviation of the short axis to the standard deviation of the long axis, the minimum area of ​​the scatter plot track, the maximum area of ​​the scatter plot track, the average area of ​​the scatter plot track, the average of the standard deviation of the horizontal coordinate of the scatter plot track, the average of the standard deviation of the vertical coordinate of the scatter plot track, the average of the third moment of the horizontal coordinate of the scatter plot track, the average of the fourth moment of the horizontal coordinate of the scatter plot track, and the average of the fourth moment of the total coordinate of the scatter plot track;

[0146] In this embodiment, the calculation formula of the minor axis standard deviation of the scatter plot is:

[0147]

[0148] Among them, SD1 is the standard deviation of the short axis of the scatter plot; SDSD PP is the standard deviation of the differences between all adjacent PP intervals;

[0149] In this embodiment, the calculation formula of the long axis standard deviation of the scatter plot is:

[0150]

[0151] Among them, SD2 is the standard deviation of the long axis of the scatter plot; SDSD PP is the standard deviation of the differences between all adjacent PP intervals; SDNN PP is the standard deviation of all adjacent PP intervals;

[0152] In this embodiment, the calculation formula of the ratio of the minor axis standard deviation to the major axis standard deviation is:

[0153]

[0154] Among them, SD 12 is the ratio of the minor axis standard deviation to the major axis standard deviation;

[0155] In this embodiment, the calculation formula of the horizontal coordinate standard deviation of the scatter plot trajectory is: Among them, σ x is the horizontal coordinate standard deviation of the scatter plot trajectory; i is the ith pulse signal data in the scatter plot; a is the mean value of the horizontal coordinate of the scatter plot track; n is the number of pulse signal data in the time window; c is the preset delay times, which can be set by the technician according to the actual situation. In this embodiment, c is set to 4;

[0156] In this embodiment, the calculation formula of the vertical coordinate standard deviation of the scatter plot trajectory is: Among them, σ y is the standard deviation of the vertical coordinate of the scatter plot trajectory; x i is the ith pulse signal data in the scatter plot; b is the mean value of the vertical coordinate of the scatter plot track; n is the number of pulse signal data in the time window; c is the preset delay times, which can be set by the technician according to the actual situation. In this embodiment, c is set to 4;

[0157] In this embodiment, the calculation formula of the cubic moment of the horizontal coordinate of the scatter plot trajectory is: in, is the third moment of the horizontal coordinate of the scatter plot trajectory; i is the ith pulse signal data in the scatter plot; a is the mean value of the horizontal coordinate of the scatter plot track; n is the number of pulse signal data in the time window; c is the preset delay times, which can be set by the technician according to the actual situation. In this embodiment, c is set to 4;

[0158] In this embodiment, the calculation formula of the ordinate cubic moment of the scatter plot trajectory is: in, is the vertical coordinate cubic moment of the scatter plot trajectory; x i is the ith pulse signal data in the scatter plot; b is the mean value of the vertical coordinate of the scatter plot track; n is the number of pulse signal data in the time window; c is the preset delay times, which can be set by the technician according to the actual situation. In this embodiment, c is set to 4;

[0159] In this embodiment, the calculation formula of the fourth moment of the horizontal coordinate of the scatter plot trajectory is: in, is the fourth moment of the horizontal coordinate of the scatter plot trajectory; i is the ith pulse signal data in the scatter plot; a is the mean value of the horizontal coordinate of the scatter plot track; n is the number of pulse signal data in the time window; c is the preset delay times, which can be set by the technician according to the actual situation. In this embodiment, c is set to 4;

[0160] In this embodiment, the calculation formula of the fourth moment of the ordinate of the scatter plot trajectory is: in, is the fourth moment of the ordinate of the scatter plot trajectory; x i is the i-th pulse signal data in the scatter plot; b is the mean of the vertical coordinate of the scatter plot trajectory; n is the number of pulse signal data in the time window; c is the preset delay times, which can be set by the technician according to the actual situation. In this embodiment, c is set to 4.

[0161] In this embodiment, the calculation formulas for the minimum area value of the scatter plot track, the maximum area value of the scatter plot track, the average area value of the scatter plot track, the average value of the standard deviation of the horizontal coordinate of the scatter plot track, the average value of the standard deviation of the vertical coordinate of the scatter plot track, the average value of the third moment of the horizontal coordinate of the scatter plot track, the average value of the third moment of the vertical coordinate of the scatter plot track, the average value of the fourth moment of the horizontal coordinate of the scatter plot track, and the average value of the fourth moment of the total coordinate of the scatter plot track are shown in Table 1 below:

[0162]

[0163] Table 1

[0164] According to the pulse signal data of each time window, the approximate entropy and sample entropy of the pulse signal data of each time window are calculated by the approximate entropy algorithm and the sample entropy algorithm to obtain the approximate entropy characteristics and sample entropy characteristics of the pulse signal data of each time window.

[0165] In this embodiment, the steps for calculating the approximate entropy are:

[0166] According to the pulse signal data of each time window, several PP intervals are determined and constructed into a PP interval sequence {pp(1), pp(2), ..., pp(N)} of length N, and the PP interval sequence of the m-dimensional reconstruction vector is calculated: pp(i) = [pp(i), pp(i+1), ..., pp(i+m-1)], i = 1, 2, ..., N-m+1;

[0167] By distance formula d[pp(i), pp(j)]=max k＝0，1，…，m-1 (|pp(i+k)-pp(j+k)|) calculates the distance between two PP interval vectors;

[0168] Given a similarity threshold r, for each i value, the ratio of the number of distances between pp(i) and pp(j) less than r to the total number is defined as The calculation formula is:

[0169]

[0170] Where i = 1, 2, 3, ..., Nm, is a step function, 1 when the parameter is greater than or equal to zero, otherwise 0;

[0171] For all values ​​of i, calculate The average value, The formula for calculating the average value of is:

[0172]

[0173] according to The approximate entropy is calculated by averaging the values ​​of ; wherein the calculation formula of the approximate entropy is:

[0174] ApEn(m, r, N) = φ m (r)-φ m+1 (r);

[0175] In this embodiment, m is set to 2, which can be adjusted according to actual conditions; r=0.2*SD, SD is the preset standard deviation of the original time series, which can be adjusted according to actual conditions.

[0176] In this embodiment, the steps for calculating sample entropy are:

[0177] According to the pulse signal data of each time window, several PP intervals are determined and constructed as a PP interval sequence {pp(1), pp(2), ..., pp(N)} of length N, and the PP interval sequence of the m-dimensional reconstruction vector is calculated: pp(i) = [pp(i), pp(i+1), ..., pp(i+m-1)], i = 1, 2, ..., N-m+1;

[0178] By distance formula d[pp(i), pp(j)]=max k＝0，1，...，m-1 (|pp(i+k)-pp(j+k)|) calculates the distance between two PP interval vectors;

[0179] Given a similarity threshold r, for each i value, the ratio of the number of distances between pp(i) and pp(j) less than r to the total number is defined as The calculation formula is:

[0180]

[0181] Where, i≠j, i=1,2,3,…,Nm, is a step function, 1 when the parameter is greater than or equal to zero, otherwise 0;

[0182] For all values ​​of i, calculate The average value, The formula for calculating the average value of is:

[0183]

[0184] according to The sample entropy is calculated by averaging the values ​​of ; wherein the sample entropy is calculated as follows:

[0185]

[0186] In this embodiment, m is set to 2, which can be adjusted according to actual conditions; r=0.2*SD, SD is the preset standard deviation of the original time series, which can be adjusted according to actual conditions.

[0187] Step S3: constructing a feature data set of the pulse signal data according to each feature of each time window;

[0188] Step S4: Calculate the product of a preset feature weight matrix and each integrated feature in the feature data set to obtain an optimized feature data set;

[0189] In this embodiment, the step of calculating the preset feature weight matrix includes:

[0190] Obtain pulse signal data with real emotion labels of different test subjects, and perform time window division and feature extraction on the pulse signal data with real emotion labels to obtain the features of each time window of the pulse signal data with real emotion labels;

[0191] According to each feature of each time window of the pulse signal data with real emotion labels, a feature data set of the pulse signal with real emotion labels is constructed;

[0192] According to the feature data set of pulse signals with real emotion labels, the preset feature weight matrix objective function is iteratively solved to obtain the feature weight matrix when the objective function is minimized.

[0193] In this embodiment, the calculation formula of the objective function of the preset feature weight matrix is:

[0194]

[0195] Where W is the feature weight matrix; x i is the feature set of the i-th time window in the feature dataset of the pulse signal with the real emotion label; y i is the true emotion label corresponding to the feature set of the i-th time window in the feature data set of the pulse signal with the true emotion label; s i is the similarity between the feature set of the i-th time window and the feature sets of all time windows; is the square of the 2-norm; || ||2，1 For L 2,1 norm regularization; μ is the control parameter.

[0196] Exemplarily, the calculation steps of the preset feature weight matrix are:

[0197] Step A1: obtaining pulse signal data with real emotion labels of different test subjects, and performing time window division and feature extraction on the pulse signal data with real emotion labels to obtain the features of each time window of the pulse signal data with real emotion labels;

[0198] Step A2: constructing a feature data set of the pulse signal with the real emotion label according to each feature of each time window of the pulse signal data with the real emotion label;

[0199] Step A3: Define the feature dataset of the pulse signal with the real emotion label as X = {x 1 , …, x n}∈R n ×d , x i ∈R d represents the i-th feature data, where d is the dimension of the feature and n is the total number of feature data in the training set; define the true emotion label matrix Y = {y 1 ,…,y n} T ∈{0,1} n×c , where c is the number of true sentiment labels and y i ∈R c is the i-th true emotion label vector; define the feature weight matrix W∈R d×c , the role of the feature weight matrix W is to map the data matrix X to the label matrix Y.

[0200] Step A4: Construct the objective function and solve W; the initial objective function is set as:

[0201] Specifically, for the first item V, the least squares loss function is chosen because of its simplicity and efficiency shown in many studies. The basic idea is to describe the relationship between the independent variable and the dependent variable by fitting a linear model. In this process, the least squares method uses an objective function, namely the residual sum of squares, which represents the difference between the model prediction value and the actual observation value. By minimizing this objective function, the optimal model parameters can be obtained, so that the model can better predict the results of unknown data; the specific expression is:

[0202]

[0203] Among them, s iRepresents sample x i Similarity with all samples, W∈R d×c is the regression coefficient vector of the training set X, which represents the weight corresponding to each feature, that is, the change in the dependent variable when the independent variable changes by one unit. By fitting the training data set, the optimal regression coefficient vector can be obtained, thereby establishing a model for predicting unknown data;

[0204] For the second term Ω, use the L-based 2,1 norm regularization, because it is robust to outliers, can better handle outliers and noise, and can effectively reveal the discriminative features between multiple labels; specifically, L 2,1 The norm penalizes the coefficients of each feature group rather than the coefficients of a single feature. This means that when some features in a feature group play an important role in the classification task, the coefficients of the feature group will be retained or amplified, while the coefficients of other feature groups may be compressed or even zeroed. This mechanism makes the model more inclined to select feature groups with significant discrimination, thereby improving the classification performance and interpretability of the model; the specific expression is

[0205] Ω(W)=‖W‖ 2,1 ;

[0206] In summary, the final objective function is obtained:

[0207]

[0208] Step A5: Since the objective function is non-convex and difficult to solve, the following iterative method is used to solve W:

[0209] First, define the matrix S, s ij Represents x i and x j The final objective function can be written as follows:

[0210]

[0211] After the deformation, the partial derivative of the final objective function with respect to W is obtained, that is, get:

[0212]

[0213] X T SXW-X T SY+μDW=0

[0214] X T SXW+μDW=X T SY

[0215] (XT SX+μD)W=X T SY;

[0216] Where D is a diagonal matrix defined as:

[0217]

[0218] Among them, w d is the dth column of the feature weight matrix W;

[0219] Use the EM-style iterative method to solve W: First, use the randomly initialized W to get D, then: W = (X T SX+μD) -1 X T SY; then update W according to this, and iterate repeatedly until the objective function converges.

[0220] Step S5: inputting the optimized feature data set into the trained emotion recognition model, so that the trained emotion recognition model performs emotion recognition classification according to the optimized feature data set to obtain an emotion recognition result.

[0221] In this embodiment, the model training of the emotion recognition model includes:

[0222] Acquire historical pulse signal data with real emotion labels of the tested person, divide the historical pulse signal data with real emotion labels into time windows and extract features to obtain the features of each time window of the historical pulse signal data with real emotion labels;

[0223] According to each feature of each time window of the historical pulse signal data with the real emotion label, a feature data set of the historical pulse signal data with the real emotion label is constructed;

[0224] Calculate the product of a preset feature weight matrix and each integrated feature in the feature data set of the historical pulse signal data to obtain an optimized feature data set of the historical pulse signal data with a true emotion label;

[0225] Inputting the optimized feature data set of historical pulse signal data with real emotion labels into the emotion recognition model to be trained, so that the emotion recognition model performs emotion recognition classification and obtains predicted emotion recognition results;

[0226] According to the predicted emotion recognition results and the historical real emotion labels, the loss value is calculated through the loss function, and the model parameters of the emotion recognition model are optimized according to the loss value until the loss value converges to obtain a trained emotion recognition model.

[0227] In this embodiment, the model structure of the emotion recognition model includes but is not limited to an SVM classifier.

[0228] In this embodiment, the loss function includes but is not limited to a mean square error function and a mean absolute error function.

[0229] In this embodiment, the model parameter optimization methods include but are not limited to stochastic gradient descent optimization algorithm, small batch gradient descent algorithm and Adam algorithm.

[0230] Embodiment 2:

[0231] Reference Figure 2 , is a system structure diagram of an emotion recognition system based on pulse signals provided by an embodiment of the present invention, the system at least comprising: a data acquisition module, a data partitioning module, a feature extraction module, a feature data set construction module, a feature data set optimization module and an emotion recognition module;

[0232] The data acquisition module is used to obtain the pulse signal data to be identified of the tested person;

[0233] The data division module is used to divide the pulse signal data to be identified into time windows to obtain pulse signal data of several time windows;

[0234] The feature extraction module is used to perform time domain feature extraction, frequency domain feature extraction and nonlinear feature extraction on the pulse signal data of each time window to obtain the features of each time window;

[0235] The feature data set construction module is used to construct a feature data set of the pulse signal according to each feature of each time window;

[0236] The feature data set optimization module is used to calculate the product of a preset feature weight matrix and each integrated feature in the feature data set to obtain an optimized feature data set;

[0237] The emotion recognition module is used to input the optimized feature data set into the trained emotion recognition model, so that the trained emotion recognition model performs emotion recognition classification according to the optimized feature data set to obtain the emotion recognition result.

[0238] In this embodiment, the feature extraction module includes: a time domain feature extraction unit, a frequency domain feature extraction unit and a nonlinear feature extraction unit;

[0239] The time domain feature extraction unit is used to extract the time domain features of the pulse signal data of each time window to obtain the time domain features of the pulse signal data of each time window; wherein the time domain features include the mean value and standard deviation of all adjacent PP intervals, the root mean square and standard deviation of the differences between all adjacent PP intervals, and the number and percentage of the differences between all adjacent PP intervals greater than 50ms;

[0240] The frequency domain feature extraction unit is used to perform frequency domain transformation on the pulse signal data of each time window by fast Fourier transform to obtain the power spectrum density of the pulse signal data of each time window, extract the peak frequency of different frequency bands for the pulse signal data of each time window according to the power spectrum density of the pulse signal data of each time window, and generate the frequency domain features of the pulse signal data of each time window according to the peak frequencies of different frequency bands; wherein the frequency domain features include the peak frequencies in the ultra-low frequency range, the low frequency range and the high frequency range, the total peak frequency, the peak frequency ratio of the low frequency to the high frequency, the normalized value of the low frequency peak frequency, the normalized value of the high frequency peak frequency, the ultra-low frequency peak frequency percentage, the low frequency peak frequency percentage and the high frequency peak frequency percentage;

[0241] The nonlinear feature extraction unit is used to perform nonlinear feature extraction on the pulse signal data of each time window through Poincare map analysis, approximate entropy algorithm and sample entropy algorithm to obtain the Poincare map features, approximate entropy features and sample entropy features of the pulse signal data of each time window.

[0242] Based on the above method embodiment, another embodiment is provided;

[0243] Another embodiment of the present invention provides an emotion recognition terminal device based on pulse signals, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements an emotion recognition method based on pulse signals as described in any one of the above method embodiments of the present invention.

[0244] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program in the pulse signal-based emotion recognition terminal device.

[0245] The pulse signal-based emotion recognition terminal device may be a computing device such as a desktop computer, a notebook, a PDA, and a cloud server. The pulse signal-based emotion recognition terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art may understand that, for example, the pulse signal-based emotion recognition terminal device may also include input and output devices, network access devices, buses, etc.

[0246] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the pulse signal-based emotion recognition terminal device, and uses various interfaces and lines to connect the various parts of the pulse signal-based emotion recognition terminal device.

[0247] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the pulse signal-based emotion recognition terminal device by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0248] Based on the above method embodiment, another embodiment is provided;

[0249] A storage medium provided in another embodiment of the present invention includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute an emotion recognition method based on a pulse signal as described in any one of the above method embodiments of the present invention.

[0250] Wherein, the above-mentioned storage medium is a computer-readable storage medium. The module / unit integrated in the emotion recognition system / terminal device based on pulse signal can be stored in a computer-readable storage medium if it is implemented in the form of a software functional unit and sold or used as an independent product. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned method embodiments can be implemented. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can include: any entity or device that can carry the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0251] It should be noted that the above-mentioned terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art will understand that the above-mentioned terminal device is merely an example and does not constitute a limitation on the terminal device. It may include more or fewer components, or a combination of certain components, or different components.

[0252] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. An emotion recognition method based on pulse signal, characterized in that: include: Obtaining the pulse signal data to be identified of the tested person, and dividing the pulse signal data to be identified into time windows to obtain pulse signal data of several time windows; Perform time domain feature extraction, frequency domain feature extraction and nonlinear feature extraction on the pulse signal data of each time window to obtain the features of each time window; According to each feature of each time window, a feature data set of the pulse signal data is constructed; Calculating the product of a preset feature weight matrix and each integrated feature in the feature data set to obtain an optimized feature data set; The optimized feature data set is input into the trained emotion recognition model, so that the trained emotion recognition model performs emotion recognition classification according to the optimized feature data set to obtain the emotion recognition result.

2. The method for emotion recognition based on pulse signal according to claim 1, characterized in that: The pulse signal data of each time window is subjected to time domain feature extraction, frequency domain feature extraction and nonlinear feature extraction respectively, including: Extracting time domain features from the pulse signal data of each time window to obtain time domain features of the pulse signal data of each time window; wherein the time domain features include the average value and standard deviation of all adjacent PP intervals, the root mean square and standard deviation of the differences between all adjacent PP intervals, and the number and percentage of the differences between all adjacent PP intervals that are greater than a preset threshold; Performing frequency domain transformation on the pulse signal data of each time window by fast Fourier transform to obtain the power spectrum density of the pulse signal data of each time window; According to the power spectrum density of the pulse signal data in each time window, the peak frequencies of different frequency bands are extracted for the pulse signal data in each time window, and the frequency domain features of the pulse signal data in each time window are generated according to the peak frequencies of different frequency bands; wherein the frequency domain features include the peak frequencies in the ultra-low frequency range, the low frequency range and the high frequency range, the total peak frequency, the peak frequency ratio of the low frequency to the high frequency, the normalized value of the low frequency peak frequency, the normalized value of the high frequency peak frequency, the peak frequency percentage of the ultra-low frequency, the peak frequency percentage of the low frequency and the peak frequency percentage of the high frequency; The nonlinear features of the pulse signal data in each time window are extracted by Poincare diagram analysis, approximate entropy algorithm and sample entropy algorithm to obtain the Poincare diagram features, approximate entropy features and sample entropy features of the pulse signal data in each time window.

3. The method for emotion recognition based on pulse signal according to claim 2, characterized in that: The calculation formula for the average value of all adjacent PP intervals is: Among them, MEAN PP is the average value of all adjacent PP intervals; PP i is the time interval between the i-th two adjacent pulse wave peaks in each time window; n is the total number of PP intervals in each time window; The calculation formula for the standard deviation of all adjacent PP intervals is: Among them, SDNN PP is the standard deviation of all adjacent PP intervals; MEAN PP is the average value of all adjacent PP intervals; PP i is the time interval between the i-th two adjacent pulse wave peaks in each time window; n is the total number of PP intervals in each time window; The calculation formula for the root mean square of the differences between all adjacent PP intervals is: Among them, RMSSD PP is the root mean square of the differences between all adjacent PP intervals; PP i is the time interval between the i-th two adjacent pulse wave peaks in each time window; PP i-1 is the time interval between the i-1th two adjacent pulse wave peaks in each time window; n is the total number of PP intervals in each time window; The calculation formula for the standard deviation of the differences between all adjacent PP intervals is: Among them, SDSD PP RMSSD is the standard deviation of the differences between all adjacent PP intervals. PP is the root mean square of the differences between all adjacent PP intervals; PP i is the time interval between the i-th two adjacent pulse wave peaks in each time window; PP i-1 is the time interval between the i-1th two adjacent pulse wave peaks in each time window; n is the total number of PP intervals in each time window; The calculation formula for the number of differences between all adjacent PP intervals that are greater than a preset threshold is: Among them, NN(A) is the number of all adjacent PP intervals with a difference greater than 50 ms; PP i is the time interval between the i-th two adjacent pulse wave peaks in each time window; PP i+1 is the time interval between the i+1th two adjacent pulse wave peaks in each time window; N is the total number of PP intervals in each time window; A is the preset threshold; The calculation formula for the percentage of the number of differences between all adjacent PP intervals that is greater than the preset threshold is: Among them, pNN(A) is the percentage of all adjacent PP intervals with a difference greater than 50 ms; PP i is the time interval between the i-th two adjacent pulse wave peaks in each time window; PP i+1 is the time interval between the i+1th two adjacent pulse wave peaks in each time window; N is the total number of PP intervals in each time window; A is the preset threshold.

4. The method for emotion recognition based on pulse signal according to claim 3, characterized in that: The calculation formula of the total peak frequency is: TP=VLF+LF+HF; Among them, TP is the total peak frequency; VLF is the peak frequency in the very low frequency range; LF is the peak frequency in the low frequency range; HF is the peak frequency in the high frequency range; The calculation formula of the peak frequency ratio of the low frequency to the high frequency is: Ratio_LH = LF / HF; Among them, Ratio_LH is the peak frequency ratio of low frequency to high frequency; LF is the peak frequency in the low frequency range; HF is the peak frequency in the high frequency range; The calculation formula of the normalized value of the low-frequency peak frequency is: LF nom =LF / (LF+HF): Among them, LF norm is the normalized value of the low-frequency peak frequency; LF is the peak frequency in the low-frequency range; HF is the peak frequency in the high-frequency range; The calculation formula of the normalized value of the high frequency peak frequency is: HF norn =HF / (LF+HF); Among them, HF norm is the normalized value of the high-frequency peak frequency; LF is the peak frequency in the low-frequency range; HF is the peak frequency in the high-frequency range; The calculation formula of the peak frequency percentage of the ultra-low frequency is: pVLF=(VLF / TP)×100%; Wherein, pVLF is the peak frequency percentage of very low frequency; TP is the total peak frequency; VLF is the peak frequency in the very low frequency range; The calculation formula of the peak frequency percentage of the low frequency is: pLF=(LF / TP)×100%; Where pLF is the peak frequency percentage of ultra-low frequency; TP is the total peak frequency; LF is the peak frequency in the low frequency range; The calculation formula of the peak frequency percentage of the high frequency is: pHF = (HF / TP) × 100%; Wherein, pHF is the peak frequency percentage of the ultra-low frequency; TP is the total peak frequency; HF is the peak frequency in the high frequency range.

5. The method for emotion recognition based on pulse signal according to claim 4, characterized in that: The step of calculating the preset feature weight matrix includes: Obtain pulse signal data with real emotion labels of different test subjects, and perform time window division and feature extraction on the pulse signal data with real emotion labels to obtain the features of each time window of the pulse signal data with real emotion labels; According to each feature of each time window of the pulse signal data with real emotion labels, a feature data set of the pulse signal with real emotion labels is constructed; According to the feature data set of pulse signals with real emotion labels, the preset feature weight matrix objective function is iteratively solved to obtain the feature weight matrix when the objective function is minimized.

6. The method for emotion recognition based on pulse signal according to claim 5, characterized in that: The calculation formula of the objective function of the preset feature weight matrix is: Where W is the feature weight matrix; x i is the feature set of the i-th time window in the feature dataset of the pulse signal with the real emotion label; y i is the true emotion label corresponding to the feature set of the i-th time window in the feature data set of the pulse signal with the true emotion label; s i is the similarity between the feature set of the i-th time window and the feature sets of all time windows; is the square of the 2-norm; || || 2,1 For L 2,1 norm regularization; μ is the control parameter.

7. The method for emotion recognition based on pulse signal according to claim 6, characterized in that: The model training of the emotion recognition model includes: Acquire historical pulse signal data with real emotion labels of the tested person, divide the historical pulse signal data with real emotion labels into time windows and extract features to obtain the features of each time window of the historical pulse signal data with real emotion labels; According to each feature of each time window of the historical pulse signal data with the real emotion label, a feature data set of the historical pulse signal data with the real emotion label is constructed; Calculate the product of a preset feature weight matrix and each integrated feature in the feature data set of the historical pulse signal data to obtain an optimized feature data set of the historical pulse signal data with a true emotion label; Inputting the optimized feature data set of historical pulse signal data with real emotion labels into the emotion recognition model to be trained, so that the emotion recognition model performs emotion recognition classification and obtains predicted emotion recognition results; According to the predicted emotion recognition results and the historical real emotion labels, the loss value is calculated through the loss function, and the model parameters of the emotion recognition model are optimized according to the loss value until the loss value converges to obtain a trained emotion recognition model.

8. An emotion recognition system based on pulse signal, characterized in that: include: Data collection module, data partitioning module, feature extraction module, feature data set construction module, feature data set optimization module and emotion recognition module; The data acquisition module is used to obtain the pulse signal data to be identified of the tested person; The data division module is used to divide the pulse signal data to be identified into time windows to obtain pulse signal data of several time windows; The feature extraction module is used to perform time domain feature extraction, frequency domain feature extraction and nonlinear feature extraction on the pulse signal data of each time window to obtain the features of each time window; The feature data set construction module is used to construct a feature data set of the pulse signal according to each feature of each time window; The feature data set optimization module is used to calculate the product of a preset feature weight matrix and each integrated feature in the feature data set to obtain an optimized feature data set; The emotion recognition module is used to input the optimized feature data set into the trained emotion recognition model, so that the trained emotion recognition model performs emotion recognition classification according to the optimized feature data set to obtain the emotion recognition result.

9. A pulse signal-based emotion recognition terminal device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements a pulse signal-based emotion recognition method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that: The storage medium includes a stored computer program, wherein when the computer program is executed, the device where the storage medium is located is controlled to execute the emotion recognition method based on pulse signals as described in any one of claims 1 to 7.

Images