CN103584872A - A psychological stress assessment method based on the fusion of multiple physiological parameters - Google Patents

Abstract

The invention discloses a psychological pressure assessment method based on fusion of multiple physiological parameters, which comprises the steps of designing a reasonable stimulation program, and collecting four electrophysiological signals of electrocardio, myoelectricity, pulse waves and electroencephalogram of a psychological pressure crowd; extracting emotional characteristics of the four electrophysiological signals; performing feature selection on the extracted features in a Relieff algorithm, a genetic algorithm optimization searching mode and the like; and obtaining a relevant fusion function based on the basic probability assignment function mass. The invention relates to a research of a method for evaluating psychological pressure based on multi-physiological parameter fusion, which is implemented by collecting multi-parameter signals, preprocessing, extracting characteristics, identifying psychological pressure emotion and fusing. Compared with single parameter classification identification, or multi-parameter data level fusion or feature level fusion, the method can more fully utilize data information and more accurately identify the psychological stress emotion.

Description

A kind of mental pressure appraisal procedure merging based on multi-physiological-parameter

Technical field

The present invention relates to mental pressure appraisal procedure technical field, particularly a kind of mental pressure appraisal procedure merging based on multi-physiological-parameter.

Background technology

Mental pressure claims again " psychological stress ", being a people is perceiveing or is recognizing that oneself is just facing to most important and while being difficult to the ambient conditions of reply, produces a kind of psychosoma tense situation of showing by psychology miscellaneous and physiological reaction tended to; That the common a kind of cognition forming of mental pressure source and Psychology of Main Body stress reaction and behavior are experienced, the emotional experience that shows as intrapsychic conflict and produce thereupon.

Suitably the mental pressure of degree there is no harm to human body, but excessive mental pressure brings series of negative impact, and chronic mental pressure can cause a series of physiology, risk of pathologies, as cardiovascular and cerebrovascular disease, and depression, abalienation etc.If can just identify accurately assessment before it becomes chronic mental pressure, will bring very large facility to us.The automatic identification of mental pressure, can help us to see clearly the factor that may cause mental pressure reaction in life; Also can be used for the individuality in mental pressure state to carry out therapeutic intervention simultaneously.Can under naturalness, carry out the analysis of mental pressure by contributing to study people's changeable in mood behavior, contribute to the objective evaluation to individual mind pressure rating simultaneously.

Psychological field is usually used in the method for mental pressure assessment and talks method, method of psychology test and questionnaire method.The remarkable response that these methods often need participant with coordinate.If mental pressure assessment can more objectively carried out under condition, its result will be more accurate reasonable, and be conducive to mental pressure and healthy research.The automatic identification technology that emotion is calculated as mental pressure provides abundant theory support.

Multi-sensor information fusion refer to the data from a plurality of sensor informations source detect, associated,

Relevant, estimation and multistage, many-sided processing such as comprehensive, to obtain accurate estimation and the assessment to measurand state, improve the overall performance of monitoring system.Process multi-sensor information integrated and that merge can be comprehensively and is accurately reflected environmental characteristic, and the physiological signal that multisensor is gathered simultaneously carries out information fusion, has improved the accuracy of mental pressure emotion recognition with respect to single signal identification.

It is the data that collect tentatively to be completed on the basis of feature extraction that decision-making level merges, apish thinking, make full use of all kinds of characteristic informations that characteristic layer merges the measuring object that extracts, by certain rule or specific algorithm, obtain the last identity of target, be a kind of high-level fusion, can draw rationally or be close to rational conclusion with coarse inference method.And assess at present substantially not having that in the research of mental pressure method, application decision level merges, so we will obtain higher discrimination in this new field.

Along with going deep into of research, research worker more and more recognizes, towards this challenge of thymopsyche Stress appraisal, the effectiveness of the reasonability of data acquisition system design and the initial data of collection, is the basis of emotion recognition and crucial.Therefore, how to set up the focus that mental pressure appraisal procedure accurately also becomes current research.

Abroad, in Germany's Augsburg university's emotion physiological data storehouse, mental pressure data base wherein, be to collect and breathe and electromyographic signal when experimenter operates famous computer picture-arrangement game " Tetris ", the object of experiment is to make experimenter alternately in high mental pressure and low mental pressure state.And at home, mental pressure affection data is also insufficient.

Summary of the invention

In order to overcome above-mentioned the deficiencies in the prior art, the object of this invention is to provide a kind of multiparameter Stress appraisal method.

Technical scheme of the present invention is based on electrocardio, myoelectricity, pulse wave and EEG signals, to obtain the mental pressure appraisal procedure of multi-parameter fusion, sets up a mental pressure assessment data storehouse based on electrocardio, myoelectricity, pulse wave and EEG signals simultaneously.

In order to solve the technical problem of above-mentioned existence, the present invention is achieved by the following technical solutions:

The mental pressure appraisal procedure that multi-physiological-parameter merges, its content comprises the steps:

The first step, stimulation programs reasonable in design, gather mental pressure crowd's electrocardio, myoelectricity, pulse wave and four kinds of electricity physiological signals of brain electricity, these gathered four kinds of electricity physiological signals carried out to the pretreatment such as denoising, the interference of filtering 50Hz power frequency;

Second step, the electricity physiological signal obtaining according to pretreatment carry out feature extraction, extract the affective characteristics of four kinds of electricity physiological signals:

The coefficient obtaining according to wavelet decomposition extracts cardiac electrical statistical nature and HRV feature, the statistical nature of myoelectricity and the statistical nature of pulse wave, brain electricity extracts respectively the features such as Kc complexity, approximate entropy and Wavelet Entropy based on Kc complexity, approximate entropy and Wavelet Entropy theory, and combination of two is characteristic vector respectively;

The 3rd step, by modes such as ReliefF algorithm, genetic algorithm optimizing, the various features of extracting are carried out to feature selection; Using characteristic vector as input message, be input in svm classifier device, obtain sample affiliated classification and classification accuracy;

The 4th step, based on basic probability assignment function mass, obtain relevant fusion function; Suppose that C is an identification space, all subsets are 2 ^c, m:2 ^c→ [0,1], is called a basic probability assignment, need to meet the following conditions:

Fusion function is as follows

$m\left(I\right)=\frac{1}{R^{-1}}\underset{A_i\∩B_j=I}{\mathrm{\Σ}}{m}_1\left(A_i\right)\·m_2\left(B_j\right)---(1-2)$

Wherein:

$R^{-1}=1-\underset{A_i\∩B_j=\mathrm{\φ}}{\mathrm{\Σ}}{m}_1\left(A_i\right)\·m_2\left(B_j\right)=\underset{A_i\∩B_j\≠\mathrm{\φ}}{\mathrm{\Σ}}{m}_1\left(A_i\right)\·m_2\left(B_j\right)$

Each sample of data space is assigned to an elementary probability, form the probability assignments value function of each signal source in target classification, according to fusion function, obtain different classes of different probability value, the classification that the classification of maximum probability is decided to be to this sample, obtain final classification results, thereby the attribute information that effectively merges electrocardio, myoelectricity, pulse and these four sensors of brain electricity, obtains higher discrimination.

Owing to adopting technique scheme, a kind of mental pressure appraisal procedure merging based on multi-physiological-parameter provided by the invention, compared with prior art has such beneficial effect:

The present invention merges based on multi-physiological-parameter the assessment mental pressure method that comprises electrocardio, myoelectricity, pulse wave and four kinds of electricity physiological signals of EEG signals, by physiological parameter being comprised to collection, pretreatment, feature extraction and the mental pressure emotion recognition of electrocardio, myoelectricity, pulse wave and four kinds of signals of EEG signals, and merge again and identify.With respect to single parameter Classification and Identification, or multiparameter pixel-based fusion or feature level fusion, the inventive method can be utilized more fully data message and identify more exactly mental pressure emotion.The foundation in multi-physiological-parameter mental pressure assessment data storehouse is abundanter with respect to previous data base, is more conducive to the research of mental pressure appraisal procedure.First gather physiological parameter, acquisition step comprises: on possibility high pressure crowd basis, then the emotion of bringing out tested person person with stimulus, make the data that obtain there is higher typicality, naturalness and controllability.At low noise, stimulate in material and gather and obtain information needed, the polygraph MP150 of employing U.S. Biopac company records tested person person's electrocardiosignal, surface electromyogram signal, pulse wave and EEG signals.

Four kinds of signal acquisition schemes that gather in experiment all induce tested person person's mental pressure emotion rationally and effectively, finally realize the foundation in mental pressure appraisal procedure and mental pressure assessment data storehouse.

Accompanying drawing explanation

Fig. 1 is the FB(flow block) that obtains electrocardiosignal, electromyographic signal, pulse wave and EEG signals mental pressure assessment data storehouse;

Fig. 2 is for obtaining the algorithm simulating flow chart of mental pressure appraisal procedure.

The specific embodiment

Below in conjunction with accompanying drawing and the specific embodiment, the present invention is described in further detail:

The mental pressure appraisal procedure that multi-physiological-parameter merges, implements the method and comprises the steps:

Electrocardiosignal, electromyographic signal, pulse wave and four kinds of mental pressure electricity physiological signals of EEG signals gather, as shown in Figure 1, that tested person person selects is three grades 16 students that are about to graduation reply of the postgraduate of University On The Mountain Of Swallows, 8 schoolgirls of 8 boy students wherein, they are all healthy, dextromanuality and all in high pressure conditions.

Research shows, the low amount noise of office is as cold air heating installation air-supply sound, computer sound and around colleague's the sound that mumbles all can be to staff's mental pressure of producing a feeling.In order to reach good effect of stimulation, survey result according to Britain's University of Salford professor Te Leifu Cox to uneasy sound, selected 5 kinds in the front ten uneasy sound of rank as mental pressure, to stimulate material, these 5 kinds of sound are the mike echo of sending, sob, the grating of Train wheel, the buzz of single child's sob and electric current that several child cries together, and stimulation order is as shown in the table:

The polygraph MP150 of employing U.S. Biopac company records tested person person's surperficial electrocardiosignal, electromyographic signal, pulse wave and EEG signals simultaneously.During image data, synchronously with earphone, to tested person person, playing mental pressure stimulates material.In test process, guarantee that test experiments indoor environment is absolutely quiet, and the equipment such as closing hand phone, reduce and disturb.Metering system: electrocardiosignal (ECG): three SMD electrodes, one is placed in right hand wrist, and one is placed in left foot ankle, and another,, as with reference to electrode, is placed in right crus of diaphragm foot wrist; Electromyographic signal (EMG): three SMD electrodes, one is placed in face's corrugator supercilii, and one is placed in left surface buccinator muscle, and another,, as with reference to electrode, is placed in right hand hands wrist; Pulse wave (PPG) a: sensor wraps left index finger the first joint and refers to abdomen; EEG signals (EEG): three electrode slices, two electrode slices are affixed on forehead two ends, and another,, as with reference to electrode, is placed in auris dextra ear-lobe.

Before experiment, first allow tested person person's rest a few minutes, it is loosened and calm down, physiological feature reaches usual state.Meanwhile, allow tested person person fill in the questionnaire survey whether personal information, Informed Consent Form and portion test tested person person have alexithymia, active and tested person person link up, and allow them understand our testing process, keeping tensions down emotion and elimination curiosity.

In order to weaken or eliminate tested person person's individual difference, after having gathered data, require tested person person to fill in a emotional experience questionnaire table, so that the screening of experimental data.To gathered various electricity physiological signal denoisings, in every group of data, intercepting is middle 2 seconds.In order to reduce the impact of noise, we adopt wavelet transformation to carry out denoising to signal.

The impact of not opened eyes and closing one's eyes due to brain electricity β ripple, and when cerebral cortex excitatory state, performance obviously, is therefore suitable as the object of study of experiment.Based on ICA method, data are carried out to denoising.EEG signals data cutout under each state is best embodied to 8 seconds of feature.By wavelet packet, adopt daubechies small echo to carry out 6 layers of decomposition to EEG signals, the decomposition node that β ripple is comprised is reconstructed, and has just obtained β ripple.

The simulation process of the mental pressure appraisal procedure algorithm based on multi-physiological-parameter, as shown in Figure 2, its basic step comprises Signal Pretreatment, feature extraction, Classification and Identification, fusion and finally identification.

Faint and be subject to the problem that extraneous noise jamming is strong based on electricity physiological signal, first to carry out denoising to electrocardio, myoelectricity, pulse wave and four kinds of signals of EEG signals, the parameter of wave filter is set respectively based on wavelet transformation, respectively four kinds of signals is carried out to filtering.Electrocardiosignal filtering 50Hz power frequency is disturbed, and electromyographic signal filtering clutter disturbs and is normalized, and pulse wave filtering clutter disturbs and unusual ripple.EEG signals filtering clutter disturbs.Setting based on wavelet transform filter parameter, first will carry out wavelet decomposition to each signal and obtain wavelet coefficient, then build filtering parameter matrix according to wavelet coefficient, carries out respectively filtering.

Electrocardiosignal feature extraction: detect electrocardiosignal P ripple, Q ripple, R ripple, S ripple, T ripple, by detecting the interval of these several ripples and crest as the data point of extracting statistical nature, the feature of extraction comprises: each waveform; QS ripple, the average between TS ripple, intermediate value, variance, value and value poor; HRV feature comprises: RR interval, the statistical value of cycle heart rate are, the statistical value of Heart rate distribution and heart rate variability power spectrum.

Electromyographic signal feature extraction: will carry out statistical nature extraction through pretreated electromyographic signal, extract its average, intermediate value, variance, value and value poor, and signal is asked to first derivative and second dervative, respectively to derivative again counting statistics value as characteristic vector.

Pulse wave feature extraction: the Time-domain Statistics feature of extracting pulse wave comprises average, intermediate value, variance, value and value poor, and signal is asked to first derivative and second dervative, then asks for respectively statistical value; Also to utilize in addition wavelet transformation to extract the pulse wave frequency spectrum of frequency domain.

EEG feature extraction: extract respectively the features such as Kc complexity, approximate entropy and Wavelet Entropy based on Kc complexity, approximate entropy and three kinds of feature extraction theories of Wavelet Entropy.

Classification and Identification: in order to remove repeatability and the redundancy of extracting feature, take respectively different feature selecting algorithm to screen, electrocardiosignal is carried out to feature selection by ReliefF algorithm, electromyographic signal is to carry out feature selection by PCA/ICA algorithm.Through the selected signal characteristic of feature selection, send into respectively svm classifier device and carry out classification based training and identification, obtain the preliminary mental pressure recognition result of four kinds of signals;

Merge and identify again: information fusion is for solving Information Complexity problem and the good method that makes full use of message complementary sense sexual clorminance in multi-sensor information analysis, especially the information fusion of decision level obtains the probability assignments value of each signal in the follow-up reforwarding of recognition result that obtains four kinds of signals with D-S evidence theory, namely the weight of signal in identification target classification, finally identifies and obtains the higher result of discrimination.

Tested person person is in high pressure or may be in high pressure conditions, adopt precision apparatus to obtain efficient signal data, the data that obtain have obvious mental pressure characteristic, except the mental pressure data base with other equally has general applicability, and himself there is the characteristic that the abundant and same mental pressure of pressure parameter is taken from same individuality under stimulating, with respect to common college student pressure data storehouse, there is abundanter researching value and use value.

The method also merging for assessment of mental pressure based on the multiple physiological signal of multi-sensor collection has more advantage and high accuracy with respect to single signal assessment mental pressure.This method has been introduced based on D-S evidence theory and decision level fusion method a plurality of electricity physiological signals source has been processed, can make full use of the information in physiological signal, by to database data emulation, result proves that this method has obviously improved classification accuracy.

Claims (1)

1. the mental pressure appraisal procedure merging based on multi-physiological-parameter, is characterized in that: its content comprises the steps:

The first step, stimulation programs reasonable in design, gather mental pressure crowd's electrocardio, myoelectricity, pulse wave, four kinds of electricity physiological signals of brain electricity, and these four kinds of electricity physiological signals that gather are carried out to the pretreatment such as denoising, the interference of filtering 50Hz power frequency;

Second step, the signal obtaining according to pretreatment carry out feature extraction, extract the affective characteristics of four kinds of electricity physiological signals:

The coefficient obtaining according to wavelet decomposition extracts cardiac electrical statistical nature and HRV feature, the statistical nature of myoelectricity and the statistical nature of pulse wave, and brain electricity extracts respectively feature based on Kc complexity, approximate entropy, Wavelet Entropy theory;

The 3rd step, by modes such as ReliefF algorithm, genetic algorithm optimizing, the feature of extracting is carried out to feature selection;

Using characteristic vector as input message, be input in svm classifier device, obtain sample affiliated classification and classification accuracy;

The 4th step, based on basic probability assignment function mass, obtain relevant fusion function; Suppose that C is an identification space, all subsets are 2 ^c, m:2 ^c→ [0,1], is called a basic probability assignment, need to meet the following conditions:

Fusion function is as follows

$m\left(I\right)=\frac{1}{R^{-1}}\underset{A_i\∩B_j=I}{\mathrm{\Σ}}{m}_1\left(A_i\right)\·m_2\left(B_j\right)---(1-2)$

Wherein:

$R^{-1}=1-\underset{A_i\∩B_j=\mathrm{\φ}}{\mathrm{\Σ}}{m}_1\left(A_i\right)\·m_2\left(B_j\right)=\underset{A_i\∩B_j\≠\mathrm{\φ}}{\mathrm{\Σ}}{m}_i\left(A_i\right)\·m_2\left(B_j\right)$

Each sample of data space is assigned to an elementary probability, form the probability assignments value function of each signal source in target classification, according to fusion function, obtain different classes of different probability value, the classification that the classification of maximum probability is decided to be to this sample, obtain final classification results, thereby the attribute information that effectively merges electrocardio, myoelectricity, pulse, these four sensors of brain electricity, obtains higher discrimination.

The QS ripple of sample, the average between TS ripple, intermediate value, variance, value and value poor; HRV feature comprises: RR interval, the statistical value of cycle heart rate are, the statistical value of Heart rate distribution and heart rate variability power spectrum;

Electromyographic signal feature extraction: will carry out statistical nature extraction through pretreated electromyographic signal, extract its average, intermediate value, variance, value and value poor, and signal is asked to first derivative and second dervative, respectively to derivative again counting statistics value as characteristic vector;

Pulse wave feature extraction: the Time-domain Statistics feature of extracting pulse wave comprises average, intermediate value, variance, value and value poor, and signal is asked to first derivative and second dervative, then asks for respectively statistical value; Also to utilize in addition wavelet transformation to extract the pulse wave frequency spectrum of frequency domain;

EEG feature extraction: extract respectively the features such as Kc complexity, approximate entropy and Wavelet Entropy based on Kc complexity, approximate entropy and three kinds of feature extraction theories of Wavelet Entropy.

Images