CN104102915A - Multiple-template matching identity recognition method based on ECG (Electrocardiogram) under electrocardiogram abnormality state - Google Patents

Abstract

The invention relates to a multiple-template matching identity recognition method based on an ECG (Electrocardiogram) under an electrocardiogram abnormality state, and belongs to the technical field of biological characteristic identity recognition. The ECG data of a user to be recognized is compared with the data of a registered user in a template library to obtain an identity recognition result. The key technology of the method comprises the following steps: carrying out electrocardiosignal preprocessing for eliminating noise interference; carrying out electrocardiosignal decomposition to separate an electrocardiogram waveform of each period; carrying out standardized processing for independently achieving standardization on time and amplitude scales; carrying out characteristic extraction: in the step, characteristics are extracted by wavelet transform, and clustering analysis is carried out by an ISODATA (Iterative Self-organizing Data Analysis Techniques Algorithm) so as to construct an ECC template library; and carrying out correlation analysis: in the step, correlation between ECG test data and each template is calculated, an optimal matching template is selected, and finally, an identity recognition result is obtained. The multiple-template matching identity recognition method provided by the invention utilizes the intrinsic electrocardiosignal of a human body to recognize an identity, and the ECG data under the abnormality state is considered.

Description

Personal identification method based on ECG multi-template matching under a kind of anomalous ecg state

Technical field

The present invention relates to the multi-template matching personal identification method based on ECG (Electrocardiogram) under a kind of anomalous ecg state, belong to biological characteristics identity recognizing technology field, utilize the electrocardiosignal identification identity of accumulateing in human body, not only for healthy population, and be applicable to have ARR individuality.

Background technology

In modern society, identification has important application in a lot of fields.Be accompanied by the continuous enhancing of society to security requirement, the drawback of traditional identity recognition method manifests gradually, such as certificate is easily lost, password easily cracks etc.Under this background, the identity recognizing technology based on biological characteristic causes people's extensive concern gradually, becomes one of focus of research.

Identification based on biological characteristic refers to physiological characteristic or the behavioral trait that utilizes human body, differentiates a technology of personal identification.Biological characteristic or behavioral trait for identification need meet the characteristics such as ubiquity, uniqueness, stability and measurability.By feat of advantages such as unique portability and reliabilities, biological characteristics identity recognizing technologies such as fingerprint recognition, recognition of face and speech recognition has been obtained fast development, and has been applied widely.

PQRST waveform state in ECG keeps relative stability within one period, even if pressing force and motion etc. easily to cause under the condition of changes in heart rate, the compound wave group of QRS is still stable.In addition, the impacts such as receptor type, age, sex, cardiac position, size, anatomical structure, thoracic cavity structure and heart physiological characteristic, ECG signal varies with each individual.Between the normal heartbeat of same individuality, between arrhythmia cordis heartbeat of the same race, have very large similarity, internal diversity is less than the difference between individual heartbeat, and therefore, ECG can be used as a kind of biological characteristic for identification.

Compare with traditional biometric identity recognition methods, by means of the ECG signal accumulateing in human body, carry out the concern that identification is subject to scholar gradually, it has many distinct advantages: 1. antifalsification, ECG signal comes from user's heart, it is a kind of living body biological feature, compare with features such as fingerprint, people's face and voice, eliminated by the hidden danger of imitating easily or stealing; 2. easily process, ECG is one-dimensional signal, and data volume is little, processes simply, saves storage space.

A lot of scholars inquire into for identification ECG signal, and early stage research work is mainly carried out around healthy population, according to normal ECG signal identification identity, and have obtained very high recognition accuracy.Yet, in actual life, thering is ARR individual ubiquity, the effect of above these methods under anomalous ecg state is also bad; Abnormal ECG signal is taken in some recent research work into account, and a lot of methods only relate to a certain or several types of arrhythmias, have significant limitation in practical application.

Summary of the invention

The present invention combines platform and considers normal and abnormal ECG signal, proposes a kind of accuracy rate high and be applicable to the identification algorithm of anomalous ecg state.As shown in Figure 1, gordian technique comprises that electrocardiosignal pre-service, signal decomposition and standardization, ecg characteristics extraction, ECG data clusters, electrocardio template database build and identification to overall procedure.Specifically can be divided into following five modules:

1. electrocardiosignal pretreatment module

1.1 obtain the multistage electrocardiogram (ECG) data under user's steady state (SS) with certain time length, with the form storage of data file, the identity information that corresponding header file comprises this user and electrocardiogram (ECG) data form;

1.2 pairs of every section of electrocardiogram (ECG) datas carry out pre-service, realize the noises such as the interference of filtering power frequency, baseline wander and myoelectricity interference by designing suitable wave filter.

2. signal decomposition and standardized module

2.1 couples of users' ECG data, detect R crest location;

2.2 adopt the method based on R crest location and adjacent R R interval, cut apart continuous ecg wave form, namely each cycle ecg wave form are successively separated;

2.3 standardizations, make all ecg wave forms of same individuality in time scale, have identical length, have equal maximum voltage value on amplitude yardstick.

3. electrocardiosignal characteristic extracting module

3.1 utilize R crest location and average RR interval information, extract the QRS complex wave of every bat ecg wave form;

3.2 extract the feature of ECG signal based on wavelet transformation (Wavelet Transform);

3.3 adopt principal component analysis (PCA) (Principal Component Analysis, PCA) to carry out dimensionality reduction to feature space.

4. electrocardio template database builds module

4.1 adopt ISODATA (Iterative Self-Organization Data Analysis Techniques) the ECG data clusters of algorithm to same user;

4.2 pairs of every class ECG data, using corresponding QRS complex wave as ECG template to be selected;

4.3 pairs of every class templates to be selected, adopt dependent thresholds method, choose the typical template of suitable number as the matching template of such data;

4.4 couples of all registered users do identical processing, thereby build electrocardio template database.

5. template matches and identification module

5.1 obtain one section of electrocardiogram (ECG) data under user's steady state (SS) to be identified with certain time length, the operations such as executing data pre-service, signal decomposition, standardization and the extraction of QRS complex wave;

5.2 adopt dependent thresholds method, and the QRS complex wave of choosing suitable number and can characterizing this user is as ECG test data;

5.3 mate each test data with all templates in ECG template base;

Related coefficient between 5.4 test datas and template data, as template matches criterion, is found the optimum matching template of test data, considers the template matching results of the whole ECG test datas of same user to be identified, determines this user's final identity.

The invention has the advantages that:

1. algorithm carries out standardization to isolated each cycle ecg wave form, and so just having eliminated the changes in heart rate causing due to external factors such as pressure, motions is well reflected in the inconsistency in ecg wave form time scale;

2. algorithm, when extracting QRS complex wave, has utilized ripe R ripple detection algorithm, has increased reliability and the precision extracted; And when extracting feature by wavelet transformation, without detection P ripple, Q, involve T ripple position, greatly reduced again the time complexity of algorithm;

3. algorithm is when building electrocardio template base, not only utilize normal ECG data, and for thering is ARR user, abnormal ECG data are also taken into account, normal and all kinds of arrhythmia cordis data are unified by ISODATA algorithm cluster, both improve identification precision, also expanded range of application;

4. when template matches and identification, in order to eliminate the impact of user's electrocardio singular value on algorithm, increase system robustness, choose a plurality of test datas and differentiate, the optimum matching template and the corresponding related coefficient that consider each test data of user, obtain final identification result.

Accompanying drawing explanation

Fig. 1 is technical solution of the present invention overview flow chart;

Fig. 2 is technical solution of the present invention modular structure schematic diagram;

Fig. 3 is electrocardiosignal pretreatment module process flow diagram;

Fig. 4 is signal decomposition and standardized module process flow diagram;

Fig. 5 is electrocardiosignal characteristic extracting module process flow diagram;

Fig. 6 is that electrocardio template database builds module process flow diagram;

Fig. 7 is template matches and identification module process flow diagram.

Embodiment

For making implementation step of the present invention, effect and advantage more clear, below in conjunction with accompanying drawing, embodiments of the present invention are described in further detail.

The present invention relates to the multi-template matching personal identification method based on ECG under a kind of anomalous ecg state, referring to Fig. 2, the method comprises:

201: electrocardiosignal pretreatment module;

202: signal decomposition and standardized module;

203: electrocardiosignal characteristic extracting module;

204: electrocardio template database builds module;

205: template matches and identification module.

Wherein, electrocardiosignal pretreatment module, for obtaining user's raw ECG data, is stored in data file, processes the noises such as myoelectricity interference in power frequency interference, baseline wander and gatherer process in erasure signal through filtering.The process flow diagram of electrocardiosignal pretreatment module, referring to Fig. 3, specifically comprises:

301: the raw ECG data of obtaining user.The present invention is with MIT-BIH arrhythmia cordis database (MIT-BIH Arrhythmia Database, MITDB) electrocardiogram (ECG) data in, as experimental data, is chosen wherein 44 electrocardiogram (ECG) data files, about 30 minutes of the duration of each file, characterize respectively each user, be designated as User _ii=1,2 ..., 44, within first 20 minutes, as training dataset, latter 10 minutes as test data set.

302: data filtering is processed.Human body electrocardio is feeble signal, and in gatherer process, is subject to various noise, and therefore, the step that is absolutely necessary is processed in filtering.

The filtering algorithm that the present invention adopts mainly completes the work of following several respects: 1. pair raw ECG signal goes equalization to process; 2. application moving average filter is eliminated high frequency noise interference; 3. eliminate baseline wander, mainly consider the low frequency signal of outside source; 4. the Butterworth filter rejection frequency that employing cutoff frequency is 30Hz is higher than the noise of 30Hz.Through as above four steps processing, can effectively remove various overriding noise in ECG signal and disturb.

Signal decomposition and standardization module.For pretreated ECG data, according to certain rule of cutting apart, user's each cycle ecg wave form is separated from continuous electrocardiosignal, and done standardization.The process flow diagram of signal decomposition and standardized module, referring to Fig. 4, specifically comprises:

401:R crest location detects.For extracting the feature of ECG signal, need to detect R crest location, for this reason, the present invention adopts the QRS ripple detecting device of ECGPUWAVE by name to realize, and the output of this detecting device is a file that comprises R crest sampling point position.

402: electrocardiosignal decomposes.For by user's ECG data clusters, and and then build electrocardio template base, the ECG signal of each cardiac cycle need to be separated from continuous electrocardiogram (ECG) data record.The present invention adopts the method based on R crest location and adjacent R R interval to realize.In MITDB, the sample frequency of electrocardiogram (ECG) data is 360Hz, to user User _ielectrocardiogram (ECG) data in training set, can be calculated periodicity is M, and sampling number is n, and this segment data duration is

$t=\frac{n}{360}s$

Average RR interval, is

$t_{\mathrm{RR}}=\frac{n}{360*M}s$

Electrocardiosignal decomposes follows following principle:

1. pair current ecg wave form, R crest location is t _r-position, is designated as t with the time interval of previous adjacent R crest _{rR_}pre, is designated as t with the time interval of a rear adjacent R crest _{rR_}next;

2. access time, interval was [t _{r_}position-0.4*t _{rR_}pre, t _{r_}position+0.6*t _{rR_}next] one piece of data characterize current ecg wave form;

3. couple all users' electrocardiogram (ECG) data is done same treatment, realizes electrocardiosignal and decomposes.

403: time scale standardization.Heart rate is easily affected by the external environment, and adverse effect arithmetic accuracy being caused in order to eliminate changes in heart rate need to be done standardization to each ecg wave form in time scale, and specific implementation is as follows:

With current period ecg-r wave peak position, be set to benchmark, and with reference to time value t between this user's average RR _rRadjust.

1. if the time interval length of current period ecg wave form is greater than average RR interval, i.e. 0.4*t _{rR_}pre+0.6*t _{rR_}next > t _rR, carrying out interval compression, interval compressibility coefficient is

$k_{\mathrm{comp}}=\frac{0.4*t_{\mathrm{RR}-}\mathrm{per}+0.6*t_{\mathrm{RR}-}\mathrm{next}-t_{\mathrm{RR}}}{0.4*t_{\mathrm{RR}-}\mathrm{pre}+0.6*t_{\mathrm{RR}-}\mathrm{next}}$

2. if the time interval length of current period ecg wave form is less than average RR interval, i.e. 0.4*t _rR-pre+0.6*t _{rR_}next < t _rR, carrying out interval extension, interval extension coefficient is

$k_{\mathrm{expd}}=\frac{t_{\mathrm{RR}}-(0.4*t_{\mathrm{RR}-}\mathrm{pre}+0.6*t_{\mathrm{RR}-}\mathrm{next})}{0.4*t_{\mathrm{RR}-}\mathrm{pre}+0.6*t_{\mathrm{RR}-}\mathrm{next}}$

3. pair all ecg wave forms are done above-mentioned interval compression or extension process, obtain standardized ECG signal in time scale.

404: amplitude scale calibration.The amplitude variation causing in order to eliminate surveying instrument, need to process in the enterprising column criterionization of amplitude yardstick each ecg wave form, and specific implementation is as follows:

1. record the amplitude at all ECG signals of same user R crest location place, averaged

2. if the amplitude at current ECG signal R crest location place is greater than mean value, , by amplitude compression, compressibility coefficient is

$k_{\mathrm{comp}}=\frac{{\mathrm{Amp}}_R-\stackrel{\&OverBar;}{{\mathrm{Amp}}_R}}{{\mathrm{Amp}}_R}$

3. if the amplitude at current ECG signal R crest location place is less than mean value, amplitude is stretched, drawing coefficient is

$k_{\mathrm{expd}}=\frac{\stackrel{\&OverBar;}{{\mathrm{Amp}}_R}-{\mathrm{Amp}}_R}{{\mathrm{Amp}}_R}$

405: electrocardiogram (ECG) data storage.To user User _ii=1,2 ..., 44 electrocardiogram (ECG) data, through signal decomposition and standardization, preserves with two-dimensional matrix, is designated as ECG ⁱ _mn, i=1,2 ..., 44, wherein m is ECG number, n is the sampling number of ECG data.

Electrocardiosignal characteristic extracting module.For ECG standardized data, extract QRS complex wave on the one hand, as user's template set to be selected, with wavelet transformation, extract primitive character on the other hand, principal component analysis (PCA) is carried out dimensionality reduction to feature space, namely eliminates redundancy feature, can reduce the time complexity of algorithm.The process flow diagram of electrocardiosignal characteristic extracting module, referring to Fig. 5, specifically comprises:

501: extract QRS complex wave.Take R crest location as benchmark extracts QRS complex wave, for standardized ECG data, only need utilize R crest location and RR interval mean value t _rRextract.If current ECG signal R crest location is t _r-position, with 0.15*t _rRduration intercept forward and backward respectively one section of electrocardiogram (ECG) data as QRS complex wave.For user User _ii=1,2 ..., 44, the QRS complex wave extracting with two-dimensional matrix storage, is designated as QRS ⁱ _mn, i=1,2 ..., 44, wherein m is QRS waveform number, n is the sampling number of each QRS waveform.

502: wavelet transformation extracts feature.Adopt the wavelet function of Daubechies by name to carry out wavelet transformation to ECG signal, Daubechies small echo is called for short dbN, and wherein N is wavelet-order, and the Support of wavelet function ψ (t) and scaling function φ (t) is 2N-1.ψ (t) can be obtained by φ (t), be φ (2t) weighted shift and, be shown below

$\mathrm{\&psi;}\left(t\right)=\underset{k}{\mathrm{\&Sigma;}}{g}_k\mathrm{\&phi;}(2t-k)$

N value is different, weights g _kalso different.φ (t) limited length, supporting domain is [0,2N-1], and therefore, the ψ obtaining (t) is finite support, and supporting domain is [1-N, N].

Theoretical and experience shows, db3 small echo is similar to ecg wave form, the principle of similarity of satisfied selection, therefore, select db3 small echo as wavelet basis, it has length is 5 supporting domain, shorter bearing length can reduce the time complexity of algorithm effectively, is conducive to the feature extraction of ECG signal.

Because different user coefficient of wavelet decomposition waveform is more more obvious than time domain waveform difference, and each heart bat coefficient of wavelet decomposition different wave shape of same user is little, more stable, ECG time domain waveform is carried out to 6 grades of wavelet decomposition, db3 is as wavelet basis, wavelet coefficient after conversion forms proper vector, is designated as x=[x ₁, x ₂..., x _p].

503:PCA carries out Feature Dimension Reduction.Principal component analysis (PCA) is a kind of data analysing method that K.Pearson proposes, and object is from primitive character, to calculate one group of new feature of arranging from big to small by importance, and they are the linear combination of primitive character, and uncorrelated mutually.

If new feature is y _i, i=1,2 ..., p, is the linear combination of above-mentioned primitive character

$y_i=\underset{j=1}{\overset{p}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_{\mathrm{ij}}{x}_j={{\mathrm{\&alpha;}}_i}^{T}x$

For unified y _iyardstick, the mould that might as well set linear combination coefficient is 1,

α _i ^Tα _i＝1

As follows by matrix representation

y＝A ^Tx

Wherein, y is new feature y _ithe proper vector forming, A is eigentransformation matrix.Need to solve optimum orthogonal transform matrix A, make new feature y _ivariance reach extreme value.

The covariance matrix of x is made as ∑, with training sample, estimates.

μ＝E{x}

∑＝E{(x-μ)(x-μ) ^T}

Covariance matrix ∑ has p eigenvalue λ _i, i=1,2 ..., p (comprise may equate and may be 0 eigenvalue), sequence is from big to small λ ₁>=λ ₂>=...>=λ _p.

PCA, as a kind of feature extracting method, is to represent data by less major component, gets a front k major component, and the variance of their representative datas accounts for the ratio of population variance and is so

$\frac{{\mathrm{\&Sigma;}}_{i=1}^k{\mathrm{\&lambda;}}_i}{{\mathrm{\&Sigma;}}_{i=1}^p{\mathrm{\&lambda;}}_i}$

In the present invention, aforementioned proportion is made as to 90%, can calculates accordingly the k value of above formula, and then realization character dimensionality reduction.

504: final characteristic storage, in two-dimensional matrix, is designated as to F ⁱ _mk, i=1,2 ..., 44, wherein, the ecg wave form number that m is user, the major component number of k for adopting.

Electrocardio template database builds module.Adopt the ECG data clusters of ISODATA algorithm to user, in classification, pass judgment under criterion and obtain rational classification results.By dependent thresholds method, in each categorical data, select typical ECG signal, and then generate such other template, the electrocardiogram (ECG) data of all each classifications of user is done to same treatment, build registered user's electrocardio template database.Electrocardio template database builds the process flow diagram of module referring to Fig. 6, specifically comprises:

601:ISODATA algorithm cluster.ISODATA (Iterative Self-Organizing Data Analysis Techniques, iteration self-organization data analysis technique) can be regarded as a kind of improved C means clustering algorithm.This algorithm is to count all kinds of average of grate after whole sample adjustment, can improve operation efficiency like this, in addition, in cluster process, introduce the judge criterion to classification, whereby can be automatically by some categories combination or division, thereby obtain more rational cluster result, also broken through to a certain extent the restriction of prior given class number.

For user User _i, establish by N ECG data and form sample set, by proper vector separately, represent, be designated as after ISODATA algorithm cluster, obtain c cluster centre, use m _j, j=1,2 ..., c represents.

602: dependent thresholds method is selected template.In order to reduce the time complexity of algorithm, adopt dependent thresholds method to select typical QRS complex wave as such matching template in every class ECG data of user, specific implementation process is as follows:

1. user User _iclassification Г _javerage m _jfor

$m_j=\frac{1}{N_j}\underset{F^i\&Element;{\mathrm{\&Gamma;}}_j}{\mathrm{\&Sigma;}}{F}^i,j=\mathrm{1,2},...,c$

Wherein, N _jit is the number of samples of j cluster.

2. at classification Г _jmiddle selection K apart from classification center m _jqRS complex wave corresponding to nearest sample, as template to be selected, is arranged in order from small to large by distance, is designated as QRS ⁱ _st, i=1,2 ..., 44s=1,2 ..., K.

3. might as well select QRS ⁱ _1tas benchmark template, calculate the related coefficient with K-1 template to be selected of residue

$r_s({{\mathrm{QRS}}^i}_{1t},{{\mathrm{QRS}}^i}_{\mathrm{st}})=\frac{\mathrm{Cov}({{\mathrm{QRS}}^i}_{1t},{{\mathrm{QRS}}^i}_{\mathrm{st}})}{\sqrt{D\left({{\mathrm{QRS}}^i}_{1t}\right)\&CenterDot;D\left({{\mathrm{QRS}}^i}_{\mathrm{st}}\right)}},s=\mathrm{2,3},...,K$

Wherein, Cov (QRS ⁱ _1t, QRS ⁱ _st) be template QRS to be selected ⁱ _1tand QRS ⁱ _stcovariance, D (QRS ⁱ _1t) and D (QRS ⁱ _st) be respectively QRS ⁱ _1tand QRS ⁱ _stvariance.

Cov(QRS ⁱ _1t，QRS ⁱ _st)＝E{[QRS ⁱ _1t-E(QRS ⁱ _1t)][QRS ⁱ _st-E(QRS ⁱ _st)]}

D(QRS ⁱ _1t)＝E{|QRS ⁱ _1t-E(QRS ⁱ _1t)| ²}

D(QRS ⁱ _st)＝E{|QRS ⁱ _st-E(QRS ⁱ _st)| ²}

And E (QRS ⁱ _1t) and E (QRS ⁱ _st) be respectively QRS ⁱ _1tand QRS ⁱ _staverage.

603: build electrocardio template base.In order to build electrocardio template database, set classification Γ _jstencil-chosen threshold value be Th ^j ₁, as select the thresholding of typical template from template to be selected, using the mean value of related coefficient as threshold value Th ^j ₁

${{\mathrm{Th}}^j}_1=\frac{1}{K-1}\underset{s=2}{\overset{K}{\mathrm{\&Sigma;}}}{r}_s({{\mathrm{QRS}}^i}_{1t},{{\mathrm{QRS}}^i}_{\mathrm{st}})$

Set such Г _jtemplate matches threshold value be Th ^j ₂, as the criterion of ECG test data template matches success or not, using the minimum value of related coefficient as threshold value Th ^j ₂

${{\mathrm{Th}}^j}_2=\underset{s}{\min }{r}_s({{\mathrm{QRS}}^i}_{1t},{{\mathrm{QRS}}^i}_{\mathrm{st}})$

ECG data to all users are done above-mentioned processing, obtain characterizing a plurality of templates of every class electrocardiogram (ECG) data, are designated as Temp ⁱ _jk, i=1,2 ..., 44, wherein j is User _icluster numbers, and the template number that k is j class.

Template matches and identification module.User User _ii=1,2, ..., 44 electrocardiogram (ECG) datas of latter 10 minutes are as test data, after data preprocessing module and signal decomposition and standardized module processing, obtain ECG standardized data, still utilize R crest location information extraction QRS complex wave, by dependent thresholds method, select ECG test data, in conjunction with the electrocardio template database building, do correlation analysis, and then obtain the optimum matching template of each test data, if corresponding related coefficient is greater than matching threshold, show that the match is successful, otherwise be considered as invalid data, refused.After all test datas have been mated, then consider the matching result of all test datas of same user, provide final identification result.The process flow diagram of template matches and identification module, referring to Fig. 7, specifically comprises:

701: extract QRS complex wave.For user User _ll=1,2 ..., 44, choose after corresponding electrocardiogram (ECG) data file the data of 10 minutes as test data, after data preprocessing module and signal decomposition and standardized module are processed, obtain ECG standardized data.

Take R crest location as benchmark extracts ECG signal QRS complex wave, due to electrocardiogram (ECG) data standardization, therefore, only need utilize R crest location and RR interval mean value t _rRextract.If current ECG signal R crest location is t _{r_}position, with 0.15*t _rRduration intercept forward and backward respectively one piece of data as QRS complex wave.For user User _ll=1,2 ..., 44 electrocardiogram (ECG) data, the QRS complex wave of extraction is stored with two-dimensional matrix, is designated as QRS ^l _uv, l=1,2 ..., 44, wherein u is QRS waveform number, v is the sampling number of QRS waveform.

702: dependent thresholds method is selected test data.For user's QRS complex wave, choose at random K waveform as test data to be selected, be designated as QRS ^l _uv, l=1,2 ..., 44u=1,2 ..., K, for guaranteeing the similarity of waveform and the typicalness of test data, adopts correlation coefficient threshold method to obtain final ECG test data.Might as well choose QRS ^l _1vas reference data, calculate the correlation coefficient r with K-1 data of residue _u(QRS ^l _1v, QRS ^l _uv), u=2,3 ..., K, the threshold value Th selecting using mean value as test data ^l

${\mathrm{Th}}^l=\frac{1}{K-1}\underset{u=2}{\overset{K}{\mathrm{\&Sigma;}}}{r}_u({{\mathrm{QRS}}^l}_{1v},{{\mathrm{QRS}}^l}_{\mathrm{uv}})$

ECG data to all users are done above-mentioned processing, obtain characterizing the ECG test data of user identity, are designated as Test ^l _uv, l=1,2 ..., 44, wherein u is User _lqRS complex wave number.

703: correlation analysis.So far, obtain 44 users' ECG template data Temp ⁱ _jk, i=1,2 ..., 44 and test data Test ^l _uv, l=1,2 ..., 44.

U the electrocardio test data Test for l position user to be identified ^l _u, the related coefficient of whole template datas, i.e. r in calculating and each class template of all registered users _{i, j, k}(Test ^l _u, Temp ⁱ _jk).

704: obtain optimum matching template.Adopt Correlation Coefficient Criteria to select the optimum matching template of test data, first in user's similar electrocardiogram (ECG) data, select the template of related coefficient maximum, i.e. max _k{ r _{i, j, k}(Test ^l _u, Temp ⁱ _jk), then in the every class electrocardiogram (ECG) data of user, select, i.e. max _j{ max _k{ r _{i, j, k}(Test ^l _u, Temp ⁱ _jk), finally in all users' electrocardiogram (ECG) data, select optimum matching template, i.e. max _i{ max _j{ max _k{ r _{i, j, k}(Test ^l _u, Temp ⁱ _jk).So far, through above-mentioned three step maximizings, process, for each test data of user to be identified finds optimum matching template.

705: consider template matches situation and draw identification result.For each test data of user to be identified searches after optimum matching template, if corresponding related coefficient is greater than default template matches threshold value, show that the match is successful, otherwise algorithm is using this test data as invalid data, is refused.

User to be identified has a plurality of ECG test datas, therefore need to consider the template matches situation of same user's total data, and the weight of matching result is directly proportional to corresponding facies relationship numerical value, obtains accordingly final identification result.To user User _ll=1,2 ..., 44 test data Test ^l _u, corresponding optimum matching template is designated as Temp ⁱ _j, i=1,2 ..., 44.

If consider the coupling of single ECG test data to template, be designated as

$\underset{u,i,j}{I}\mathrm{match}({{\mathrm{Test}}^l}_u,{{\mathrm{Temp}}^i}_j)$

If consider the coupling of single ECG test data to registered user, be designated as

$\underset{u,i}{I}\mathrm{match}({{\mathrm{Test}}^l}_u,{\mathrm{Temp}}^i)=\underset{i}{\max }\left\{\underset{j}{\max }\right\{\underset{k}{\max }\left\{\underset{i,j,k}{r}({{\mathrm{Test}}^l}_u,{{\mathrm{Temp}}^i}_{\mathrm{jk}})\right\}\left\}\right\}$

If consider the coupling of user to be identified to registered user, be designated as

$\underset{l,i}{I}\mathrm{match}({\mathrm{Test}}^l,{\mathrm{Temp}}^i)=\underset{i}{\max }\{i*\underset{u,i}{I}\mathrm{match}({{\mathrm{Test}}^l}_u,{\mathrm{Temp}}^i)\}$

Finally, algorithm is exported final identification result.

Claims (7)

1. the multi-template matching personal identification method based on ECG under anomalous ecg state, is characterized in that, comprises the steps:

Step 1: raw ECG data acquisition, pre-service;

Step 2: electrocardiosignal decomposes and standardization;

Step 3: wavelet transformation extracts the characteristic feature of electrocardiosignal, and principal component analysis (PCA) is carried out dimensionality reduction to feature space, obtains characterizing the proper vector of ECG signal;

Step 4: utilize ISODATA algorithm to ECG data clusters, and then build registered user ECG template base;

Step 5: user's to be identified ECG test data is mated with the electrocardio template in template base one by one, adopt related coefficient as similarity criteria, obtain identification result.

2. the multi-template matching personal identification method based on ECG under anomalous ecg state according to claim 1, is characterized in that, described personal identification method comprises following five modules:

2.1 electrocardiosignal pretreatment module;

2.2 signal decomposition and standardized module;

2.3 electrocardiosignal characteristic extracting module;

2.4 electrocardio template databases build module;

2.5 template matches and identification module.

3. the electrocardiosignal pretreatment module that the personal identification method based on ECG according to claim 2 comprises, is characterized in that, comprises following steps:

3.1 obtain the multistage electrocardiogram (ECG) data under user's steady state (SS) with certain time length;

3.2 pairs of every section of electrocardiogram (ECG) datas carry out pre-service, design suitable wave filter, the noises such as the interference of filtering power frequency, baseline wander and myoelectricity interference.

4. the personal identification method based on ECG according to claim 2 comprises signal decomposition and standardized module, is characterized in that, comprises following steps:

4.1 couples of users' ECG data, detect R crest location;

4.2 adopt the method based on R crest location and adjacent R R interval, isolate the waveform of each cardiac electrical cycle;

4.3 standardizations make it to have identical length in time scale, make it to have equal maximum voltage value on amplitude yardstick.

5. the electrocardiosignal characteristic extracting module that the personal identification method based on ECG according to claim 2 comprises, is characterized in that, comprises following steps:

5.1 utilize the information of R crest location and RR interval, extract the QRS complex wave of every bat ecg wave form;

5.2 extract the primitive character of ecg wave form based on wavelet transformation;

5.3 adopt principal component analysis (PCA) to carry out dimensionality reduction to ecg characteristics space, obtain characterizing the proper vector of ecg wave form.

6. the ecg characteristics template base that the personal identification method based on ECG according to claim 2 comprises builds module, it is characterized in that, comprises following steps:

6.1 adopt ISODATA algorithm by same user's ECG data clusters, obtain c class data;

6.2 templates to be selected using QRS complex wave as every class ECG data;

6.3 adopt dependent thresholds method, choose the typical QRS complex wave of suitable number, as the matching template of such ECG data from every class template to be selected;

6.4 do same treatment to each registered user, thereby build electrocardio template database.

7. the personal identification method based on ECG according to claim 2 comprises template matches and identification module, is characterized in that, comprises following steps:

7.1 obtain the electrocardiogram (ECG) data that next section of user's steady state (SS) to be identified has certain time length, complete the steps such as data pre-service, electrocardiosignal decomposition, standardization and the extraction of QRS complex wave;

7.2 adopt dependent thresholds method, and the QRS complex wave of choosing suitable number and can characterizing user to be identified is as ECG test data;

7.3 mate each test data of user to be identified with all templates in ECG template base;

7.4 adopt related coefficient as template matches similarity criteria, find optimum matching template, consider the match condition of each test data, determine user's to be identified identity.