CN104055522A - Electrocardiosignal identity recognition method under arrhythmia condition - Google Patents

Abstract

An electrocardiosignal identity recognition method under the arrhythmia condition comprises the following steps that electrocardiosignal collection, filtering and standardized processing are carried out, electrocardiosignal sinus heart beats are judged, if electrocardiosignals are judged as the sinus heart beats, next processing is carried out, and otherwise the electrocardiosignals are removed; a heart beat feature vector generated after feature extraction and dimensionality reduction is obtained through the judged heart beats and autocorrelation and factor analysis; hierarchical clustering is carried out on heart beat templates, the templates with the larger similarity are classified as a class, and central points of all classes are used as templates to be recognized; in the recognition process, similarity matching recognition is carried out on the collected heart beats and the centers of the templates of all the classes. According to the electrocardiosignal identity recognition method under the arrhythmia condition, heart beat features of individuals are effectively distinguished and stored, and interference of arrhythmia waveforms and other waveforms with large deformation is eliminated; individual heart beat information is perfected, the identity recognition accuracy under the arrhythmia condition is improved.

Description

Electrocardiosignal personal identification method under a kind of arrhythmia conditions

Technical field

The present invention relates to electrocardiosignal personal identification method under a kind of arrhythmia conditions, belong to bio signal identification field, particularly utilize human ecg signal to carry out identification.

Background technology

Bio information recognition technology has been subject to scholar's extensive concern with its uniqueness, convenience, safety etc., this technology also becomes a kind of safety verification mode of being accepted by society gradually.Common biological identification technology has: fingerprint recognition, palmmprint identification, iris identification, Application on Voiceprint Recognition etc.In recent years, a kind of new biological characteristic is used as identification, and this feature is exactly electrocardiosignal.Electrocardiosignal is the weak biological signal of telecommunication that cardiac muscle produces before and after beating, and because individual physiological structure has its uniqueness, therefore this bioelectrical signals also has uniqueness in the projection of body surface.Through a large amount of statistical data research, it has uniqueness to prove electrocardiosignal; With respect to above several biological characteristics, adopt electrocardiosignal to there is following advantage as recognition feature: electrocardiosignal belongs to one-dimensional signal, is compared to several two dimensional characters above, and its data volume reduces greatly, only takies memory space seldom; Electrocardiosignal can only, from live body collection, therefore be difficult to forge.

Because electrocardiosignal feature has above advantage, therefore can be applied to the units higher to security protection requirement such as bank, insurance company, army.Electrocardiosignal is the concentrated expression of electrical activity of the myocardial cell of heart; human heart is constantly beated again; heart there will be abnormal conditions in work process for a long time unavoidably; such as common artrial premature beat, premature ventricular beat, leakage are fought etc.; these situations not only occur in patients with arrhythmia; even normal person also has other arrhythmia waveform, occur, normal person there will be above several frequently seen arrhythmia often at fatigue, irriate, excited, drug administration.The form difference of the form of these arrhythmia waveforms and normal sinus rhythm waveform is very large, and arrhythmia is of a great variety, and ARR appearance often has randomness.Just because of the existence of these situations, if the electrocardiosignal of collection exists arrhythmia when carrying out identification, if checkout gear is not clapped and processed the ARR heart, will have a strong impact on identification.Existing technology is generally to adopt shaping filter, and the waveform that electrocardiosignal wave distortion is comparatively serious carries out Regularization, or the feature of extracting on ecg wave form frequency domain is identified.But the shortcoming of doing is like this: no matter be shaping filter or the feature of extracting other frequency domains, the normal waveform of electrocardiosignal and unusual waveforms are still mixed in together, due to what adopt, be same approach, fundamentally do not eliminate the normal heart and clap the difference of clapping with the arrhythmia heart.Therefore, these class methods in ARR situation discrimination generally less than 90%, the discrimination far below normal waveform more than 97%.

Summary of the invention

The present invention considered under normal circumstances with arrhythmia conditions under the feature of electrocardiosignal, provide a kind of arrhythmia with all there is under normal circumstances the electrocardiosignal identification algorithm of high discrimination.As shown in Figure 1, major technique comprises electrocardiosignal pretreatment, arrhythmia differentiation, ecg characteristics extraction, template cluster and coupling to overall procedure.Specifically can be divided into following three modules:

1 arrhythmia discrimination module

1.1 electrocardiosignal pretreatment.Pretreatment is mainly the interference that may exist in electrocardiosignal in order to remove.Electrocardiosignal is other body surface signal of millivolt level, is easily subject to external interference, as myoelectricity interference, power frequency interference, baseline drift etc. are the common interference that affects electrocardiosignal waveform.Therefore in preprocessing process, need it to carry out filtering and noise reduction.

1.2 arrhythmia are differentiated.The object of carrying out arrhythmia differentiation is to distinguish the hole heart to clap and non-hole heart bat.The rhythm of the heart that the impulsion of being initiated by sinuatrial node forms is medically being called sinus rhythm.Sinus rhythm waveform is rule comparatively, and waveform is comparatively complete, has the QRS ripple of narrow shape, meets the standard of electrocardio identification.Otherwise the non-hole heart is clapped waveform, and often form is irregular, larger with the difference of Dou Xingxin bat waveform, in identifying, need its processing.The non-hole heart claps waveform and Fig. 2 is shown in the concrete contrast of Dou Xingxin bat waveform.Through the judgement of algorithm, if being delivered to next step, Dou Xingxin bat processes, if other non-hole hearts are clapped by its rejecting.

2 ecg characteristics extraction modules

2.1 auto-correlation functions (Autocorrelation, AC) are processed.Traditional electrocardio identification generally adopts datum mark as its recognition feature, adopts the method for datum mark to have its inherent shortcoming.The first, datum mark is detected, amount of calculation is larger, comparatively expends system resource; The second, medical circle does not have unified standard to datum mark particular location yet at present, and different dimensioning algorithm accepted standards is different.Therefore, in order to reduce the complexity of algorithm, reduce identification cost, the present invention adopts the heart after self correlation (AC) is processed to clap Wave data as feature.Electrocardiosignal is carried out to the amount of calculation that auto-correlation processing can reduce feature greatly, and the robustness of Enhanced feature.

2.2 Feature Dimension Reductions are processed.We adopt factorial analysis (Factor Analysis, FA) to carry out dimension-reduction treatment to feature.Factor analysis is mainly from data inside, analyzes its degree that influences each other, and utilizes a few factors to represent the Global Information of data.Factor analysis is PCA relatively, and it is explanatory that advantage is that its analysis result has more.Factor analysis is carried out the restructuring of information according to primary data information (pdi), a selected part information is used for representing data message originally, has played the effect of dimensionality reduction.

3 template matching

Template is carried out cluster.Utilize the method for hierarchical clustering that the electrocardiosignal template collecting is classified, object is that the decentraction of the same individuality of difference is clapped feature.The advantage of hierarchical clustering is the number that does not need artificial appointment cluster classification as K-Means, can adapt to different data characteristicses and automatically determine classification by algorithm.Template can better show the different heart of same identification person through cluster and clap feature, has avoided the less consequence of difference between the template characteristic of each sample that the existing meansigma methods that seeks template causes.

The invention has the advantages that

1. algorithm was just rejected arrhythmia waveform or clutter before identification, thus the minimizing of minimum degree the interference of arrhythmia or clutter, improved robustness and the discrimination of recognizer.

2. in the process and sample characteristics leaching process of the judgement of arrhythmia waveform, all need not check the characteristic points such as P ripple, Q ripple, T ripple, therefore reduce largely the complexity of algorithm.

3. adopt self correlation and factorial analysis to carry out dimension-reduction treatment, can when reducing sample dimension, retain most sample informations.

4. in template storing process, adopted non-supervisory hierarchical clustering, can preserve the heart that same identification person is different and clap characteristic, showed to greatest extent its heart and clap feature.

Accompanying drawing explanation

Fig. 1 is overview flow chart of the present invention.

Fig. 2 is that arrhythmia is differentiated flow chart.

Fig. 3 is that ecg characteristics extracts flow chart.

Fig. 4 is template level clustering flow journey figure.

Specific implementation

For further explaining in detail implementation step of the present invention, effect and advantage, below in conjunction with accompanying drawing, embodiment of the present invention are further described.

Electrocardiosignal personal identification method under the arrhythmia conditions that the present invention proposes, as shown in Figure 1, comprises the following steps:

Electrocardiosignal arrhythmia is differentiated:

201: before differentiating, need to carry out pretreatment to electrocardiosignal, eliminate power frequency, baseline drift and High-frequency Interference.Utilize five rank to go mean filter to remove the interference of high-frequency noise; Utilize band filter that cut-off frequency is 1-30Hz to eliminate the interference of power frequency and baseline drift.After obtaining the waveform of processing after filtering, need to carry out the detection of R ripple to it, the present invention adopts threshold detection method to detect R ripple position.According to the position of R ripple, get R wavefront and clap as a heart with the electrocardiosignal of latter 0.4 second for 0.4 second.

202: the method that the present invention adopts template matching is clapped the bat of the hole heart and arrhythmia or other hearts to make a distinction.Choosing of template comes from MIT-BIH Normal Sinus Rhythm Database (nsrdb).In this data base, include the electrocardiosignal of 18 normal sinus rhythms.Arrhythmia algorithm is clapped each heart gathering to mate with standard form, if the match is successful, carries out next step, unsuccessful by this heart bat rejecting.The construction process of standard hole heart beat template is as follows:

(a) every section of electrocardiogram (ECG) data in MIT-BIH Normal Sinus Rhythm Database is chosen to 30 hearts at random and clap, obtain altogether 540 hearts and clap.

(b) heart is clapped and is normalized.If primary signal is y _i, normalized signal is x _i, have:

$x_i=\frac{y_i-{\min }_y}{{\max }_y-{\min }_y}$

Wherein, max _ywith min _ybe respectively maximum and minima in primary signal.

203: obtain characteristic vector.From above-mentioned data base, obtain P ripple, QRS ripple, T ripple position and amplitude that the heart is clapped.Utilize P wave amplitude H _p, PR interval T _pR, QRS ripple interval T _qRS, QT interval T _qT, ST section amplitude H _sTthese five features form feature vector, X={ H that the heart is clapped _p, T _pR, T _qRS, T _qT, H _sT.

204: above 540 hearts are clapped, utilized its characteristic vector to adopt K-Means clustering algorithm to carry out cluster.The classification of K-Means cluster is made as 5 classes, gets the standard form that the electrocardiogram (ECG) data at the 5 Ge Lei centers that cluster obtains is clapped as the hole heart.

205: template matching adopts correlation coefficient process as the discrimination standard of similarity degree.If T[N] be hole heart beat template signal, S[N] be acquired signal, its correlation coefficient is defined as:

$r(T,S)=\frac{\mathrm{Cov}(T,S)}{\sqrt{D\left(T\right)\·D\left(S\right)}}$

D is the variance of signal sequence:

D(X)＝E{(X-E(X)) ²}

Cov (T, S) is the covariance of signal sequence:

Cov(T，S)＝E[(T-E(T))(S-E(S))]

Each heart beat of data collecting and 5 formwork calculation correlation coefficienies in hole template base, as long as one of them correlation coefficient r (T _i, S), i=1,2,3,4,5 are greater than threshold value T _dthink and clap for the hole heart, otherwise differentiate into other arrhythmia hearts, clap and reject.The arrhythmia heart is clapped differentiation flow process and is seen Fig. 2.

Electrocardiosignal feature extraction:

301: obtain through arrhythmia detection and after differentiating the data of clapping for the hole heart, first carry out auto-correlation function (AC) and process, auto-correlation function formula is as follows:

$y\left(m\right)=\frac{\underset{i=0}{\overset{N-\left|m\right|-1}{\mathrm{\Σ}}}x\left(i\right)x(i+m)}{y\left(0\right)}$

Wherein x (i) is original electrocardiographicdigital data, y (m), and m=0,2 ..., M-1 is the electrocardiogram (ECG) data after auto-correlation processing, M < N.

302: data after processing through auto-correlation function need it to adopt factor analysis (FA) to carry out dimension-reduction treatment.In factor analysis, each initial data represents by the linear sum of common factor and specific factor, and expression formula is as follows:

y _i＝a _i1F ₁+a _i2F ₂+…+a _ipF _p+ε _i(i＝1，2，…，m)

F ₁, F ₂..., F _p(p < M) is called common factor, ε _ibe called specific factor.Be expressed in matrix as:

Y＝AF+ε

$\left[\begin{array}{c}{y}_1\\ y_2\\ .\\ .\\ .\\ y_m\end{array}\right]=\left[\begin{array}{cccc}{a}_{11}& a_{12}& ...& a_{1p}\\ a_{21}& a_{22}& ...& a_{2p}\\ ...& ...& ...& ...\\ a_{m1}& a_{m2}& ...& a_{\mathrm{mp}}\end{array}\right]\left[\begin{array}{c}{F}_1\\ F_2\\ .\\ .\\ .\\ F_p\end{array}\right]+\left[\begin{array}{c}{\mathrm{\&epsiv;}}_1\\ {\mathrm{\&epsiv;}}_2\\ .\\ .\\ .\\ {\mathrm{\&epsiv;}}_m\end{array}\right]$

Wherein, A is factor loading matrix, its element representation the correlation coefficient in the factor and original data.The concrete steps that factor-analysis approach is carried out dimensionality reduction are as follows:

(a) initial data is carried out to standardization computing.Standardization formula is as follows:

$x_i=\frac{y_i-\mathrm{\&mu;}}{\mathrm{\&sigma;}}$

Wherein, the average that μ is signal, the standard deviation that σ is signal, x _ifor signal after standardization, y _ifor standardization front signal.After above-mentioned standardization, the average of signal is 0, and variance is 1.

(b) determine common factor number calculated factor loading matrix.

Because primary signal has been passed through standardization, the covariance matrix of signal just equals correlation matrix ∑.

Calculate the eigenvalue λ of ∑ ₁>=λ ₂>=...>=λ _m>=0, characteristic of correspondence vector is u ₁, u ₂..., u _m.Ignore the impact of model specific factor, omit rear m-p item and obtain factor loading matrix A, computing formula is as follows:

$A=[\sqrt{{\mathrm{\&lambda;}}_1}{u}_1,\sqrt{{\mathrm{\&lambda;}}_2}{u}_2,...,\sqrt{{\mathrm{\&lambda;}}_p}{u}_p]$

(c) adopt the rule for the treatment of of variance to carry out quadrature rotation to factor loading matrix, the square value of every row of load in the factor or every row element is broken up to 0 and 1.

If spin matrix T is:

Wherein, I is unit matrix, r, and g=1,2 ..., p.If rotation formula is: B=AT, B is rotation after load matrix.Order:

d _ij＝a _ij/t _i，i＝1，2，…，m

${\stackrel{\&OverBar;}{d}}_j=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{d_{\mathrm{ij}}}^2$

Have:

$V\left(\mathrm{\&theta;}\right)=\underset{j=1}{\overset{p}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{({d_{\mathrm{ij}}}^2-{\stackrel{\&OverBar;}{d}}_j)}^2/m=\max$

Meet above formula and meet rotating condition.Order can solve θ ₀, utilize θ ₀solve spin matrix T, thereby try to achieve rotation load matrix B according to rotation formula B=AT.

(d) ignore the impact of specific factor, utilize matrix F after formula reverse dimensionality reduction.Formula is as follows:

F＝B ^-1Y

Template matching:

Template hierarchical clustering.Through feature extraction and dimension-reduction treatment, the dimension of data has met the requirement of template storage, at this, template of collecting is carried out to hierarchical clustering.Hierarchical clustering specific implementation step is as follows:

401: initialize, each template is classified as to a class separately, calculate the distance between template between two.If template data is x _i, i=1,2 ..., p, x ^jwith x ^krepresent respectively different templates, the distance between two templates adopts manhatton distance (Manhattan Di stance), and expression formula is as follows:

$d_{\mathrm{jk}}=\underset{i=1}{\overset{p}{\mathrm{\&Sigma;}}}|{x_i}^j-{x_i}^k|$

402: find immediate two classes between each class, judge whether to meet distance threshold T _d, be less than threshold value and be classified as a class.And compute classes central sample position, otherwise algorithm termination is chosen the central point of each template class as the matching template in matching process.Class center sample position is by being attributed to this type of template average generation:

$z_i=\underset{k=1}{\overset{c}{\mathrm{\&Sigma;}}}{x}_k,i=\mathrm{1,2},...,p$

Wherein, z is cluster centre, and x is template data in cluster for this reason, and c is the number of template in cluster for this reason.

403: the similarity recalculating between newly-generated class and old class is the manhatton distance between compute classes central point, repeat 402.Hierarchical clustering flow process as shown in Figure 4.

The heart is clapped identification.The sample collecting through arrhythmia differentiate, its correlation coefficient of all kinds of center calculation carries out similarity differentiation in the data of feature extraction after processing and template, as long as wherein in some sample datas and template, a certain class similarity matching result is greater than threshold value, the match is successful for result, otherwise it fails to match.

Claims (6)

1. an electrocardiosignal personal identification method under arrhythmia conditions, comprises the following steps:

Step 1, ecg signal acquiring, filtering and standardization;

Step 2, electrocardiosignal arrhythmia is differentiated, and whether differentiate is that the hole heart is clapped;

Step 3, electrocardiosignal feature extraction and dimensionality reduction, obtain heart bat characteristic vector through self correlation and factorial analysis;

Step 4, heart beat template carries out hierarchical clustering, and all kinds of centers are clapped and are mated identification with the heart of collection.

2. electrocardiosignal personal identification method under arrhythmia conditions according to claim 1, is characterized in that, the step of described arrhythmia method of discrimination is: electrocardiosignal pretreatment; Set up hole heart beat template; Electrocardiosignal is carried out to Dou Xingxin and clap differentiation.

3. arrhythmia method of discrimination according to claim 2, is characterized in that, obtains standard hole electrocardio from hole electrocardio standard database; Electrocardio is normalized; Location R ripple, cuts the heart according to R wavelength-division and claps; Extract P ripple, QRS ripple, T ripple position and amplitude that the heart is clapped; Utilize P wave amplitude, PR interval, QRS ripple interval, QT interval, these five features of ST section amplitude to form characteristic vectors that the heart is clapped; To the heart bat feature obtaining, adopting K-Means clustering method to gather is 5 classes; Get the heart beat of data at above 5 class template centers as standard hole heart beat template.

4. arrhythmia method of discrimination according to claim 2, is characterized in that, heart beat of data to be identified is clapped and carried out respectively correlation coefficient measurement with 5 standard hole hearts that obtain; If the threshold value that the correlation coefficient that wherein has a group is greater than setting is clapped this heart to differentiate and clapped for the hole heart, otherwise differentiate into that other arrhythmia hearts are clapped and by its rejecting.

5. electrocardiosignal personal identification method under arrhythmia conditions according to claim 1, is characterized in that, in electrocardiosignal feature extraction, factorial analysis is carried out Feature Dimension Reduction and mainly comprised following step:

Step 1, carries out standardization computing to the electrocardiogram (ECG) data through self correlation conversion (AC);

Step 2, determines the quantity of common factor according to information reservation amount, calculate the factor loading matrix of electrocardiogram (ECG) data;

Step 3, adopts the rule for the treatment of of difference to carry out quadrature rotation to factor loading matrix, and the square value of the every row of factor loading battle array or every row element is broken up to 0 and 1;

Step 4, ignores the impact of specific factor, according to electrocardiogram (ECG) data after factor loading Matrix Calculating dimensionality reduction.

6. electrocardiosignal personal identification method under arrhythmia conditions according to claim 1, is characterized in that, heart beat template hierarchical clustering key step comprises:

Step 1, initial phase, is classified as a class separately by each heart beat template, distance between two between compute classes;

Step 2, according to class centre distance find all kinds of between immediate two classes, judge whether minimum range is less than threshold value, if meet above-mentioned condition, this two class is classified as a class, otherwise stops hierarchical clustering, preserves all kinds of heart beat templates;

Step 3, recalculates newly-generated Lei center, and recalculates the distance between two between all kinds of, repeating step 2.