CN106419936A - Emotion classifying method and device based on pulse wave time series analysis - Google Patents
Emotion classifying method and device based on pulse wave time series analysis Download PDFInfo
- Publication number
- CN106419936A CN106419936A CN201610803574.1A CN201610803574A CN106419936A CN 106419936 A CN106419936 A CN 106419936A CN 201610803574 A CN201610803574 A CN 201610803574A CN 106419936 A CN106419936 A CN 106419936A
- Authority
- CN
- China
- Prior art keywords
- pulse wave
- wave time
- emotion
- time serieses
- support vector
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/16—Devices for psychotechnics; Testing reaction times ; Devices for evaluating the psychological state
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Veterinary Medicine (AREA)
- Biophysics (AREA)
- Pathology (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Psychiatry (AREA)
- Child & Adolescent Psychology (AREA)
- Social Psychology (AREA)
- Developmental Disabilities (AREA)
- Educational Technology (AREA)
- Hospice & Palliative Care (AREA)
- Physiology (AREA)
- Psychology (AREA)
- Artificial Intelligence (AREA)
- Cardiology (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention discloses an emotion classifying method and device based on pulse wave time series analysis. The emotion classifying method comprises the steps that a pulse wave time sequence is acquired through a photoplethysmography; a characteristic representation of the sequence is established according to the pulse wave time sequence; a support vector machine is trained according to the characteristic representation of the pulse wave time sequence and corresponding emotion labels; emotion classification is conducted on the pulse wave time sequence through the support vector machine. By implementing the emotion classifying method, nonlinear characteristics of the pulse wave time sequence can be accurately described and are utilized to achieve accurate emotion identification. In addition, the emotion classifying method is not limited to the acquisition process of the pulse wave time sequence and is also not limited to places such home and working places, a user can accurately perform emotion analysis on a tested person under the perception-free situation by himself/herself in real time.
Description
Technical field
Embodiment of the present invention is related to Medical Instruments technical field, and more particularly to one kind is divided based on pulse wave time serieses
The emotion sorting technique of analysis and device.
Background technology
Emotion is a kind of physiology of transient state and psychological phenomenon, and it represents the adaptation row that the individual environment to change is taken
For.Emotion is the individual psychology causing because of local environment, the reflection of physiological statuss, and it is different from position, attitude and disposition.Feelings
Thread is not only a subjective concept, is affected by the individual factor such as social, cultural that is located, emotion also can be in most of environment
Under, there are some determinations, identical composition between most people.For example when achieving successfully, people can compare
Happiness, glad degree is then depending on the height according to successful size and individual satisfaction, when suffering from heavy losses, Ren Mentong
The sadness that often can compare.
In prior art, extract maximally effective feature from physiological signal to identify emotion, the physiology letter that may be related to
Number include：
ECG (Electrocardiogram, electrocardiogram),
EMG (Electromyography, electromyogram),
RSP (Respiratory, breath signal),
SC (Skin conductance, the skin signal of telecommunication), wherein：
Detection method based on ECG：When the emotion such as angry, frightened, the heart rate of people is the fastest, takes second place, work as compassion when glad
Decreased heart rate when hindering and be surprised, when detesting, heart rate touches the bottom.The change of heart rate is by sex and emotion reciprocal effect,
As tested heart rate response level in women is tested higher than male.Relatively low heart rate variability rate (HRV) is shown to be the state loosened, and
Enhanced HRV shows possibly psychentonia and the state suffering setbacks.
Detection method based on EMG：EMG is a kind of time at skin surface for the electrical activity and the space of epidermis muscle
On synthesis result, it can be collected by surface electrode, and can avoid piercing the wound brought in muscle as needle electrode
Property defect.So it be guided by electrode from muscle surface, the activity of the neuromuscular system recorded when biological telecommunications
Number, the comprehensive effect of electrical activity mainly in superficial muscular and nerve trunk.Between the active state of it and muscle and functional statuses
The relatedness of various degrees, thus nervimuscular activity can be reflected on certain degree.Test table according to forefathers
Bright, electromyographic signal is a kind of very faint signal, its amplitude in l00～5000uV, its peak-to-peak value typically in 0～6mV, all
Root 0～l.5mv, typically useful signal frequency composition is located in the range of 0～500Hz, and wherein main energetic concentrates on 50
In the range of～150Hz.Electromyogram signal is One-dimension Time Series signal, and it is multiple motion lists that surface leading electrode is touched
The result of produced Electrical change superposition over time and space during the activity of position, from the ginseng under different functional statuss and active state
Plus the synchronization degree of the motor unit number of activity, the discharge frequency of different motion unit, motor unit activity, motor unit
Raise the factor such as pattern and surface electrode placement location, subcutaneous fat thickness, Temperature changing relevant.Electromyographic signal is closed with emotion
Closely, when emotion is more exciting, electromyographic signal performance is more active for system；When emotion is more gentle, electromyographic signal shows relatively
For inactive.
Detection method based on RSP：Breathing RSP refers to the process of that human body and external environment carry out gas exchange.Human body leads to
Cross Repiration constantly nutrient substance in external environment picked-up oxygen supply donor, maintain energy and body temperature, simultaneously by oxygen
The C02 producing during change excretes, thus ensureing metabolic being normally carried out.So, breathing is important one of human body
Individual physiological process.When measuring RSP signal, a resilient braces that placed piezoresistance sensor is wound on chest, when
During the expansion of people thoracic cavity, belt will be strained, and piezoresistance sensor exports corresponding magnitude of voltage.Breath signal is in close relations with emotion,
When emotion is more exciting, breath signal performance is more active；When emotion is more gentle, breath signal performance is not more lived
Jump.
Detection method based on SC：The skin signal of telecommunication (SC) is the instruction of SC, can inject one between the finger
Subtle small voltage, then measures its conductance.It is it desired to handss and do not fettered by sensor it is also possible to from being placed on foot
Electrode measurement obtains equally reliable signal.During different emotional state, the diastole of blood vessel and contraction and sweat gland secretion in skin
Deng change, the change of skin resistance can be caused, neurovegetative emotional response is measured with this.Generally SC is with the feelings of people
Thread awakening degree (Arousal) is relevant, and according to the theory of Schachter and Singer, same physiological signal is in different feels
Under awake degree, represented emotion is also different.The individual variation of galvanic skin response foundation level substantially and with personal characteristics phase
Close：The higher person of foundation level, more introversive, nervous, anxious, emotional instability, reaction sensitivity；And the low person of foundation level, more
Optimistic, export-oriented, phychology compares balance, and self-confident, mental adaptation is preferable.
But, the emotion analysis method being currently based on ECG, EMG, SC typically requires the collecting device wearing specialty, such
Equipment generally wears complexity, is not easy to tested individual's complete independently and wears, wears at any time and carry out real-time monitoring.Simultaneously as RSP
The emotional information that signal packet contains is limited, is generally difficult to carry out accurate emotion analysis only in accordance with this single modal information.
Therefore, in prior art, also do not have one kind to be both convenient for measuring, and can accurately distinguish and identify common emotion
Scheme.
Content of the invention
Embodiment of the present invention mainly solving the technical problems that provide a kind of pulse measurement device and based on pulse wave when
Between the emotion sorting technique of sequence analysis and device, can to user bring a kind of accurately, the good emotion of Consumer's Experience divides
Analysis method and apparatus.
For solving above-mentioned technical problem, the technical scheme that embodiment of the present invention adopts is：One kind is provided to be based on arteries and veins
Fight the emotion sorting technique of ripple time series analysis.
A kind of emotion sorting technique based on pulse wave time series analysis, the method includes：
Pulse wave time serieses are gathered by photoplethaysmography；
Build its character representation according to described pulse wave time serieses；
According to described pulse wave seasonal effect in time series character representation and corresponding emotion label Training Support Vector Machines；
By described support vector machine, emotion classification is carried out to described pulse wave time serieses.
Preferably, described build its character representation according to described pulse wave time serieses and include：
Obtain Chaos characteristic parameter according to described pulse wave time serieses, wherein, described Chaos characteristic parameter includes Li Ya
Pu Nuofu index, correlation dimension, approximate entropy and complexity.
Preferably, described according to described pulse wave time serieses obtain Chaos characteristic parameter also include：
Described Liapunov exponent, described correlation dimension, described approximate entropy and described complexity are normalized
Process, to build described pulse wave time serieses character representation.
Preferably, described according to described pulse wave seasonal effect in time series character representation and corresponding emotion label training
Hold vector machine to include：
According to the pulse wave time serieses of happy and angry two kinds of emotions, the nonlinear characteristic extracted and emotion are divided
The label joined, Training Support Vector Machines；
By described support vector machine to the nonlinear characteristic corresponding to the pulse wave time serieses of Unknown Label thus it is speculated that institute
State emotion label.
Preferably, described by described support vector machine, emotion classification is carried out to described pulse wave time serieses after also wrap
Include：
Gather new pulse wave time serieses；
Described new pulse wave time serieses are sent into described support vector machine；
By described support vector machine, emotion classification is carried out to described new pulse wave time serieses.
The invention allows for a kind of emotion sorter based on pulse wave time series analysis, this device includes：
Pulse wave time serieses acquisition module, for gathering pulse wave time serieses by photoplethaysmography；
First processing module, for building its character representation according to described pulse wave time serieses；
Second processing module, for according to described pulse wave seasonal effect in time series character representation and corresponding emotion label instruction
Practice support vector machine；
Emotion sort module, for carrying out emotion classification by described support vector machine to described pulse wave time serieses.
Preferably, described first processing module includes first processing units, and described first processing units are used for according to described
Pulse wave time serieses obtain Chaos characteristic parameter, and wherein, described Chaos characteristic parameter includes Liapunov exponent, correlation dimension
Number, approximate entropy and complexity.
Preferably, described first processing module also includes second processing unit, and described second processing unit is used for will be described
Liapunov exponent, described correlation dimension, described approximate entropy and described complexity are normalized, described to build
Pulse wave time serieses character representation.
Preferably, described Second processing module includes the 3rd processing unit and fourth processing unit：
Described 3rd processing unit is used for according to happily non-with what the pulse wave time serieses of angry two kinds of emotions were extracted
The label that linear character and emotion are distributed, Training Support Vector Machines；
Described fourth processing unit is used for right to the pulse wave time serieses institute of Unknown Label by described support vector machine
The nonlinear characteristic answered is thus it is speculated that described emotion label.
Preferably, described emotion sort module is additionally operable to：
Gather new pulse wave time serieses；
Described new pulse wave time serieses are sent into described support vector machine；
By described support vector machine, emotion classification is carried out to described new pulse wave time serieses.
Implement the present invention, with accurate description pulse wave seasonal effect in time series nonlinear characteristic, and can be realized using these features
Emotion is accurately identified, meanwhile, the present invention is not limited to pulse wave seasonal effect in time series gatherer process, be also not limited to be in or
Person work etc. place, user can voluntarily, accurately, in real time, in the case of unaware, emotion analysis is carried out to testee.
Brief description
Fig. 1 is pulse wave Time series analysis method first embodiment flow chart provided in an embodiment of the present invention；
Fig. 2 is pulse wave Time series analysis method fourth embodiment flow chart provided in an embodiment of the present invention；
Fig. 3 is pulse wave Time series analysis method the 5th embodiment flow chart provided in an embodiment of the present invention；
Fig. 4 is pulse wave time series analysis device sixth embodiment structured flowchart provided in an embodiment of the present invention；
Fig. 5 is pulse wave time series analysis device the 7th example structure block diagram provided in an embodiment of the present invention.
Specific embodiment
In order that the purpose of the present invention, technical scheme and advantage become more apparent, below in conjunction with drawings and Examples,
The present invention will be described in further detail.It should be appreciated that specific embodiment described herein is only in order to explain the present invention,
It is not intended to limit the present invention.
As long as additionally, involved technical characteristic in each embodiment of invention described below is each other not
The conflict of composition just can be combined with each other.
Embodiment 1：
The embodiment of the present invention 1 provides the first preferred embodiment of pulse wave Time series analysis method, is illustrated in figure 1
Pulse wave Time series analysis method first embodiment flow chart provided in an embodiment of the present invention.
A kind of pulse wave Time series analysis method providing refering to Fig. 1, the present embodiment, this method comprises the following steps：
S1, by photoplethaysmography gather pulse wave time serieses.The present embodiment is by the way of photoplethaysmography
Collection pulse wave time serieses, that is, utilizing photoelectric sensor, reflective light intensity after blood of human body and tissue resorption for the detection
The difference of degree, traces out change in cardiac cycle for the capacity of blood vessel.It is understood that the present embodiment can apply to portable
The pulse wave time serieses acquisition terminal of formula, is easy to user and implements this analysis method whenever and wherever possible, or, the present embodiment also may be used
To be applied to the equipment of fixation, consequently facilitating this analysis method of having a try in public places.Because this programme uses photocapacitance
Long-pending tracing gathers pulse wave time serieses, and therefore, the enforcement of this programme is not limited by environment, can be according to industrial functional need
Customization is implemented.
S2, build its character representation according to pulse wave time serieses.This step is passed through to analyze pulse wave time serieses, extracts
Pulse wave seasonal effect in time series nonlinear characteristic, determines its character representation by multiple nonlinear characteristics.
S3, according to pulse wave seasonal effect in time series character representation and corresponding emotion label Training Support Vector Machines.Wherein,
Emotion label is the classifying ruless to emotion, for example, emotion is divided into happy and angry two types, is both type of emotion
Determine its corresponding label respectively.This step adopts support vector machine, features described above is represented and corresponding emotion label
Parameter is trained, in order to train this support vector machine.
S4, by support vector machine, emotion classification is carried out to pulse wave time serieses.In this step, support vector machine are worked as
After the completion of trained, you can the pulse wave time serieses of input are analyzed, obtain emotion label.Test lead is according to emotion mark
Sign and can achieve that the classification to emotion is processed.
Having the beneficial effects that of the present embodiment, gathers pulse wave time serieses by photoplethaysmography；According to described arteries and veins
Ripple time serieses of fighting build its character representation；According to described pulse wave seasonal effect in time series character representation and corresponding emotion label
Training Support Vector Machines；By described support vector machine, emotion classification is carried out to described pulse wave time serieses.Can accurately retouch
State pulse wave seasonal effect in time series nonlinear characteristic, and realize emotion is accurately identified using these features, meanwhile, the present invention
Be not limited to pulse wave seasonal effect in time series gatherer process, be also not limited to the place such as be in or work, user can voluntarily, accurate
Really, in real time, in the case of unaware, emotion analysis is carried out to testee.
Embodiment 2：
The embodiment of the present invention 2 provides the second preferred embodiment of pulse wave Time series analysis method.
On the basis of above-described embodiment 1, build its character representation according to pulse wave time serieses and include：
Obtain Chaos characteristic parameter according to pulse wave time serieses, wherein, Chaos characteristic parameter includes Liapunov and refers to
Number, correlation dimension, approximate entropy and complexity.
In the present embodiment, it is related to the extraction operation to Chaos characteristic parameter, before extracting operation, need clearly to be wanted
The Chaos characteristic parameter extracting, for this Chaos characteristic parameter, the preferred version that the present embodiment is adopted is that Liapunov refers to
These four Chaos characteristic parameters of number, correlation dimension, approximate entropy and complexity.These four nonlinear characteristics, different aspect describes
Seasonal effect in time series nonlinear characteristic.Hereinafter these four Chaos characteristic parameters are briefly described：
(1) largest Lyapunov exponent
Chaotic motion shows as to this feature of sensitivity of initial condition：Apart from two close tracks due to initial strip
The minor alteration of part, thus their movement locus are away from each other, finally becomes do not have any relatedness.Liapunov exponent
Just it is used to describe this phenomenon.
System is maintained for a n, the mathematical definition that Liapunov exponent is is as follows：If the initial condition of this system is
One infinitesimal n ties up ball, and this n dimension ball develops into ellipsoid at leisure, and the main axis length according to ellipsoid arranges them, then and i-th
Individual Liapunov exponent is defined as
Wherein piT () is advancing the speed of i-th main shaft.
(2) correlation dimension
The movement locus with the dynamical system of chaotic characteristic through separation many times and are drawn close, and constantly stretch and roll over
Folded, may eventually form the geometric figure of ingredient and global similarity.This geometric figure referred to as divides shape, and the feature of point shape is
There is dimension.
(3) approximate entropy
Approximate entropy can carry out quantitative analysis of nonlinear to signal randomness, and the entropy of periodic signal is zero, chaotic signal
Approximate entropy is a nonnegative number, and therefore we can calculate this chaos characteristic parameter of approximate entropy of physiological signal, and physiology is described
The nonlinear characteristic of signal.Approximate entropy is insensitive for the maximum value minimum in signal, so robustness is stronger；To threshold value
It is adjusted just to realize noise filtering, such approximate entropy is hardly affected by noise,
(4) complexity
Although the characteristic quantities such as correlation dimension, Liapunov exponent and approximate entropy can represent the complexity of system,
Because correlation dimension is the static distribution in space for the describing system, Liapunov exponent has then pertained only to behavioral characteristics, approximately
The definition of entropy excessively theorizes, and is unsuitable for by the real data of sound pollution.One sequence is made up of " 0 ", " 1 ", this sequence
Complexity be exactly the minimal path sequence producing it bit number.For different sequences, producing their minimal path sequence length is
Different, not homotactic complexity therefore can be weighed with it.
The having the beneficial effects that of the present embodiment, by obtaining Chaos characteristic parameter according to described pulse wave time serieses, really
Pulse wave seasonal effect in time series Liapunov exponent, correlation dimension, approximate entropy and these four Chaos characteristic parameters of complexity are determined.
Above-mentioned four kinds of nonlinear characteristics, different aspect describes seasonal effect in time series nonlinear characteristic, thus being follow-up pulse wave time sequence
Row analysis operation provides data basis.
Embodiment 3：
The embodiment of the present invention 3 provides the third preferred embodiment of pulse wave Time series analysis method.
On the basis of above-described embodiment 2, obtain Chaos characteristic parameter according to pulse wave time serieses and also include：
Liapunov exponent, correlation dimension, approximate entropy and complexity are normalized, to build pulse wave
Time serieses character representation.
In the present embodiment, using normalized method, four kinds of nonlinear indicators of pulse wave seasonal effect in time series are calculated
Come.The preferred version that the present embodiment is adopted is, by carrying out the collection of experimental data, contrived experiment covering as much as possible
Multiple emotions, the maximum that each nonlinear indicator is occurred in an experiment, as each referring to target normalization factor.If after
Data more than normalization factor occurs in continuous experiment, new maximum is substituted former maximum, as normalization factor.
The having the beneficial effects that of the present embodiment, by by described Liapunov exponent, described correlation dimension, described approximate
Entropy and described complexity are normalized, to build described pulse wave time serieses character representation.Achieve to normalizing
Change the determination of the factor, thus providing data basis for the training operation of follow-up support vector machine.
Embodiment 4：
The embodiment of the present invention 4 provides the 4th preferred embodiment of pulse wave Time series analysis method.It is illustrated in figure 2
Pulse wave Time series analysis method fourth embodiment flow chart provided in an embodiment of the present invention.
A kind of pulse wave Time series analysis method providing refering to Fig. 2, the present embodiment.
On the basis of above-described embodiment 1, according to pulse wave seasonal effect in time series character representation and corresponding emotion label
Training Support Vector Machines include：
Nonlinear characteristic and emotion that the pulse wave time serieses of the happy and angry two kinds of emotions of S31, basis are extracted
The label being distributed, Training Support Vector Machines；
S32, by described support vector machine to the nonlinear characteristic corresponding to the pulse wave time serieses of Unknown Label, push away
Survey described emotion label.
In above-mentioned steps S31, after the completion of the experimental data of testee gathers, record the emotion of current subjectss certainly
Comment result (for example, happy and angry both type of emotion), and result is expressed as label 0 and label 1, thus build
Found the corresponding relation between experimental data and label.
In step s 32, obtain described emotion label by support vector machine.Including following step：
The first step, determines Liapunov exponent：
Hypothesis pulse wave time serieses are x1,x2,...,xn, Embedded dimensions are m, and time delayses are τ, then reconstruct is mutually empty
Between be Y (ti)=(x (ti),x(ti+τ),...,x(ti+(m-1)τ)), wherein i=1,2 ..., N, N=n-m+1 are vectorial number.
Take initial point Y (t0), if itself and closest point Y0(t0) distance be L0, follow the trail of this 2 points of temporal evolution, until t1Moment, its
Spacing exceedes threshold epsilon>0,Retain Y (t1), and in Y (t1) neighbouring in addition look for a point Y1
(t1) so that L1=| Y (t1)-Y1(t1)|<ε, and angle is as little as possible therewith, proceeds as described above, until when Y (t) reaches
Between sequence terminal N, at this moment following the trail of the total iterationses of evolutionary process is M, then this pulse wave seasonal effect in time series Liapunov
Index is
Second step, determines correlation dimension：
As it was previously stated, pulse wave time serieses are x1,x2,...,xn, using the method for moving to right, τ gradually steps up at regular intervals
The subscript of sequential element, builds the point set Y (t of the m dimension phase space that the exhibition of pulse wave time serieses is opened upi)=(x (ti),x
(ti+τ),...,x(ti+(m-1)τ)), optional m ties up one of the point set of phase space point YiAs a reference point, calculate other N-1 point
With its distance, then can count and grow with point YiCentered on, the number of the point in volume element with little scalar r as radius, from
And obtain correlation functionWherein, H (×) be Heaviside jump function.Order
dmaxFor maximum extension distance in m-dimensional space for the attractor, then work as r3dmaxWhen, C (r)=N (N-1)/N2=(N-1)/
N, as N → ∞, Cm(r) ＞ ＞ 1.There it can be seen that correlation function reflect dot spacing in attractor from distribution
Probability, therefore should haveWherein r ￡ dmax, D2(m, r) is the constant relevant with m and r.To little
Apart from r1And r2Have：Take the logarithm simultaneously and obtain in both sides：When | r1-r2| during very little, D2(m,r2) ＞ ＞ D2
(m,r1).Therefore, it can simplify further and obtainI.e. D2(m,r2) it is lnCmR ()～lnr is bent
The slope of line.WhenWhen, can get correlation dimension
3rd step, determines approximate entropy：
As it was previously stated, pulse wave time serieses are x1,x2,...,xnCarry out phase space reconfiguration, obtain by N=n-m+1
The phase space that vector is constituted, for each point Y in phase spacei, calculate and meet condition d (Yi,Yj) ￡ r number of vectors, and will
Count the data drawing and be expressed as Nm(i), to each i=1,2 ..., N, N=n-m+1, all count NmThe numerical value of (i), so
Calculate N afterwardsmI () and the ratio of vector distance total number N, is designated asTo all ofTake from so
Logarithm, then calculates the meansigma methodss of itself and the number for all iDimension m is changed into
M+1, repeats above calculating process, obtains φm+1(r).So, this pulse wave seasonal effect in time series approximate entropy is
4th step, determines complexity
L-Z complexity can characterize the wave character of Modulation recognition characteristic, reflects time serieses long with sequence
The speed of new model in the increase of degree, and complexity is bigger, illustrates that the new model that data occurred within the length of window time is more.
The extraction of L-Z complexity is based on signal codeization reconstruct.
As it was previously stated, to pulse wave time serieses x1,x2,...,xnAsk for minima and maximum, be designated as respectively
A=min (xi) and b=max (xi), carrying out symbolization reconstruct, to obtain new sequence s (i) as follows：IfSo s (i)=j；If f (i)=b, then s (i)=n-1.Thus
Obtain a symbolization reproducing sequence { s (i) } containing n symbol.According to L-Z complexity construction method, { s (i) } is decomposed into
The individual different substring of c (n).CalculateSo pulse wave time serieses x1,x2,...,xnL-
Z complexity can be by CLZ=c (n)/b (n).
5th step, determines normalization factor using method for normalizing
Through above-mentioned steps, by pulse wave time serieses x1,x2,...,xnFour kinds of nonlinear indicators calculate, point
Wei not Liapunov exponent：Correlation dimensionApproximate entropy isL-Z complexity can be by CLZ=c (n)/b (n).By carrying out adopting of experimental data
Collection, contrived experiment is as much as possible to cover multiple emotions, and the maximum that each nonlinear indicator is occurred in an experiment, as each
Normalization factor from index.If the data more than normalization factor occurs in subsequent experimental, new maximum is substituted former
Maximum, as normalization factor.
Through experimental data collection, such as GP method calculates correlation dimension, and result is the arithmetic number less than or equal to 10, then here
The 10 i.e. normalization factor as correlation dimension.
In above-mentioned steps S32, its corresponding emotional characteristics is determined according to emotion label.As above described in example, in experiment number
According to gatherer process in, record currently tested emotion self-appraisal result (happy and angry), and result be expressed as label 0 and 1,
Thus establish the corresponding relation between experimental data and label.Carry out many people and experiment be repeated several times, by experimental data each
Feature is normalized operation respectively, and together with label send into support vector machine carry out parameter training in order to train support to
The parameter of amount machine.
Preferably, this programme adopt LIBSVM support vector machine, so the model parameter obtaining be stored automatically for
Train.scale.model file, this document comprises to carry out needed for unknown data Tag Estimation using LIBSVM support vector machine
The parameter wanted：Nr_class represents the classification number that training sample set comprises, and rho is constant term b of decision function, and nr_sv is
Each class declines in borderline vector number, and obj is the value of the optimization object function to SVM support vector machine problem, and nSV is
The number of supporting vector, nBSV is the number of border supporting vector.
The having the beneficial effects that of the present embodiment, by determination and the normalization operation of characteristic parameter it is achieved that according to
Support vector machine determine pulse wave seasonal effect in time series emotion label, further, it is determined that the emotion corresponding to emotion label is special
Levy.
Embodiment 5：
The embodiment of the present invention 5 provides the 5th preferred embodiment of pulse wave Time series analysis method, is illustrated in figure 3
Pulse wave Time series analysis method the 5th embodiment flow chart provided in an embodiment of the present invention.
A kind of pulse wave Time series analysis method providing refering to Fig. 3, the present embodiment, this method comprises the following steps：
The new pulse wave time serieses of S41, collection.
S42, new pulse wave time serieses are sent into support vector machine.
S43, by support vector machine, emotion classification is carried out to new pulse wave time serieses.
In above three step, after the completion of to the new new sequence acquisition of pulse, due to not knowing specific emotion mark
Sign, the present embodiment, by after new for pulse sequential extraction procedures feature, sends into the prediction that support vector machine carry out emotion class label.Specifically
Process：1) gather the new sequence of pulse；2) extract pulse characteristics and be normalized；3) feature after normalization is sent into LIBSVM
Support vector machine, the model parameter that this support vector machine obtains according to the training of training set before, thus realize to emotion label
Prediction.
Pulse wave time serieses are carried out nonlinear analyses using nonlinear characteristic by having the beneficial effects that of the present embodiment,
Build the corresponding characteristic vector of pulse wave seasonal effect in time series, can the happy and angry two kinds of emotions of Accurate Prediction, accuracy exists
More than 95%；Sad and happy two kinds of emotions can accurately be predicted, accuracy is more than 95%.
The preferred version of the present embodiment is that contrived experiment is predicted the statistics of accuracy rate：1) experiment is by trial number 30 people；
2) when guaranteeing to occur happy two kinds of emotions with anger in the middle of tested one day, gather tested pulse data each once；3) add up
Collection natural law is not less than 10 days；4) add up at least each 200 groups of the data containing happy and angry two kinds of labels for the collection；5) enter line number
Data preprocess, rejects the more data of interference；6) carry out data characteristicses extraction；7) SVM support vector machine are trained；8) use SVM
Support vector machine carry out the prediction of new gathered data label.
Preferably, in carrying out SVM support vector machine training process, using 10 folding Cross-Validation technique it is ensured that SVM supports
The parameter stability of vector machine is accurate.
Embodiment 6：
The embodiment of the present invention 6 provides the 6th preferred embodiment of pulse wave time series analysis device, is illustrated in figure 4
Pulse wave time series analysis device sixth embodiment structured flowchart provided in an embodiment of the present invention.
A kind of pulse wave time series analysis device providing refering to Fig. 4, the present embodiment, this device includes：
Pulse wave time serieses acquisition module 10, for gathering pulse wave time serieses by photoplethaysmography；
First processing module 20, for building its character representation according to pulse wave time serieses；
Second processing module 30, for according to pulse wave seasonal effect in time series character representation and corresponding emotion label training
Support vector machine；
Emotion sort module 40, for carrying out emotion classification by support vector machine to pulse wave time serieses.
Embodiment 7：
The embodiment of the present invention 7 provides the 7th preferred embodiment of pulse wave time series analysis device, is illustrated in figure 5
Pulse wave time series analysis device the 7th example structure block diagram provided in an embodiment of the present invention.
On the basis of above-described embodiment 6：
Preferably, first processing module 20 includes first processing units 21, and described first processing units 21 are used for according to institute
State pulse wave time serieses and obtain Chaos characteristic parameter, wherein, described Chaos characteristic parameter includes Liapunov exponent, association
Dimension, approximate entropy and complexity.
Preferably, described first processing module 20 also includes second processing unit 22, and described second processing unit 22 is used for
Described Liapunov exponent, described correlation dimension, described approximate entropy and described complexity are normalized, with structure
Build described pulse wave time serieses character representation.
Preferably, described Second processing module 30 includes the 3rd processing unit 31 and fourth processing unit 32：
The pulse wave time serieses that described 3rd processing unit 31 is used for the happy and angry two kinds of emotions of basis are extracted
The label that nonlinear characteristic and emotion are distributed, Training Support Vector Machines；
Described fourth processing unit 32 is used for by described support vector machine the pulse wave time serieses institute to Unknown Label
Corresponding nonlinear characteristic is thus it is speculated that described emotion label.
Preferably, described emotion sort module 40 is additionally operable to：
Gather new pulse wave time serieses；
Described new pulse wave time serieses are sent into described support vector machine；
By described support vector machine, emotion classification is carried out to described new pulse wave time serieses.
Implement the present invention, with accurate description pulse wave seasonal effect in time series nonlinear characteristic, and can be realized using these features
Emotion is accurately identified, meanwhile, the present invention is not limited to pulse wave seasonal effect in time series gatherer process, be also not limited to be in or
Person work etc. place, user can voluntarily, accurately, in real time, in the case of unaware, emotion analysis is carried out to testee.
The foregoing is only embodiments of the present invention, not thereby limit the scope of the claims of the present invention, every utilization is originally
Equivalent structure or equivalent flow conversion that description of the invention and accompanying drawing content are made, or directly or indirectly it is used in other correlations
Technical field, is included within the scope of the present invention.
Claims (10)
1. a kind of pulse wave Time series analysis method is it is characterised in that methods described includes：
Pulse wave time serieses are gathered by photoplethaysmography；
Build its character representation according to described pulse wave time serieses；
According to described pulse wave seasonal effect in time series character representation and corresponding emotion label Training Support Vector Machines；
By described support vector machine, emotion classification is carried out to described pulse wave time serieses.
2. pulse wave Time series analysis method according to claim 1 it is characterised in that described according to described pulse wave
Time serieses build its character representation and include：
Obtain Chaos characteristic parameter according to described pulse wave time serieses, wherein, described Chaos characteristic parameter includes Li Yapunuo
Husband's index, correlation dimension, approximate entropy and complexity.
3. pulse wave Time series analysis method according to claim 2 it is characterised in that described according to described pulse wave
Time serieses obtain Chaos characteristic parameter and also include：
Described Liapunov exponent, described correlation dimension, described approximate entropy and described complexity are normalized,
To build described pulse wave time serieses character representation.
4. pulse wave Time series analysis method according to claim 1 it is characterised in that described according to described pulse wave
Seasonal effect in time series character representation and corresponding emotion label Training Support Vector Machines include：
The nonlinear characteristic extracted according to the pulse wave time serieses of happy and angry two kinds of emotions and emotion are distributed
Label, Training Support Vector Machines；
By described support vector machine to the nonlinear characteristic corresponding to the pulse wave time serieses of Unknown Label thus it is speculated that described feelings
Thread label.
5. pulse wave Time series analysis method according to claim 1 it is characterised in that described by described support to
Amount machine also includes after carrying out emotion classification to described pulse wave time serieses：
Gather new pulse wave time serieses；
Described new pulse wave time serieses are sent into described support vector machine；
By described support vector machine, emotion classification is carried out to described new pulse wave time serieses.
6. a kind of pulse wave time series analysis device is it is characterised in that described device includes：
Pulse wave time serieses acquisition module, for gathering pulse wave time serieses by photoplethaysmography；
First processing module, for building its character representation according to described pulse wave time serieses；
Second processing module, for according to described pulse wave seasonal effect in time series character representation and corresponding emotion label training
Hold vector machine；
Emotion sort module, for carrying out emotion classification by described support vector machine to described pulse wave time serieses.
7. pulse wave time series analysis device according to claim 6 is it is characterised in that described first processing module bag
Include first processing units, described first processing units are used for obtaining Chaos characteristic parameter according to described pulse wave time serieses, its
In, described Chaos characteristic parameter includes Liapunov exponent, correlation dimension, approximate entropy and complexity.
8. pulse wave time series analysis device according to claim 7 it is characterised in that described first processing module also
Including second processing unit, described second processing unit is used for described Liapunov exponent, described correlation dimension, described near
It is normalized like entropy and described complexity, to build described pulse wave time serieses character representation.
9. pulse wave time series analysis device according to claim 6 is it is characterised in that described Second processing module bag
Include the 3rd processing unit and Unit the 4th：
Described 3rd processing unit is used for according to happily non-linear with what the pulse wave time serieses of angry two kinds of emotions were extracted
The label that feature and emotion are distributed, Training Support Vector Machines；
Described fourth processing unit is used for by described support vector machine to corresponding to the pulse wave time serieses of Unknown Label
Nonlinear characteristic is thus it is speculated that described emotion label.
10. pulse wave time series analysis device according to claim 6 is it is characterised in that described emotion sort module
It is additionally operable to：
Gather new pulse wave time serieses；
Described new pulse wave time serieses are sent into described support vector machine；
By described support vector machine, emotion classification is carried out to described new pulse wave time serieses.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610803574.1A CN106419936A (en) | 2016-09-06 | 2016-09-06 | Emotion classifying method and device based on pulse wave time series analysis |
| PCT/CN2016/103783 WO2018045623A1 (en) | 2016-09-06 | 2016-10-28 | Emotion classification method and device based on pulse wave time sequence analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610803574.1A CN106419936A (en) | 2016-09-06 | 2016-09-06 | Emotion classifying method and device based on pulse wave time series analysis |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106419936A true CN106419936A (en) | 2017-02-22 |
Family
ID=58164692
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610803574.1A Pending CN106419936A (en) | 2016-09-06 | 2016-09-06 | Emotion classifying method and device based on pulse wave time series analysis |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN106419936A (en) |
| WO (1) | WO2018045623A1 (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107356417A (en) * | 2017-06-30 | 2017-11-17 | 暨南大学 | A kind of bolted joint damnification recognition method for merging Time-Series analysis and comentropy |
| CN107582037A (en) * | 2017-09-30 | 2018-01-16 | 深圳前海全民健康科技有限公司 | Method based on pulse wave design medical product |
| CN108523906A (en) * | 2018-04-27 | 2018-09-14 | 合肥工业大学 | Personality analysis method and system, storage medium based on pulse characteristics |
| CN109472290A (en) * | 2018-10-11 | 2019-03-15 | 南京邮电大学 | Mood swing model analysis method based on finite state machine |
| CN110090029A (en) * | 2018-01-28 | 2019-08-06 | 北京师范大学 | Emotional measurement system, Intelligent bracelet and portable terminal |
| CN110520048A (en) * | 2017-03-28 | 2019-11-29 | 国立大学法人九州工业大学 | Mood estimation device |
| CN111144436A (en) * | 2019-11-15 | 2020-05-12 | 北京点滴灵犀科技有限公司 | Emotional stress screening and crisis early warning method and device based on wearable equipment |
| CN111419250A (en) * | 2020-04-08 | 2020-07-17 | 恒爱高科(北京)科技有限公司 | Emotion recognition method based on pulse waves |
| FR3099615A1 (en) | 2019-07-30 | 2021-02-05 | Advestis | INTERPRETABLE PREDICTION PROCESS BY LEARNING USING TIME SERIES |
| FR3109232A1 (en) | 2020-04-10 | 2021-10-15 | Advestis | INTERPRETABLE PREDICTION PROCESS BY LEARNING OPERATING WITH LIMITED MEMORY RESOURCES |
| CN114334090A (en) * | 2022-03-02 | 2022-04-12 | 博奥生物集团有限公司 | Data analysis method and device and electronic equipment |
| WO2023221068A1 (en) * | 2022-05-19 | 2023-11-23 | 道本妙用科技(北京)有限公司 | Method and system for intelligent pulse wave analysis based on human body ordinal parameter model |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| RU2723732C1 (en) * | 2019-10-23 | 2020-06-17 | Акционерное общество «Нейротренд» | Method of analysing emotional perception of audiovisual content in group of users |
| CN115813389A (en) * | 2022-12-02 | 2023-03-21 | 复旦大学 | Pregnant woman delivery fear detection method and system based on skin electric signal analysis |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101887721A (en) * | 2010-07-19 | 2010-11-17 | 东南大学 | Electrocardiosignal and voice signal-based bimodal emotion recognition method |
| CN105496371A (en) * | 2015-12-21 | 2016-04-20 | 中国石油大学(华东) | Method for emotion monitoring of call center service staff |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2004237022A (en) * | 2002-12-11 | 2004-08-26 | Sony Corp | Information processing device and method, program and recording medium |
| CN105193431B (en) * | 2015-09-02 | 2017-12-01 | 杨静 | A kind of human body stress state analysis device |
-
2016
- 2016-09-06 CN CN201610803574.1A patent/CN106419936A/en active Pending
- 2016-10-28 WO PCT/CN2016/103783 patent/WO2018045623A1/en active Application Filing
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101887721A (en) * | 2010-07-19 | 2010-11-17 | 东南大学 | Electrocardiosignal and voice signal-based bimodal emotion recognition method |
| CN105496371A (en) * | 2015-12-21 | 2016-04-20 | 中国石油大学(华东) | Method for emotion monitoring of call center service staff |
Non-Patent Citations (2)
| Title |
|---|
| 韩清鹏: "脉搏信号的非线性分析及其不同情绪和环境的影响研究", 《浙江大学》 * |
| 颜红博: "基于多生理信息融合的情绪识别方法", 《中国优秀硕士学位论文全文数据库（电子期刊）》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110520048A (en) * | 2017-03-28 | 2019-11-29 | 国立大学法人九州工业大学 | Mood estimation device |
| CN107356417A (en) * | 2017-06-30 | 2017-11-17 | 暨南大学 | A kind of bolted joint damnification recognition method for merging Time-Series analysis and comentropy |
| CN107356417B (en) * | 2017-06-30 | 2019-10-18 | 暨南大学 | A kind of bolted joint damnification recognition method merging Time-Series analysis and comentropy |
| CN107582037A (en) * | 2017-09-30 | 2018-01-16 | 深圳前海全民健康科技有限公司 | Method based on pulse wave design medical product |
| CN110090029A (en) * | 2018-01-28 | 2019-08-06 | 北京师范大学 | Emotional measurement system, Intelligent bracelet and portable terminal |
| CN108523906A (en) * | 2018-04-27 | 2018-09-14 | 合肥工业大学 | Personality analysis method and system, storage medium based on pulse characteristics |
| CN109472290A (en) * | 2018-10-11 | 2019-03-15 | 南京邮电大学 | Mood swing model analysis method based on finite state machine |
| FR3099615A1 (en) | 2019-07-30 | 2021-02-05 | Advestis | INTERPRETABLE PREDICTION PROCESS BY LEARNING USING TIME SERIES |
| CN111144436A (en) * | 2019-11-15 | 2020-05-12 | 北京点滴灵犀科技有限公司 | Emotional stress screening and crisis early warning method and device based on wearable equipment |
| CN111419250A (en) * | 2020-04-08 | 2020-07-17 | 恒爱高科(北京)科技有限公司 | Emotion recognition method based on pulse waves |
| FR3109232A1 (en) | 2020-04-10 | 2021-10-15 | Advestis | INTERPRETABLE PREDICTION PROCESS BY LEARNING OPERATING WITH LIMITED MEMORY RESOURCES |
| CN114334090A (en) * | 2022-03-02 | 2022-04-12 | 博奥生物集团有限公司 | Data analysis method and device and electronic equipment |
| CN114334090B (en) * | 2022-03-02 | 2022-07-12 | 博奥生物集团有限公司 | Data analysis method and device and electronic equipment |
| WO2023221068A1 (en) * | 2022-05-19 | 2023-11-23 | 道本妙用科技(北京)有限公司 | Method and system for intelligent pulse wave analysis based on human body ordinal parameter model |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2018045623A1 (en) | 2018-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106419936A (en) | Emotion classifying method and device based on pulse wave time series analysis | |
| Bobade et al. | Stress detection with machine learning and deep learning using multimodal physiological data | |
| CN109620262B (en) | Emotion recognition system and method based on wearable bracelet | |
| Zenonos et al. | HealthyOffice: Mood recognition at work using smartphones and wearable sensors | |
| Boukhechba et al. | ActiPPG: Using deep neural networks for activity recognition from wrist-worn photoplethysmography (PPG) sensors | |
| Saganowski et al. | Emotion recognition for everyday life using physiological signals from wearables: A systematic literature review | |
| Camara et al. | Real-time electrocardiogram streams for continuous authentication | |
| Canento et al. | Multimodal biosignal sensor data handling for emotion recognition | |
| TW201918222A (en) | Method and apparatus for high accuracy photoplethysmogram based atrial fibrillation detection using wearable device | |
| Khalili et al. | Emotion detection using brain and peripheral signals | |
| CN108113663A (en) | Cardiac event detection system and method | |
| CN107405087A (en) | A kind of Wearable and its method for being used to assess the possibility of heart arrest generation | |
| CN109691994A (en) | A kind of rhythm of the heart analysis method based on electrocardiogram | |
| Chang et al. | Emotion recognition with consideration of facial expression and physiological signals | |
| Chandra et al. | Comparative study of physiological signals from empatica e4 wristband for stress classification | |
| Rosales et al. | Physiological-based smart stress detector using machine learning algorithms | |
| Boateng et al. | ActivityAware: an app for real-time daily activity level monitoring on the amulet wrist-worn device | |
| Ranganathan et al. | Estimation of heart rate signals for mental stress assessment using neuro fuzzy technique | |
| KR20070087787A (en) | Emotion classification system of real time bio signals and emotion classification method | |
| Mesin | A neural algorithm for the non-uniform and adaptive sampling of biomedical data | |
| Li et al. | Multi-modal emotion recognition based on deep learning of EEG and audio signals | |
| Ni et al. | Deep multi-branch two-stage regression network for accurate energy expenditure estimation with ECG and IMU data | |
| CN113749656A (en) | Emotion identification method and device based on multi-dimensional physiological signals | |
| Schellenberger et al. | Detecting respiratory effort-related arousals in polysomnographic data using lstm networks | |
| Rachakonda et al. | Stress-Lysis: An IoMT-enabled device for automatic stress level detection from physical activities |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170222 |