Physiological signal individual difference removal method and device, edge computing equipment and medium

Abstract

The invention provides a physiological signal individual difference removal method and device, edge computing equipment and a storage medium, and the method comprises the steps: obtaining physiological signals of a plurality of subjects, the physiological signals being signals collected by physiological signal collection equipment when the plurality of subjects execute the same task; extracting frequency domain characteristics based on the physiological signal, wherein the frequency domain characteristics are used for representing frequency characteristics of the physiological signal; extracting cepstrum features based on the frequency domain features; and performing normalization processing on the cepstrum features to obtain the cepstrum features of the physiological signals from which at least part of individual differences of the multiple subjects are removed. According to the method provided by the invention, the normalization of the physiological signals is facilitated, the individual difference of the physiological signals is removed, the data distribution of the physiological signals does not need to be limited, and the application range is wider.

Description

Technical Field

[0001] The present application relates to the technical field of physiological signal processing, and in particular to a method, apparatus, edge computing device and storage medium for removing individual differences in physiological signals. Background Art

[0002] Physiological signals are naturally generated by the human body. The range and baseline of each person's physiological signals are different. If you want to extract a common property from multiple individuals, you need to remove individual differences. One way to remove individual differences is to normalize the data, that is, unify the signals of different individuals into one unit of measurement before conducting research.

[0003] Related technologies include min-max normalization and Z-Score normalization for normalizing physiological signals. However, the min-max normalization method is more suitable for situations where data distribution is relatively concentrated. When new data is added, max and min may need to be redefined. The Z-Score normalization method requires the assumption that the data distribution follows a normal distribution. Summary of the invention

[0004] The present application provides a method, apparatus, edge computing device and storage medium for removing individual differences in physiological signals, which helps to normalize physiological signals and remove individual differences in physiological signals without restricting the data distribution of physiological signals, and has a wider scope of application.

[0005] In a first aspect, the present application provides a method for removing individual differences in physiological signals, comprising:

[0006] Acquiring physiological signals of a plurality of subjects, wherein the physiological signals are signals acquired by a physiological signal acquisition device when the plurality of subjects perform the same task;

[0007] Extracting frequency domain features based on the physiological signal, wherein the frequency domain features are used to characterize the frequency characteristics of the physiological signal;

[0008] Extracting cepstrum features based on the frequency domain features;

[0009] The cepstrum features are normalized to obtain cepstrum features of the physiological signal from which at least part of the individual differences of the multiple subjects are removed.

[0010] In one possible implementation, the frequency domain feature includes a power spectral density of the physiological signal, and extracting the frequency domain feature based on the physiological signal includes:

[0011] Dividing the physiological signal into signals of a plurality of windows by using a sliding window method;

[0012] Extract the autocorrelation function of the signal in each window;

[0013] Performing a fast Fourier transform on the autocorrelation function to obtain a power spectrum density of the physiological signal.

[0014] In one possible implementation, extracting the cepstrum feature based on the frequency domain feature includes:

[0015] The power spectrum density of the physiological signal is logarithmically taken, and then inverse Fourier transform is performed to obtain the cepstrum feature.

[0016] In one possible implementation, normalizing the cepstral features includes: normalizing the cepstral features using a cepstral mean normalization model, wherein a training method for the cepstral mean normalization model includes:

[0017] Acquire training data, wherein the training data includes physiological data of multiple subjects, wherein the physiological data is obtained by performing cepstrum feature extraction on the physiological signals collected by the physiological signal collection device, and the physiological data of each subject includes multiple segments of signals;

[0018] Calculate the mean and variance of the signals of all segments of each subject respectively;

[0019] Based on the mean and the variance, respectively calculate a first average mean and a first average standard deviation of each subject, wherein the first average mean is the average mean between different segments of each subject, and the first average standard deviation is the average standard deviation between different segments of each subject;

[0020] A second average value and a second average standard deviation are calculated based on the first average value and the first average standard deviation, wherein the second average value is the average mean of signals among all subjects, and the second average standard deviation is the average standard deviation of signals among all subjects.

[0021] In one possible implementation, the normalizing the cepstrum features using the cepstrum mean normalization model includes:

[0022] The cepstrum feature is normalized according to a second average mean value and a second average standard deviation obtained by training the cepstrum mean normalization model to obtain the cepstrum feature of the physiological signal with at least part of the individual differences of the multiple subjects removed.

[0023] In one possible implementation manner, before the step of extracting frequency domain features based on the physiological signal, the method further includes:

[0024] Performing filtering processing and / or independent component analysis processing on the physiological signal, wherein the filtering processing includes high-pass filtering and / or low-pass filtering and / or notch filtering; and / or,

[0025] The physiological signal includes at least one of an electroencephalogram signal, an electrocardiogram signal, an electromyography signal, and an electrical skin signal; and / or,

[0026] The physiological signal collection process includes:

[0027] The multiple subjects wear or attach the physiological signal acquisition device and perform the same task under different fatigue states, and the physiological signals are acquired by the physiological signal acquisition device. The different fatigue states include at least two of the following: mild fatigue, moderate fatigue and severe fatigue.

[0028] In one possible embodiment, the method further includes:

[0029] Determining the fatigue state of the subject by using a subjective scale and / or objective physiological signal analysis, wherein the fatigue state includes mild fatigue, moderate fatigue or severe fatigue;

[0030] A corresponding relationship between the fatigue state of the subject and the corresponding physiological signal is established in the acquired physiological signal.

[0031] In a second aspect, the present application provides a device for removing individual differences in physiological signals, comprising:

[0032] An acquisition module, used for acquiring physiological signals of a plurality of subjects, wherein the physiological signals are signals acquired by a physiological signal acquisition device when the plurality of subjects perform the same task;

[0033] A first extraction module, configured to extract frequency domain features based on the physiological signal, wherein the frequency domain features are used to characterize the frequency characteristics of the physiological signal;

[0034] A second extraction module, used for extracting cepstrum features based on the frequency domain features;

[0035] The normalization module is used to perform normalization processing on the cepstrum feature to obtain the cepstrum feature of the physiological signal with at least part of the individual differences of the multiple subjects removed.

[0036] In a third aspect, the present application provides an edge computing device, comprising: a processor and a memory, the memory being used to store a computer program; the processor being used to run the computer program to implement the method for removing individual differences in physiological signals as described in the first aspect.

[0037] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer-readable storage medium is executed on a computer, the computer implements the method for removing individual differences in physiological signals as described in the first aspect.

[0038] Compared with the prior art, this application has at least the following technical effects:

[0039] The present application provides a method for removing individual differences in physiological signals, by extracting frequency domain features from physiological signals of multiple subjects, extracting cepstral features based on the frequency domain features, and inputting the cepstral features into a cepstral mean normalization model for normalization, that is, using the cepstral mean normalization technology, using the mean and standard deviation parameters obtained by training the cepstral mean normalization model, to normalize the cepstral features input into the model, which helps to achieve normalization of physiological signals and remove individual differences in physiological signals, without restricting the data distribution of physiological signals, so that the scope of application is wider and the model is more generalizable. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A schematic diagram of a process for removing individual differences in physiological signals provided in an embodiment of the present application;

[0041] Figure 2 A schematic diagram of a flow chart of a method for extracting cepstrum features provided in an embodiment of the present application;

[0042] Figure 3 A schematic diagram of a flow chart of a training method for a cepstral mean normalization model provided in an embodiment of the present application;

[0043] Figure 4 A schematic diagram of the structure of a device for removing individual differences in physiological signals provided in an embodiment of the present application;

[0044] Figure 5 A schematic diagram of the structure of an edge computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0045] In the embodiments of the present application, unless otherwise specified, the character " / " indicates that the objects before and after the association are in an or relationship. For example, A / B can represent A or B. "And / or" describes the association relationship of the associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone.

[0046] It should be pointed out that the words "first", "second", etc. involved in the embodiments of the present application are only used to distinguish the description purpose, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated, nor can they be understood as indicating or implying order.

[0047] In the embodiments of the present application, "at least one" refers to one or more, and "plurality" refers to two or more. In addition, "at least one of the following" or similar expressions refers to any combination of these items, which may include any combination of single items or plural items. For example, at least one of A, B, or C may represent: A, B, C, A and B, A and C, B and C, or A, B and C. Among them, each of A, B, and C may be an element itself, or a set containing one or more elements.

[0048] In the embodiments of the present application, "exemplary", "in some embodiments", "in another embodiment", etc. are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" in the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present concepts in a concrete way.

[0049] In the embodiments of the present application, "of", "corresponding", and "corresponding" can sometimes be used interchangeably. It should be noted that when the distinction between them is not emphasized, the meanings to be expressed are consistent. In the embodiments of the present application, communication and transmission can sometimes be used interchangeably. It should be noted that when the distinction between them is not emphasized, the meanings to be expressed are consistent. For example, transmission can include sending and / or receiving, which can be a noun or a verb.

[0050] An individual's emotional state, cognitive load level when completing a task, physical and mental stress and fatigue will seriously affect their work attitude and work efficiency. With the development of cognitive neuroscience, psychophysiology and computer science, the use of physiological signals to monitor an individual's state in real time has gradually come into the researchers' attention, and has had a huge impact on the fields of traffic human factors, aerospace, industrial construction, etc.

[0051] In the field of traffic human factors, it can not only be used in driver assistance systems to identify different braking intentions of drivers, use the classification results to perform braking operations, and reduce braking distances, but can also be used in the driver's smart cockpit to warn drivers of fatigue and dangerous driving behaviors, thereby reducing the accident rate. In the field of aerospace, it can monitor the cognitive load of drivers at different stages of aircraft takeoff, ascent, cruising, descent, and landing, design more reasonable operating consoles, and improve the driving safety of various types of pilots. In the field of industrial construction, it can be used to identify the stress of construction workers, helping to improve workers' safety, health, happiness, and productivity.

[0052] When a person receives external stimulation or has a certain emotional state, the excitement of the sympathetic nervous system will cause the dilation and contraction of blood vessels in the skin, the secretion of sweat glands, and changes in the heart's pumping rhythm, thereby causing changes in skin resistance (Electrodermalactivity, EDA) and heart rate variability (Heart Rate Variability, HRV). Physiological signals such as EEG signals, EMG signals, galvanic skin signals, ECG signals, respiratory rate, and blood pressure can reflect a person's physiological fatigue state.

[0053] The following is an explanation of EEG signals and HRV signals.

[0054] EEG signal: EEG is derived from the synchronous discharge of the postsynaptic membrane of a large number of neurons in the cerebral cortex. It can be divided into evoked EEG and spontaneous EEG according to whether it receives external stimulation. The former is induced by external stimulation and has obvious time-locked and phase-locked properties. Event-related potentials (ERP) can be obtained by superposition to reflect the cognitive processing of the brain. Spontaneous EEG has a specific physiological rhythm and can be divided into 5 frequency bands: δ, θ, α, β and γ. Each frequency band has its specific scalp distribution and physiological significance, and will be modulated by stimulation to produce an increase or decrease in the energy of a specific frequency band, which can reflect the individual's physical and mental state.

[0055] HRV signal: Heart rate variability is used to assess the autonomic regulation of the heart by quantifying the variability of sinus rhythm. The sinus rhythm time series is derived from the QRS (QRS complex, ventricular depolarization wave group) to QRS interval (or RR interval, which refers to the time distance between two consecutive R waves) sequence of the electrocardiogram (ECG) by extracting the normal sinus to normal sinus heartbeat interval (or NN interval, which refers to the time interval between two consecutive normal sinus rhythms). The relatively high frequency changes in sinus rhythm reflect parasympathetic (vagal) regulation, while the slower changes reflect parasympathetic and sympathetic regulation and non-autonomic factors.

[0056] The methods for normalizing physiological signals in related technologies include:

[0057] (1) Min-max normalization, also known as deviation normalization, is a linear transformation of the original data so that the resulting value is mapped to the range [0,1]. The conversion function is:

[0058]

[0059] where x max is the maximum value of the sample data, x min is the minimum value of the sample data.

[0060] This normalization method is more suitable for situations where the values ​​are relatively concentrated. However, if x max and x min Unstable, it is easy to make the normalization result unstable, making the subsequent use effect unstable. In actual use, you can use the empirical constant value to replace x max and x min , and when new data is added, it may cause x max and x min changes need to be redefined.

[0061] (2) Z-Score Standardization

[0062] This normalization method standardizes the data by giving the original data a mean and standard deviation. The processed data conforms to the standard normal distribution, that is, the mean is 0 and the standard deviation is 1. The conversion function is:

[0063]

[0064] The advantage of this normalization method is that the standardized data retains useful information in the outliers, making the algorithm less sensitive to outliers. The disadvantage of this method is that it assumes that the distribution of the data follows a normal distribution.

[0065] Based on the above problems, the embodiment of the present application proposes a method for removing individual differences in physiological signals, which helps to achieve normalization of physiological signals and remove individual differences in physiological signals without restricting the data distribution of physiological signals and has a wider scope of application.

[0066] Now combined Figure 1-Figure 3 The method for removing individual differences in physiological signals provided in an embodiment of the present application is described.

[0067] Figure 1 The flowchart of the method for removing individual differences in physiological signals provided in the embodiment of the present application specifically includes the following steps:

[0068] Step S11, acquiring physiological signals of multiple subjects.

[0069] In this step, according to the research purpose, multiple subjects are designed to perform the same repetitive tasks under different fatigue states, and physiological signals are collected by physiological signal acquisition devices worn or worn by multiple subjects. For example, in order to study the fatigue state of the driver, multiple subjects can be designed to perform tasks simulating driving behavior in at least two states of mild fatigue, moderate fatigue, and severe fatigue, and a portable electroencephalograph with dry electrodes (or wet electrodes or water electrodes) is used to collect electroencephalogram (EEG) signals of multiple drivers under different fatigue states, and the sampling rate can be 256Hz. The physiological signals are transmitted to an edge computing device for analysis and research. The edge computing device can be a desktop computer, a tablet computer, a laptop computer, a PDA, an ultra-mobile personal computer (UMPC) or a netbook, etc. The embodiment of the present application does not limit the type of edge computing device.

[0070] Optionally, the method provided in the embodiment of the present application may also include: using a subjective scale and / or objective physiological signal analysis to determine the fatigue state of the subject, the fatigue state including mild fatigue, moderate fatigue or severe fatigue; establishing a correspondence between the fatigue state of the subject and the corresponding physiological signal in the acquired physiological signal.

[0071] After each task, the subjects will fill in a subjective scale for subjective evaluation, that is, after each task is completed, a self-assessment questionnaire is presented to the subjects to collect the subjects' subjective feelings during the task. For example, in the simulated driving behavior task, the subjects' self-evaluation of their mental state is collected, where the mental state can be alert, a little alert, neither alert nor sleepy, a little sleepy, sleepy, need to work hard to stay awake, very sleepy, etc. The subjects fill in the questionnaire themselves and evaluate the current fatigue state of the subjects based on the questionnaire. For example, the fatigue state can be graded according to the mental state, and the mental state of alertness, a little alertness, neither alert nor sleepy can be classified as non-fatigue state, the mental state of a little sleepiness can be classified as mild fatigue, the mental state of sleepiness and needing to work hard to stay awake can be classified as moderate fatigue, and the mental state of very sleepiness can be classified as severe fatigue.

[0072] Optionally, objective physiological signal analysis can also be used, that is, the subject's current fatigue state can be determined by collecting and analyzing the subject's physiological signals. The subjective scale and objective physiological signal analysis can be used alone or in combination. After the subject's current fatigue state is determined, the correspondence between the fatigue state and the physiological signal can be annotated in the collected physiological signal, so that the cepstral feature normalization model can be trained on data sets under different fatigue states, thereby improving the accuracy and generalization of cepstral feature normalization.

[0073] In other embodiments, performing the same task may also be at least one of a pressure tolerance test task, a two-hand coordination task, and a motion tracking task. Among them, the pressure tolerance test task is used to test the subject's reaction ability, for example, pressing the corresponding color button or stepping on the corresponding pedal according to the screen prompt. Operational tasks such as two-hand coordination tasks and motion tracking tasks are used to test the subject's coordination ability, arm stability, and hand flexibility. When performing a two-hand coordination task, for example, the joystick can be manipulated to control the ball to move within the track. When performing a motion tracking task, for example, the joystick can be manipulated to control the crosshairs to aim at the target. The crosshairs is part of the aiming device. The crosshairs are usually located at the upper end of the aiming device. The crosshairs are cylindrical, triangular, rectangular, etc., and together with the sight and the notch form an aiming baseline. In the process of multiple subjects performing the same task, physiological signals are collected by the physiological signal acquisition device. Among them, the physiological signals can be photoplethysmogram (PPG), EDA (ie skin electrical signal), respiratory signal (RESP), skin temperature (SKT), periphery capillary oxygen saturation (SpO2), ECG (ie electrocardiogram signal) and electromyogram (Electromyogram, EMG, electromyography signal) as well as nine-axis posture data of the human body.

[0074] Step S12: extracting frequency domain features based on the physiological signal.

[0075] Specifically, the frequency domain feature is a feature extracted from the frequency domain of the physiological signal, which may include power spectrum density, spectrum distribution, etc., and is used to characterize the frequency characteristics of the physiological signal. In other embodiments, extracting frequency domain features from the physiological signal may also include extracting features of multiple frequency bands. For example, for EEG signals, features within a specific frequency band may be extracted from the EEG signal, such as extracting features of 5 frequency bands of δ (1-4 Hz), θ (4-8 Hz), α (8-12 Hz), β (12-31 Hz), and γ (31-75 Hz). The extraction of frequency band features can help identify specific patterns or behaviors in the signal. For example, in electroencephalogram analysis, different frequency band features reflect the electrical activity patterns of the brain in different states. For example, α waves are usually associated with relaxation and closed eyes, while β waves are associated with alertness and concentration. It should be noted that the division of frequency bands here is not unique, and the number and type of extracted frequency band features can be selected according to the research purpose, and this application does not limit it. By extracting features of multiple frequency bands from the physiological signal, the normalized physiological signal is more targeted and contains more information useful for subsequent research.

[0076] In one possible embodiment, before step S12, the method provided by the present application further includes: performing filtering processing and / or independent component analysis processing on the physiological signal. Optionally, the filtering processing includes high-pass filtering and / or low-pass filtering and / or notch filtering. Among them, the high-pass filtering can be set to 0.5 Hz, the low-pass filtering can be set to 70 Hz, and the notch filtering can be set to 50 Hz.

[0077] Filtering is mainly used to remove unnecessary frequency components in physiological signals to extract useful signal features or reduce the impact of noise. Among them, high-pass filtering is used to remove low-frequency noise, such as drift, baseline fluctuation, etc. For example, in electroencephalogram (EEG) signal processing, high-pass filtering can remove low-frequency components caused by eye movement or ECG interference. Low-pass filtering is used to remove high-frequency noise, such as electromyographic interference or power line interference. In electrocardiogram (ECG) signal processing, low-pass filtering can remove high-frequency noise and artifacts. Notch filtering is specifically used to remove interference at specific frequencies, such as removing 50Hz or 60Hz power line interference in EEG signals. The specific method of filtering needs to consider the characteristics of the signal and the type of noise. For example, if the signal mainly contains low-frequency components, a low-pass filter may be required, and for periodic interference, notch filtering is required.

[0078] Independent component analysis is used to separate statistically independent source signals from multivariate signals. In physiological signal processing, it is mainly used to separate different physiological sources in mixed signals, such as separating brain electrical activity and other bioelectric interference in EEG signals. Independent component analysis is applicable to multi-channel data and can effectively separate independent sources from multidimensional signals.

[0079] When the frequency range of the noise of the physiological signal is known, and these frequency ranges are significantly different from the useful frequency range of the signal, or when specific frequency interference, such as power line interference, needs to be removed, the physiological signal is filtered. When multiple independent sources are mixed in the physiological signal, and the statistical characteristics of these sources are different, or when specific physiological or non-physiological components need to be separated from multi-channel physiological signals, such as separating the electrooculogram and electrocardiogram components in the EEG signal, independent component analysis is required.

[0080] Optionally, when the physiological signal is an EEG signal, it is necessary to perform high-pass filtering, low-pass filtering and notch filtering on the EEG signal, which can effectively eliminate interference such as mains power and motion artifacts; it is also necessary to perform independent component analysis processing, which can remove artifact components such as electrooculography and electromyography in the EEG signal. The present application can remove interference signals and improve the normalization effect of physiological signals by performing filtering processing and / or independent component analysis processing on physiological signals.

[0081] Step S13, extracting cepstrum features based on the frequency domain features.

[0082] In this step, the frequency domain features include the power spectral density (PSD, hereinafter referred to as power spectrum) of the physiological signal. The power spectrum can be used to describe the distribution of signal power at different frequencies. Therefore, the present application extracts frequency domain features by calculating the power spectrum. The Wiener-Hinchin theorem proves that the power spectrum of a signal is equal to the Fourier transform of the autocorrelation function of the signal, so there are two ways to calculate the power spectrum: 1. (square of the Fourier transform) / (interval length); 2. Fourier transform of the autocorrelation function. These two methods are called the direct method and the correlation function method. Because the correlation function method has a better noise suppression effect and a smoother graph, the present application uses the autocorrelation function method to calculate the power spectrum.

[0083] like Figure 2 As shown, Figure 2 The flowchart of the method for extracting cepstrum features provided in the embodiment of the present application specifically includes the following steps:

[0084] Step S21: Divide the physiological signal into signals of a plurality of windows by using a sliding window method.

[0085] Specifically, the physiological signal is divided into multiple 4-second windows by sliding the window, with 4 seconds as one window. It is understandable that the window length can also be 2 seconds, 3 seconds, etc., and the application does not limit the window length.

[0086] Step S22: extracting the autocorrelation function of the signal in each window.

[0087] Specifically, the autocorrelation function of the signal in each window is extracted. The autocorrelation function is usually defined as the expected value of the product of the signal and itself at different time delays. Taking the signal length of each window as 4 seconds and the original sampling rate as 256Hz as an example, the length of the extracted data is 256*4=1024 points, that is, each window includes 1024 sample points.

[0088] Step S23, performing fast Fourier transform on the autocorrelation function to obtain the power spectrum density of the physiological signal.

[0089] Specifically, a Fast Fourier Transform (FFT) is performed on the autocorrelation function of each window to convert the time domain signal into a frequency domain signal to obtain the frequency domain representation of the window.

[0090] It should be noted that the above steps S21-S23 are used to obtain the power spectrum using the autocorrelation function method.

[0091] Step S24, taking the logarithm of the power spectrum density of the physiological signal, and then performing inverse Fourier transform to obtain the cepstrum feature.

[0092] Specifically, the power spectral density of the physiological signal is logarithmized, which helps to enhance the frequency characteristics of the signal and reduce the dynamic range differences between different frequency components. The frequency domain data after logarithmization is inversely transformed by Fourier to obtain the cepstrum feature. The cepstrum is the result of nonlinear transformation of the frequency domain of the signal, and is usually used for certain specific applications in signal processing, such as feature extraction. After the above steps S21-S24, each window will obtain a 1024-dimensional cepstrum feature vector.

[0093] Step S14, normalizing the cepstrum features to obtain cepstrum features of the physiological signals from which at least part of the individual differences of the multiple subjects are removed.

[0094] Cepstrum features refer to signal features obtained by performing logarithmic transformation and inverse Fourier transformation on the power spectrum of a signal. These features can capture the periodicity or frequency distribution characteristics of a signal, reveal the periodic components and linear prediction characteristics of a signal, reduce redundant information in the spectrum, and retain important features of the signal, which helps to extract useful information from complex signals. Using cepstrum features for normalization of physiological signals can improve the efficiency and accuracy of normalization. Normalization is to scale these cepstrum features to a uniform scale, remove individual differences between different subjects, that is, remove individual differences in physiological signals, so that in the subsequent comparison and analysis of physiological signals, individual differences of the subjects themselves can be excluded, and important features in physiological signals that are relatively common to different subjects, such as periodic changes in frequency, can be studied.

[0095] The normalized cepstral features can be applied to the field of multimodal physiological signal processing. For example, the normalized cepstral features can be combined with visual information to improve the accuracy of emotion recognition through a multi-level fusion method. The normalized cepstral features can also be fused at the feature level with the features of other physiological signals to improve the performance and robustness of the recognition system. In the classification stage, the normalized cepstral features can be fused at the decision level with other modal information to achieve more accurate classification and recognition.

[0096] Therefore, the normalized cepstrum features play a multi-faceted role in the field of physiological signal processing, which not only helps to improve the accuracy of signal processing, but also enhances the robustness and anti-noise ability of the system. By fusing the normalized cepstrum features with multimodal features, the performance of the physiological signal processing system can be further improved.

[0097] The present application obtains normalized cepstral features by normalizing the cepstral features of physiological signals, that is, obtaining cepstral features of physiological signals with at least part of individual differences of multiple subjects removed. These normalized cepstral features can be used for subsequent comparison and analysis of physiological signals.

[0098] In this step, normalizing the cepstral features includes: normalizing the cepstral features using a cepstral mean normalization model, wherein the training method of the cepstral mean normalization model is as follows: Figure 3 As shown, the specific steps include:

[0099] Step S31, obtaining training data, wherein the training data includes physiological data of multiple subjects, and the physiological data of each subject includes signals of multiple segments.

[0100] Specifically, the physiological signal acquisition device acquires physiological signals of multiple subjects, and the physiological data obtained after performing cepstrum feature extraction on the physiological signals is used as the training data of the model, wherein the training data includes physiological data of multiple subjects, and the physiological data of each subject includes multiple segments of signals. Optionally, 3 seconds can be taken as a window, with a window shift of 1 second, and finally a signal segment with a duration of 3 seconds can be generated every 1 second.

[0101] The following is an explanation of model training using EEG signals as an example. It can be understood that different types of physiological signals and different data have different feature dimensions, and the model supports the input of physiological signals with multiple feature dimensions.

[0102] For example, the training data includes EEG signals of multiple subjects (physiological data obtained after cepstrum feature extraction), and each subject's EEG signal includes signals of different segments. The file format of the data can be:

[0103] sub1: eeg1.csv, eeg2.csv, eeg3.csv

[0104] sub2: eeg4.csv, eeg5.csv

[0105] sub3: eeg6.csv, eeg7.csv ......

[0107] Among them, sub1 represents subject 1, sub3 represents subject 2, sub3 represents subject 3, eeg1.csv, eeg2.csv, eeg3.csv represent the EEG signals of three segments of subject 1, eeg4.csv and eeg5.csv represent the EEG signals of two segments of subject 2, and eeg6.csv and eeg7.csv represent the EEG signals of two segments of subject 3.

[0108] Generally, during the data collection stage, the training data can be grouped according to different subjects. If the grouping information is not provided, the data can be grouped using a clustering algorithm, or the EEG signal of each segment can be regarded as a physiological signal of a subject.

[0109] Step S32, respectively calculating the mean sum and variance sum of the signals of all segments of each subject.

[0110] At the beginning of model training, the number of segments, mean value, and variance value of each subject are initialized. The number of segments is the number of segments of the EEG signal. All training data are traversed, and for each subject, the mean value and variance value of all segments of the signal are calculated.

[0111] Step S33, respectively calculating the first average mean and the first average standard deviation of each subject based on the mean sum and the variance sum.

[0112] Specifically, the first average mean is the average mean between different segments of each subject, and the first average standard deviation is the average standard deviation between different segments of each subject. Wherein, the first average mean of the subject = the mean sum of the subject / the number of segments of the subject, and the first average standard deviation of the subject = square root (the variance sum of the subject / the number of segments of the subject-the first average mean of the subject*the first average mean of the subject), and the subject here refers to the same subject.

[0113] For example, the sum of the mean of subject 1 is A1, the sum of the variance of subject 1 is B1, the first average mean of subject 1 is a1, the first average standard deviation of subject 1 is b1, and the number of segments of subject 1 is 5, then

[0114]

[0115] Step S34, calculating a second average value and a second average standard deviation based on the first average value and the first average standard deviation.

[0116] Specifically, the second average mean is the average mean of the signals among all the subjects, and the second average standard deviation is the average standard deviation of the signals among all the subjects. Wherein, the second average mean=the first average mean of all the subjects and / the number of subjects, and the second average standard deviation=the first average standard deviation of all the subjects and / the number of subjects.

[0117] For example, the training data includes three subjects, the first average means of each subject are a11, a12, a13, the first average standard deviation of each subject are b11, b12, b13, the second average means are a2, and the second average standard deviation is b2.

[0118]

[0119] Optionally, normalizing the cepstral features using a cepstral mean normalization model may specifically include: normalizing the cepstral features according to a second average mean value and a second average standard deviation obtained by training the cepstral mean normalization model to obtain normalized cepstral features, that is, obtaining cepstral features of the physiological signal from which at least part of the individual differences of the multiple subjects are removed.

[0120] In this embodiment, two feature vectors μ and λ can be obtained according to the training of the model, where μ is the second average mean and λ is the square of the second standard deviation mean. Then the specific implementation method of normalizing the cepstrum feature by the cepstrum mean normalization model can be expressed as:

[0121] According to the formula The cepstrum feature is normalized,

[0122] Among them, x is the cepstral feature of the cepstral mean normalization model input, and y is the normalized result of the cepstral mean normalization model output.

[0123] In the present application, frequency domain features are extracted from the physiological signals of multiple subjects, cepstral features are extracted based on the frequency domain features, and the cepstral features are input into the cepstral mean normalization model for normalization. That is, the cepstral mean normalization technology is adopted, and the mean and standard deviation parameters obtained by training the cepstral mean normalization model are used to normalize the cepstral features input into the model. This helps to achieve the normalization of physiological signals and remove individual differences in physiological signals. There is no need to restrict the data distribution of physiological signals, so the scope of application is wider and the model is more generalizable.

[0124] In other embodiments, the present application also provides a device for removing individual differences in physiological signals, such as Figure 4 As shown, the physiological signal individual difference removal device 40 may include:

[0125] An acquisition module 41 is used to acquire physiological signals of multiple subjects, where the physiological signals are signals acquired by a physiological signal acquisition device when the multiple subjects perform the same task;

[0126] A first extraction module 42, configured to extract frequency domain features based on the physiological signal, wherein the frequency domain features are used to characterize the frequency characteristics of the physiological signal;

[0127] A second extraction module 43, used for extracting cepstrum features based on the frequency domain features;

[0128] The normalization module 44 is used to perform normalization processing on the cepstrum features to obtain cepstrum features of the physiological signals from which at least part of the individual differences of the multiple subjects are removed.

[0129] In one possible implementation, the frequency domain feature includes a power spectrum density of the physiological signal, and the first extraction module 42 is further configured to:

[0130] Dividing the physiological signal into signals of a plurality of windows by using a sliding window method;

[0131] Extract the autocorrelation function of the signal in each window;

[0132] The autocorrelation function is subjected to fast Fourier transform to obtain the power spectrum density of the physiological signal.

[0133] In one possible implementation, the second extraction module 43 is further configured to:

[0134] The power spectral density of the physiological signal is logarithmized and then inverse Fourier transformed to obtain the cepstrum feature.

[0135] In one possible implementation, normalizing the cepstral features includes: normalizing the cepstral features using a cepstral mean normalization model, wherein the training method of the cepstral mean normalization model includes:

[0136] Acquire training data, wherein the training data includes physiological data of multiple subjects, wherein the physiological data is obtained by performing cepstrum feature extraction on the physiological signals collected by the physiological signal collection device, and the physiological data of each subject includes multiple segments of signals;

[0137] Calculate the mean and variance of the signals of all segments of each subject respectively;

[0138] Based on the mean and the variance, respectively calculate a first average mean and a first average standard deviation for each subject, wherein the first average mean is the average mean between different segments of each subject, and the first average standard deviation is the average standard deviation between different segments of each subject;

[0139] A second average value and a second average standard deviation are calculated based on the first average value and the first average standard deviation, wherein the second average value is the average mean of signals among all subjects, and the second average standard deviation is the average standard deviation of signals among all subjects.

[0140] In one possible implementation, normalizing the cepstrum features using a cepstrum mean normalization model includes:

[0141] The cepstrum feature is normalized according to a second average mean value and a second average standard deviation obtained by training the cepstrum mean normalization model to obtain the cepstrum feature of the physiological signal from which at least part of the individual differences of the multiple subjects are removed.

[0142] In one possible implementation manner, before the step of extracting frequency domain features based on the physiological signal, the method further includes:

[0143] Performing filtering and / or independent component analysis on the physiological signal. The filtering includes high-pass filtering and / or low-pass filtering and / or notch filtering; and / or,

[0144] The physiological signal includes at least one of an electroencephalogram signal, an electrocardiogram signal, an electromyography signal, and an electrical skin signal; and / or,

[0145] The physiological signal collection process includes:

[0146] The multiple subjects wear or attach the physiological signal acquisition device and perform the same task under different fatigue states, and the physiological signals are acquired by the physiological signal acquisition device. The different fatigue states include at least two of the following: mild fatigue, moderate fatigue and severe fatigue.

[0147] In one possible embodiment, the method further includes:

[0148] Determining the fatigue state of the subject by using a subjective scale and / or objective physiological signal analysis, wherein the fatigue state includes mild fatigue, moderate fatigue or severe fatigue;

[0149] A corresponding relationship between the fatigue state of the subject and the corresponding physiological signal is established in the acquired physiological signal.

[0150] Figure 4 The physiological signal individual difference removal device 40 provided in the illustrated embodiment can be used to implement the technical solution of the method embodiment shown in the present application, and its implementation principle and technical effects can be further referred to the relevant description in the method embodiment.

[0151] It should be understood that the above Figure 4 The division of the various modules of the physiological signal individual difference removal device 40 shown is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. And these modules can all be implemented in the form of software calling through processing elements; they can also be all implemented in the form of hardware; some modules can also be implemented in the form of software calling through processing elements, and some modules can be implemented in the form of hardware. For example, the acquisition module can be a separately established processing element, or it can be integrated in a chip of an edge computing device. The implementation of other modules is similar. In addition, all or part of these modules can be integrated together, or they can be implemented independently. In the implementation process, each step of the above method or each of the above modules can be completed by an integrated logic circuit of hardware in a processor element or instructions in the form of software.

[0152] For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as one or more application specific integrated circuits (ASIC), or one or more microprocessors (DSP), or one or more field programmable gate arrays (FPGA). For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0153] In the above embodiments, the processor involved may include, for example, a CPU, a DSP, a microcontroller or a digital signal processor, and may also include a GPU, an embedded neural network processor (Neural-network Process Units; hereinafter referred to as: NPU) and an image signal processor (Image Signal Processing; hereinafter referred to as: ISP). The processor may also include necessary hardware accelerators or logic processing hardware circuits, such as ASIC, or one or more integrated circuits for controlling the execution of the program of the technical solution of the present application. In addition, the processor may have the function of operating one or more software programs, and the software programs may be stored in a storage medium.

[0154] An embodiment of the present application also provides a computer-readable storage medium, in which a computer program is stored. When the computer-readable storage medium is run on a computer, the computer executes the method provided by the embodiment shown in the present application.

[0155] An embodiment of the present application also provides a computer program product, which includes a computer program. When the computer program is run on a computer, it enables the computer to execute the method provided by the embodiment shown in the present application.

[0156] Combine the following Figure 5 The exemplary edge computing device provided in the embodiments of the present application is further introduced. Figure 5 A schematic diagram of the structure of an edge computing device 5000 is shown.

[0157] The above-mentioned edge computing device 5000 may include: at least one processor; and at least one memory communicatively connected to the above-mentioned processor, wherein: the above-mentioned memory stores program instructions that can be executed by the above-mentioned processor, and the processor calls the above-mentioned program instructions to execute the method for removing individual differences of physiological signals provided in the embodiment shown in this application.

[0158] Figure 5A block diagram of an exemplary edge computing device 5000 suitable for implementing the embodiments of the present application is shown. Figure 5 The edge computing device 5000 shown is merely an example and should not bring any limitations to the functionality and scope of use of the embodiments of the present application.

[0159] like Figure 5 As shown, the edge computing device 5000 is in the form of a general computing device. The components of the edge computing device 5000 may include, but are not limited to: one or more processors 5010, a memory 5020, a communication bus 5040 connecting different system components (including the memory 5020 and the processor 5010), and a communication interface 5030.

[0160] In an optional example, the edge computing device is deployed on the end side, connected to or integrated with the terminal such as the signal acquisition device, and acquires the physiological signals collected by the signal acquisition device, and performs individual difference elimination processing (i.e., normalization processing), and performs subsequent processing such as state classification based on the processed signal, or the edge computing device can transmit the processed signal to the server or cloud for subsequent processing such as state classification. This application migrates at least part of the computing tasks to the end-side device, which can give full play to the advantages of edge computing, improve data processing efficiency, reduce latency and network resource occupation and dependence, and is also conducive to data security and privacy protection.

[0161] The communication bus 5040 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor or a local bus using any of a variety of bus structures. For example, these architectures include but are not limited to Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MAC) bus, Enhanced ISA bus, Video Electronics Standards Association (VESA) local bus and Peripheral Component Interconnection (PCI) bus.

[0162] The edge computing device 5000 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the edge computing device, including volatile and non-volatile media, removable and non-removable media.

[0163] The memory 5020 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The edge computing device may further include other removable / non-removable, volatile / non-volatile computer system storage media. Figure 5 Not shown in the figure, a disk drive for reading and writing a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing a removable non-volatile optical disk (e.g., a compact disc read only memory (CD-ROM), a digital versatile disc read only memory (DVD-ROM), or other optical media) may be provided. In these cases, each drive may be connected to the communication bus 5040 via one or more data medium interfaces. The memory 5020 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present application.

[0164] A program / utility having a set (at least one) of program modules may be stored in memory 5020, such program modules including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. The program modules typically perform the functions and / or methods of the embodiments described herein.

[0165] The edge computing device 5000 may also communicate with one or more external devices (e.g., keyboards, pointing devices, displays, etc.), one or more devices that enable a user to interact with the edge computing device, and / or any device that enables the edge computing device to communicate with one or more other computing devices (e.g., network cards, modems, etc.). Such communication may be performed via the communication interface 5030. In addition, the edge computing device 5000 may also communicate with the network adapter ( Figure 5 The network adapter may communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via the communication bus 5040. It should be understood that although Figure 5Not shown, other hardware and / or software modules may be used in conjunction with the edge computing device 5000, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, disk arrays (Redundant Arrays of Independent Drives; hereinafter referred to as: RAID) systems, tape drives, and data backup storage systems.

[0166] The processor 5010 executes various functional applications and data processing by running the programs stored in the memory 5020, such as implementing the method provided in the embodiment of the present application.

[0167] It is understandable that the interface connection relationship between the modules illustrated in the embodiment of the present application is only a schematic illustration and does not constitute a structural limitation on the edge computing device 5000. In other embodiments of the present application, the edge computing device 5000 may also adopt different interface connection methods in the above embodiments, or a combination of multiple interface connection methods.

[0168] In the above embodiments, the processor involved may include, for example, a CPU, a DSP, a microcontroller or a digital signal processor, and may also include a GPU, an embedded neural network processor (Neural-network Process Units; hereinafter referred to as: NPU) and an image signal processor (Image Signal Processing; hereinafter referred to as: ISP). The processor may also include necessary hardware accelerators or logic processing hardware circuits, such as ASIC, or one or more integrated circuits for controlling the execution of the program of the technical solution of the present application. In addition, the processor may have the function of operating one or more software programs, and the software programs may be stored in a storage medium.

[0169] Those of ordinary skill in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented in a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0170] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0171] In several embodiments provided in the present application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory; hereinafter referred to as: ROM), random access memory (Random Access Memory; hereinafter referred to as: RAM), disk or optical disk, and other media that can store program codes.

[0172] The above is only a specific implementation of the present application. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. The protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for removing individual differences in physiological signals, characterized in that: The method comprises: Acquiring physiological signals of a plurality of subjects, wherein the physiological signals are signals acquired by a physiological signal acquisition device when the plurality of subjects perform the same task; Extracting frequency domain features based on the physiological signal, wherein the frequency domain features are used to characterize the frequency characteristics of the physiological signal; Extracting cepstrum features based on the frequency domain features; The cepstrum features are normalized to obtain cepstrum features of the physiological signal from which at least part of the individual differences of the multiple subjects are removed.

2. The method according to claim 1, characterized in that The frequency domain feature includes the power spectral density of the physiological signal, and extracting the frequency domain feature based on the physiological signal includes: Dividing the physiological signal into signals of a plurality of windows by using a sliding window method; Extract the autocorrelation function of the signal in each window; Performing a fast Fourier transform on the autocorrelation function to obtain a power spectrum density of the physiological signal.

3. The method according to claim 2, characterized in that The extracting of cepstrum features based on the frequency domain features comprises: The power spectrum density of the physiological signal is logarithmically taken, and then inverse Fourier transform is performed to obtain the cepstrum feature.

4. The method according to any one of claims 1 to 3, characterized in that: Normalizing the cepstral features includes: normalizing the cepstral features using a cepstral mean normalization model, wherein the training method of the cepstral mean normalization model includes: Acquire training data, wherein the training data includes physiological data of multiple subjects, wherein the physiological data is obtained by performing cepstrum feature extraction on the physiological signals collected by the physiological signal collection device, and the physiological data of each subject includes multiple segments of signals; Calculate the mean and variance of the signals of all segments of each subject respectively; Based on the mean and the variance, respectively calculate a first average mean and a first average standard deviation of each subject, wherein the first average mean is the average mean between different segments of each subject, and the first average standard deviation is the average standard deviation between different segments of each subject; A second average value and a second average standard deviation are calculated based on the first average value and the first average standard deviation, wherein the second average value is the average mean of signals among all subjects, and the second average standard deviation is the average standard deviation of signals among all subjects.

5. The method according to claim 4, characterized in that The normalizing process of the cepstrum features by using the cepstrum mean normalization model includes: The cepstrum feature is normalized according to a second average mean value and a second average standard deviation obtained by training the cepstrum mean normalization model to obtain the cepstrum feature of the physiological signal with at least part of the individual differences of the multiple subjects removed.

6. The method according to claim 1, characterized in that Before the step of extracting frequency domain features based on the physiological signal, the method further includes: performing filtering processing and / or independent component analysis processing on the physiological signal, wherein the filtering processing includes high-pass filtering and / or low-pass filtering and / or notch filtering; and / or, The physiological signal includes at least one of an electroencephalogram signal, an electrocardiogram signal, an electromyography signal, and an electrical skin signal; and / or, The physiological signal collection process includes: the multiple subjects wear or wear the physiological signal collection equipment, perform the same task under different fatigue states, and the physiological signals are collected by the physiological signal collection equipment. The different fatigue states include at least two of the following: mild fatigue, moderate fatigue and severe fatigue.

7. The method according to claim 6, characterized in that The method further comprises: Determining the fatigue state of the subject by using a subjective scale and / or objective physiological signal analysis, wherein the fatigue state includes mild fatigue, moderate fatigue or severe fatigue; A corresponding relationship between the fatigue state of the subject and the corresponding physiological signal is established in the acquired physiological signal.

8. A device for removing individual differences in physiological signals, characterized in that: include: An acquisition module, used for acquiring physiological signals of a plurality of subjects, wherein the physiological signals are signals acquired by a physiological signal acquisition device when the plurality of subjects perform the same task; A first extraction module, configured to extract frequency domain features based on the physiological signal, wherein the frequency domain features are used to characterize the frequency characteristics of the physiological signal; A second extraction module, used for extracting cepstrum features based on the frequency domain features; The normalization module is used to perform normalization processing on the cepstrum feature to obtain the cepstrum feature of the physiological signal with at least part of the individual differences of the multiple subjects removed.

9. An edge computing device, characterized in that: include: A processor and a memory, wherein the memory is used to store a computer program; and the processor is used to run the computer program to implement the method for removing individual differences in physiological signals as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed on a computer, the method for removing individual differences in physiological signals as described in any one of claims 1 to 7 is implemented.

Images