CN114419657A - ECG identity recognition method based on local segment sparse representation - Google Patents

ECG identity recognition method based on local segment sparse representation Download PDF

Info

Publication number
CN114419657A
CN114419657A CN202111495097.4A CN202111495097A CN114419657A CN 114419657 A CN114419657 A CN 114419657A CN 202111495097 A CN202111495097 A CN 202111495097A CN 114419657 A CN114419657 A CN 114419657A
Authority
CN
China
Prior art keywords
matrix
ecg
sparse
sparse representation
dictionary
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
Application number
CN202111495097.4A
Other languages
Chinese (zh)
Inventor
秦曙光
郑强荪
董宁
李宁
申盼盼
孙雪敏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xian University of Technology
Second Affiliated Hospital School of Medicine of Xian Jiaotong University
Original Assignee
Xian University of Technology
Second Affiliated Hospital School of Medicine of Xian Jiaotong University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Xian University of Technology, Second Affiliated Hospital School of Medicine of Xian Jiaotong University filed Critical Xian University of Technology
Priority to CN202111495097.4A priority Critical patent/CN114419657A/en
Publication of CN114419657A publication Critical patent/CN114419657A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Probability & Statistics with Applications (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)

Abstract

The invention discloses an ECG (electrocardiogram) identity recognition method based on local segment sparse representation, which comprises the steps of collecting electrocardiosignals, denoising the signals by adopting wavelet transform based on a weight threshold value to obtain denoised ECG signals, dividing the denoised ECG signals by adopting a sliding window method, extracting local segments, obtaining local characteristics of each signal by adopting a principal component analysis method to realize the dimensionality reduction of the data, searching optimal matching atoms in the sparse representation process of the data subjected to dimensionality reduction by adopting an orthogonal matching tracking algorithm, constructing a dictionary for sparse representation by adopting a K singular value decomposition algorithm to obtain a processed sparse coefficient matrix, and carrying out probability neural network recognition on the final sparse coefficient matrix to obtain the recognition precision of the ECG signals. The invention can well capture global and local information, and the sparse representation and the dictionary construction can improve the reliability of the identification based on the electrocardiosignal.

Description

ECG identity recognition method based on local segment sparse representation
Technical Field
The invention belongs to the technical field of biological feature recognition, and relates to an ECG identity recognition method based on local segment sparse representation.
Background
In recent years, scientific technology is the first productivity of China, and people's lives also tend to be informationized gradually. The prior means for identifying identity are based on knowledge, and some means are based on tokens, and the methods gradually have some disadvantages in the aspects of related use, such as that secret numbers, passwords and the like are easy to forget, or the conditions of stealing, tampering, transaction, sharing and the like occur like certificates for identity authentication, so that the biometric data serving as the basis of identity identification is safer and more reliable.
The biological characteristics have the characteristics of uniqueness, stability, portability, good anti-counterfeiting performance and the like, and become a new medium in the field of personal identification. Currently, the use of physiological methods such as face recognition, fingerprint recognition, and voice recognition in society has been widely studied. The facial recognition is realized by utilizing the facial features of people, the fingerprint recognition is realized by utilizing the fingerprint uniqueness of each person to ensure the distinguishability between people, and the voice recognition is realized according to the different tone characteristics of each person. However, with the development of the related art, biometric features such as face, fingerprint and voice are counterfeit, which results in the security and reliability of biometric identification being reduced, so it is necessary to find a non-counterfeit and distinguishable biometric feature for identification.
The development of technology has led to the rise of living standard of people, and simultaneously, many bad living habits are caused, which can cause some heart diseases. Electrocardiograms are an effective means for diagnosing heart diseases. Originally, cardiac electrical signals were intended to diagnose diseases, but have been widely used for biometric identification in the past decade because multiple functions of cardiac electrical signals enable biometric identification. Different individuals have different physiological and cardiac structures, so that the individuals can be identified from the electrocardiosignals. Other conventional biological features such as gait, face, voice, etc. are easily marked in life, and although they are widely used in the market, they are not very safe because they are easily collected by cameras and recording devices, and can be copied. In contrast, the electrocardiosignal belongs to an internal physiological characteristic, which means that the safety of the electrocardiosignal is higher. Moreover, most of the devices for collecting electrocardiosignals depend on medical devices, are not easy to steal, and have weak signals, low frequency and variability, so that the electrocardiosignals are not easy to copy. Therefore, the identification based on the electrocardiosignals is more beneficial to the information management of the modern society.
Although people have achieved certain achievements in the aspect of electrocardiosignal identity recognition, when the existing electrocardiosignal identity recognition method is adopted for identity recognition, the electrocardiosignal identity recognition method is affected by the heart rate change of a human body, the environment acquisition equipment and the like, certain deviation occurs in identity recognition, and the reliability of the electrocardiosignal identity recognition is reduced.
Disclosure of Invention
The invention aims to provide an ECG (electrocardiogram) identity recognition method based on local segment sparse representation, which solves the problem of larger deviation of the existing electrocardiosignal identity recognition method.
The technical scheme adopted by the invention is that the ECG identity recognition method based on the sparse representation of the local segments comprises the following steps:
step 1, acquiring electrocardiosignals, and denoising the signals by adopting wavelet transform based on a weight threshold value to obtain denoised ECG signals;
step 2, dividing the denoised ECG signal by adopting a sliding window method, extracting local segments, and then obtaining the local characteristics of each signal by adopting a principal component analysis method to realize the dimensionality reduction of data;
step 3, searching optimal matching atoms in the sparse representation process of the data subjected to dimensionality reduction by adopting an orthogonal matching pursuit algorithm, and constructing a sparse representation dictionary by adopting a K singular value decomposition algorithm to obtain a processed sparse coefficient matrix;
and 4, carrying out probabilistic neural network identification on the final sparse coefficient matrix obtained in the step 3 to obtain the identification precision of the ECG signal.
Wherein, the specific process of the step 1 is as follows:
step 1.1, obtaining original electrocardiographic data in a mode of equipment reading or database acquisition, and then carrying out drawing processing on the obtained original electrocardiographic data by using a drawing algorithm to obtain a matrix storing ECG data, wherein the matrix is an ECG signal needing to be processed;
and step 1.2, decomposing and reconstructing the ECG signal obtained in the step 1.1 by adopting a denoising method based on wavelet weight threshold shrinkage to obtain a denoised ECG signal.
The specific process of step 1.2 is as follows:
step 1.2.1, decomposing the ECG signal obtained in the step 1.1 by using a mallat wavelet algorithm to obtain wavelet coefficient values of each layer;
step 1.2.2, screening the wavelet coefficient value obtained in the step 1.2.1 by using a dynamic soft threshold formula to obtain a processed wavelet coefficient;
and step 1.2.3, combining the wave coefficients processed in the step 1.2.2 into a wave structure to receive signals, and obtaining denoised ECG signals.
The dynamic soft threshold formula in step 1.2.2 is:
Figure BDA0003399823890000041
in which j represents the decomposed scale, TjRepresents the critical threshold, W represents the wave coefficient, sign (ω) is a sign function.
The specific process of step 2 is as follows:
step 2.1, dividing the denoised ECG signal by adopting a sliding window method, and extracting local segments to obtain a segmentation matrix corresponding to the local segments;
and 2.2, processing the segmented matrix obtained in the step 2.1 by using a principal component analysis method to obtain the local characteristics of each signal, thereby realizing the dimension reduction of the data.
The specific process of step 2.2 is as follows:
step 2.2.1, calculating the mean value of each row of the segmented matrix obtained in the step 2.1, and then subtracting the corresponding mean value from each row of data to calculate a characteristic covariance matrix;
step 2.2.2, solving an eigenvalue and an eigenvector according to the characteristic covariance matrix, and then carrying out normalization processing on the eigenvector;
and 2.2.3, selecting the eigenvector corresponding to the maximum eigenvalue obtained in the step 2.2.2, and projecting the sample point onto the selected eigenvector to obtain a matrix Y after dimensionality reduction.
The specific process of step 3 is as follows:
3.1, searching the optimal matching atom in the matrix Y in the sparse representation process by adopting an orthogonal matching pursuit algorithm on the matrix Y after the dimension reduction;
and 3.2, constructing a dictionary of sparse representation by adopting a K singular value decomposition algorithm according to the optimal matching atoms to obtain a processed sparse coefficient matrix.
The specific process of step 3.2 is as follows:
step 3.2.1, describing the optimal matching atom obtained in the step 3.1 by using a linear expression of a basic element to form a basic atom dictionary, initializing the dictionary, and sparsely representing a given sample by using the dictionary to obtain a corresponding matrix;
step 3.2.2, when the matrix obtained in the step 3.2.1 is sparse, the error between the sample data and the data expressed by the coefficient matrix is continuously reduced by the iterative dictionary column by column, and the optimal base atom dictionary is infinitely approximated;
step 3.2.3, updating the basic atoms obtained in the step 3.2.2 row by row;
step 3.2.4, after the dictionary is updated in step 3.2.3, performing sparse coding on the new dictionary, and stopping updating the dictionary when the iteration number reaches a specified value or the error rate reaches a specified range, so as to obtain a processed sparse coefficient matrix, that is, after sparse representation, reconstructed electrocardiosignals Y 'are Dx, Y' are to-be-processed dimensionality-reduced data signal matrices, and D belongs to Rm×nIs a dictionary matrix, and x is a sparse coefficient x obtained after processing and belongs to Rm。
The specific process of step 4 is as follows:
step 4.1, inputting the final sparse coefficient matrix obtained in the step 3 into an input layer of the probabilistic neural network, and then entering a hidden layer;
step 4.2, inputting the data output from the hidden layer into the summation layer for calculation to obtain the maximum category calculated in the summation layer;
and 4.3, normalizing the maximum category data calculated in the summation layer to obtain probability estimation of each category, namely the identification precision of the ECG signal.
The method has the advantages that the local features are extracted from the training signals to construct a compact and discriminable dictionary, the method can well capture global and local information, and the reliability of the identification based on the electrocardiosignal identity can be improved by sparse representation and dictionary construction; the filtering processing of the electrocardiosignals can be effectively finished by using a soft threshold wavelet denoising method to denoise the signals, the algorithm is stable, and the identification accuracy is high; the ECG identification recognition method based on the ECG comprises the steps of preprocessing of the electrocardiosignals, feature extraction and classification recognition, and is added with the link of sparse representation, so that invalid information of the electrocardiosignals can be well filtered, the complexity of the signals is reduced, and meanwhile, the cost and the time are reduced.
Drawings
FIG. 1 is a general flow chart of the ECG identification method based on sparse representation of local segments according to the present invention;
FIG. 2 is a specific flow chart of wavelet threshold denoising in the ECG identification method based on local segment sparse representation according to the present invention;
FIG. 3 is a specific flow chart of feature extraction of a central electrical signal of the ECG identification method based on local segment sparse representation according to the present invention;
FIG. 4 is a flow chart of sparse representation in the ECG identification method based on sparse representation of local segments according to the present invention;
FIG. 5 is a diagram of the basic structure of the Probabilistic Neural Network (PNN) used in the ECG identification method based on sparse representation of local segments according to the present invention.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
The invention relates to an ECG (electrocardiogram) identity recognition method based on sparse representation of local segments, which comprises the following steps of:
step 1, referring to fig. 2, acquiring an electrocardiosignal, and denoising the electrocardiosignal by adopting wavelet transform based on a weight threshold value to obtain a denoised ECG signal, wherein the specific process is as follows:
step 1.1, acquiring electrocardio original data through an arrhythmia electrocardio database (MIT-BIH), and then drawing the acquired electrocardio original data by using a drawing algorithm to obtain a matrix in which ECG data is stored, wherein the matrix is an ECG signal to be processed;
step 1.2, decomposing and reconstructing the ECG signal obtained in the step 1.1 by adopting a denoising method based on wavelet weight threshold shrinkage to obtain a denoised ECG signal, wherein the specific process is as follows:
step 1.2.1, firstly selecting a wavelet base scale function db5 which can most highlight the characteristics of the ECG signal, simultaneously selecting a decomposition principle of 8 scales, and applying a fast binary orthogonal wavelet transform algorithm based on multi-resolution analysis, namely performing wavelet decomposition and reconstruction of 8 scales on the ECG signal obtained in the step 1.1 by using a mallat wavelet algorithm to obtain wavelet coefficient values of each layer;
step 1.2.2, screening the wavelet coefficient value obtained in the step 1.2.1 by using a dynamic soft threshold formula to obtain a processed wavelet coefficient;
the dynamic soft threshold formula in step 1.2.2 is:
Figure BDA0003399823890000071
in which j represents the decomposed scale, TjRepresenting a critical threshold, W represents a wave coefficient, sign (ω) is a sign function;
and step 1.2.3, combining the wave coefficients processed in the step 1.2.2 into a wave structure to receive signals, and obtaining denoised ECG signals.
Step 2, referring to fig. 3, dividing the denoised ECG signal by using a sliding window method, extracting local segments, then obtaining local features of each signal by using a principal component analysis method, and implementing the dimensionality reduction of data, specifically:
step 2.1, adopting a non-reference point-based method, namely a sliding window method, on the denoised ECG signal, firstly presetting a window with a fixed length, wherein the width of the sliding window is more than the width of a period, then moving from the head section to the tail end of the denoised ECG signal obtained in the step 1, extracting local segments, and obtaining a segmentation matrix X corresponding to the local segmentsm×n＝{x1,x2,...,xn}，xnFor each column of data of the data matrix;
step 2.2, processing the segmented matrix obtained in the step 2.1 by using a principal component analysis method to obtain local characteristics of each signal, and realizing the dimension reduction of data, wherein the specific process is as follows:
step 2.2.1, calculating the mean value of each row of the segmented matrix obtained in the step 2.1
Figure BDA0003399823890000081
Then, the corresponding mean value is subtracted from each column of data to obtain
Figure BDA0003399823890000082
Finally, the characteristic covariance matrix is solved
Figure BDA0003399823890000083
Step 2.2.2, solving an eigenvalue and an eigenvector according to the characteristic covariance matrix: a. thecovλ＝λv,Acov＝E∑E-1Sigma is a diagonal matrix, lambda is an eigenvalue, v is an eigenvector, and the eigenvector is normalized to obtain a matrix Em×n'；
Step 2.2.3, selecting eigenvectors corresponding to the largest eigenvalues obtained in the step 2.2.2 to form a matrix En×k' projecting the sample point to the selected characteristic vector to obtain the matrix Y ═ X after dimension reductionm×k＝Xm×n×En×k′。
Step 3, referring to fig. 4, for the dimensionality reduced data, an orthogonal matching pursuit algorithm is adopted to search for an optimal matching atom in the sparse representation process, a K singular value decomposition algorithm is adopted to construct a sparse representation dictionary, and a processed sparse coefficient matrix is obtained, wherein the specific process is as follows:
step 3.1, searching the optimal matching atoms in the matrix Y in the sparse representation process by adopting an orthogonal matching pursuit algorithm on the matrix Y after the dimension reduction, specifically comprising the following steps:
step 3.1.1, Y is the raw signal to be decomposed from step 2, D ═ xi}∈Rn×KAnd R × f is a data set of the signal after the kth iteration, wherein the overcomplete dictionary matrix is formed, norms of atoms in D are all equal to 1. When initializing, can let f0＝0，R′f＝f，x＝0，a0And k is 0. It can be assumed that the signal data set after the decomposition in the k-th step becomes:
Figure BDA0003399823890000091
and xn，Rkf＝0n＝1,2,,k...an kRepresenting the coefficient obtained after the k-th decomposition;
,
step 3.1.2, it can be deduced from the formula (3-1) obtained in step 3.1.1 that at step k +1, the above formula can be decomposed to obtain:
Figure BDA0003399823890000092
and is<xn,Rk+1f>＝0 n＝1,2,k+1... (3-2)
Figure BDA0003399823890000093
And is<γk,xn>＝0,n＝1,2,...,k (3-3)
Wherein the content of the first and second substances,
Figure BDA0003399823890000094
represents xk+1At { x1,x2,...,xkAnd (c) projecting.
Figure BDA0003399823890000095
Represents xk+1Relative to matrix { x1,x2,...,xkThe resulting components are mapped vertically, wherein,
an k+1＝an k-akbn kn is 1,2,.., k, and
Figure BDA0003399823890000096
Figure BDA0003399823890000097
step 3.1.3, based on the remaining unprocessed signal R in formula (3-2) in step 3.1.2k+1f has certain satisfied conditions, can get:
Rk+1f＝Rkf-αkγkand is and
Figure BDA0003399823890000101
and when the iteration meets the requirement of the control condition in the k times of circulation, ending the iteration: k is K, and K is sparsity, so that the optimal matching atom can be obtained;
step 3.2, constructing a sparse representation dictionary by adopting a K singular value decomposition algorithm according to the optimal matching atoms to obtain a processed sparse coefficient matrix, wherein the method specifically comprises the following steps:
step 3.2.1, the optimal matching atom obtained in step 3.1 is described by a linear expression of a base element to form a base atom dictionary, the dictionary is initialized, a given sample is sparsely represented by the dictionary to obtain a corresponding matrix, and the base atom dictionary is multiplied by the sparse matrix, namely:
Figure BDA0003399823890000102
dicolumn, x, representing DiRepresenting the rows of X, and then updating the columns of the dictionary.
Step 3.2.2, the matrix obtained in step 3.2.1 is often not optimal, there is a certain error between the sample data and the data expressed by using the coefficient matrix, when the matrix obtained in step 3.2.1 is sparse, the column-by-column iterative dictionary is used to continuously reduce the error between the sample data and the data expressed by using the coefficient matrix, the optimal base atom dictionary is infinitely approximated, and the formula for calculating the error is as follows:
Figure BDA0003399823890000103
step 3.2.3, updating the base atoms obtained in the step 3.2.2 column by column because the dictionary is updated by column units; that is, assuming that X and D are known before derivation, column-by-column updating the kth column of the dictionary yields DkCorresponding coefficient ofThe k-th row in the matrix X is then Xk TThus, the multiplication term of the formula (3-8) obtained in step 3.2.2 can be rewritten into the formula:
Figure BDA0003399823890000111
solving the formula (3-9), if the decomposition is directly carried out by SVD, the obtained result is diverged, and the formula (3-9) is converted to obtain the formula shown as follows:
Figure BDA0003399823890000112
in the formula (3-10), ΩkIs Nx | omegakI matrix, the characteristics of which are defined in (omega)k(i) The values at i) are all equal to 1 and the other points are all equal to 0, defined by the characteristics of their matrix
Figure BDA0003399823890000113
Wherein
Figure BDA0003399823890000114
Are respectively
Figure BDA0003399823890000115
Y,EkThe contraction result obtained after removing zero input vector, pair EkPerforming SVD decomposition to obtain ER k＝UΔVTUpdating the dictionary column by column;
step 3.2.4, use the new dictionary after the dictionary updating is finished
Figure BDA0003399823890000116
Making sparse coding, and stopping updating the dictionary after the iteration times reach a specified value or the error rate reaches a specified range, so as to obtain a processed sparse coefficient matrix, namely, after sparse representation, the reconstructed electrocardiosignal Y 'is Dx, Y' is a data signal matrix to be processed after dimensionality reduction, and D belongs to Rm×nIs a dictionary matrix, and x is a sparse coefficient x obtained after processingRm。
Step 4, referring to fig. 5, performing Probabilistic Neural Network (PNN) recognition on the final sparse coefficient matrix obtained in step 3 to obtain the ECG signal recognition accuracy, which specifically comprises the following steps:
step 4.1, inputting the final sparse coefficient matrix obtained in the step 3 into an input layer of a Probabilistic Neural Network (PNN), and then entering a hidden layer;
PNN by input layer x1,x2,x3,...,xfHidden layer (calculated distance), summation layer Ψ1,Ψ2,Ψ3,...,ΨfAnd the output layer y is sequentially arranged from left to right, the sample signals firstly enter the hidden layer through the input layer, and the distance between the vectors and the center of the neuron can be calculated after input data is detected because each neuron node of the hidden layer has a central point. The input layer vector is first multiplied by a weighting value, calculated using the corresponding function:
Zi＝xqi (4-1)
wherein, i is 1,2, and N, N is the total class of training data, and if the input layer vector and the weighting value are normalized, the data relationship determined by the jth pattern neuron of the ith class of the hidden layer is determined by the formula (4-2), as follows:
Figure BDA0003399823890000121
step 4.2, inputting the data output from the hidden layer into a summation layer for calculation,
Figure BDA0003399823890000122
wherein v isiThe calculation result of the ith class is output, H is the number of neurons of the ith class, and the number N of the classes is the same as the number of the neurons of the summation layer;
the maximum class of computation in the summation layer is obtained,
y＝argmax(vi) (4-4)
in the above formula (4-2), σ is a smoothing factor, which affects the final network and the calculation result, wherein each kind of the summation layer has a neuron corresponding to it, each kind of the hidden layer also has a neuron corresponding to it, PNN uses the mechanism of Online superior Learning, and the data obtained by the output layer is positively correlated with the probability estimates of different kinds;
and 4.3, normalizing the maximum category data calculated in the summation layer to obtain probability estimation of each category, namely the identification precision of the ECG signal.
The number of the categories is the same as the number of the neurons of the summation layer, and the classification and identification mainly obtains the output of the summation layer and distinguishes samples according to the output. Some of all neurons in the hidden layer have the highest probability density, the output of the neurons is 1, other neurons are 0, the final identification category of the sample is finally output, and a label vector is obtained, and under the condition of the optimal parameters, the identification accuracy after sparse representation is 95.33% according to the label vector.

Claims (9)

1. The ECG identification method based on the sparse representation of the local segments is characterized by comprising the following steps of:
step 1, acquiring electrocardiosignals, and denoising the signals by adopting wavelet transform based on a weight threshold value to obtain denoised ECG signals;
step 2, dividing the denoised ECG signal by adopting a sliding window method, extracting local segments, and then obtaining the local characteristics of each signal by adopting a principal component analysis method to realize the dimensionality reduction of data;
step 3, searching optimal matching atoms in the sparse representation process of the data subjected to dimensionality reduction by adopting an orthogonal matching pursuit algorithm, and constructing a sparse representation dictionary by adopting a K singular value decomposition algorithm to obtain a processed sparse coefficient matrix;
and 4, carrying out probabilistic neural network identification on the final sparse coefficient matrix obtained in the step 3 to obtain the identification precision of the ECG signal.
2. The ECG identification method based on local segment sparse representation according to claim 1, wherein the specific process of the step 1 is as follows:
step 1.1, obtaining original electrocardiographic data in a mode of equipment reading or database acquisition, and then carrying out drawing processing on the obtained original electrocardiographic data by using a drawing algorithm to obtain a matrix storing ECG data, wherein the matrix is an ECG signal needing to be processed;
and step 1.2, decomposing and reconstructing the ECG signal obtained in the step 1.1 by adopting a denoising method based on wavelet weight threshold shrinkage to obtain a denoised ECG signal.
3. An ECG identification method based on local segment sparse representation according to claim 2, wherein the specific process of the step 1.2 is as follows:
step 1.2.1, decomposing the ECG signal obtained in the step 1.1 by using a mallat wavelet algorithm to obtain wavelet coefficient values of each layer;
step 1.2.2, screening the wavelet coefficient value obtained in the step 1.2.1 by using a dynamic soft threshold formula to obtain a processed wavelet coefficient;
and step 1.2.3, combining the wave coefficients processed in the step 1.2.2 into a wave structure to receive signals, and obtaining denoised ECG signals.
4. The method for ECG identification based on sparse representation of local segments according to claim 3, wherein the dynamic soft threshold formula in step 1.2.2 is as follows:
Figure FDA0003399823880000021
in which j represents the decomposed scale, TjRepresents the critical threshold, W represents the wave coefficient, sign (ω) is a sign function.
5. The ECG identification method based on local segment sparse representation according to claim 1, wherein the specific process of the step 2 is as follows:
step 2.1, dividing the denoised ECG signal by adopting a sliding window method, and extracting local segments to obtain a segmentation matrix corresponding to the local segments;
and 2.2, processing the segmented matrix obtained in the step 2.1 by using a principal component analysis method to obtain the local characteristics of each signal, thereby realizing the dimension reduction of the data.
6. An ECG identification method based on local segment sparse representation according to claim 5, wherein the specific process of the step 2.2 is as follows:
step 2.2.1, calculating the mean value of each row of the segmented matrix obtained in the step 2.1, and then subtracting the corresponding mean value from each row of data to calculate a characteristic covariance matrix;
step 2.2.2, solving an eigenvalue and an eigenvector according to the characteristic covariance matrix, and then carrying out normalization processing on the eigenvector;
and 2.2.3, selecting the eigenvector corresponding to the maximum eigenvalue obtained in the step 2.2.2, and projecting the sample point onto the selected eigenvector to obtain a matrix Y after dimensionality reduction.
7. The ECG identification method based on local segment sparse representation according to claim 6, wherein the specific process of the step 3 is as follows:
3.1, searching the optimal matching atom in the matrix Y in the sparse representation process by adopting an orthogonal matching pursuit algorithm on the matrix Y after the dimension reduction;
and 3.2, constructing a dictionary of sparse representation by adopting a K singular value decomposition algorithm according to the optimal matching atoms to obtain a processed sparse coefficient matrix.
8. An ECG identification method based on local segment sparse representation according to claim 7, wherein the specific process of the step 3.2 is as follows:
step 3.2.1, describing the optimal matching atom obtained in the step 3.1 by using a linear expression of a basic element to form a basic atom dictionary, initializing the dictionary, and sparsely representing a given sample by using the dictionary to obtain a corresponding matrix;
step 3.2.2, when the matrix obtained in the step 3.2.1 is sparse, the error between the sample data and the data expressed by the coefficient matrix is continuously reduced by the iterative dictionary column by column, and the optimal base atom dictionary is infinitely approximated;
step 3.2.3, updating the basic atoms obtained in the step 3.2.2 row by row;
step 3.2.4, after the dictionary is updated in step 3.2.3, performing sparse coding on the new dictionary, and stopping updating the dictionary when the iteration number reaches a specified value or the error rate reaches a specified range, so as to obtain a processed sparse coefficient matrix, that is, after sparse representation, reconstructed electrocardiosignals Y 'are Dx, Y' are to-be-processed dimensionality-reduced data signal matrices, and D belongs to Rm×nIs a dictionary matrix, and x is a sparse coefficient x obtained after processing and belongs to Rm。
9. The ECG identification method based on the sparse representation of the local segments as claimed in claim 8, wherein the specific process of the step 4 is as follows:
step 4.1, inputting the final sparse coefficient matrix obtained in the step 3 into an input layer of the probabilistic neural network, and then entering a hidden layer;
step 4.2, inputting the data output from the hidden layer into the summation layer for calculation to obtain the maximum category calculated in the summation layer;
and 4.3, normalizing the maximum category data calculated in the summation layer to obtain probability estimation of each category, namely the identification precision of the ECG signal.
CN202111495097.4A 2021-12-08 2021-12-08 ECG identity recognition method based on local segment sparse representation Pending CN114419657A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111495097.4A CN114419657A (en) 2021-12-08 2021-12-08 ECG identity recognition method based on local segment sparse representation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111495097.4A CN114419657A (en) 2021-12-08 2021-12-08 ECG identity recognition method based on local segment sparse representation

Publications (1)

Publication Number Publication Date
CN114419657A true CN114419657A (en) 2022-04-29

Family

ID=81264820

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111495097.4A Pending CN114419657A (en) 2021-12-08 2021-12-08 ECG identity recognition method based on local segment sparse representation

Country Status (1)

Country Link
CN (1) CN114419657A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116509414A (en) * 2023-04-14 2023-08-01 中国科学院大学 Electrocardiosignal denoising classification system and method

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116509414A (en) * 2023-04-14 2023-08-01 中国科学院大学 Electrocardiosignal denoising classification system and method
CN116509414B (en) * 2023-04-14 2023-12-15 中国科学院大学 Electrocardiosignal denoising classification system and method

Similar Documents

Publication Publication Date Title
Li et al. Toward improving ECG biometric identification using cascaded convolutional neural networks
Hejazi et al. ECG biometric authentication based on non-fiducial approach using kernel methods
Irvine et al. eigenPulse: Robust human identification from cardiovascular function
Li et al. Robust ECG biometrics using GNMF and sparse representation
Boumbarov et al. ECG personal identification in subspaces using radial basis neural networks
Prakash et al. BAED: A secured biometric authentication system using ECG signal based on deep learning techniques
CN107122643B (en) Identity recognition method based on feature fusion of PPG signal and respiratory signal
Hammad et al. End-to-end data authentication deep learning model for securing IoT configurations
Huang et al. Learning joint and specific patterns: A unified sparse representation for off-the-person ECG biometric recognition
Hadiyoso et al. One-lead electrocardiogram for biometric authentication using time series analysis and Support Vector Machine
Huang et al. Robust multi-feature collective non-negative matrix factorization for ECG biometrics
Pinto et al. Deep neural networks for biometric identification based on non-intrusive ECG acquisitions
Al-Juboori et al. Palm vein verification using multiple features and locality preserving projections
Deshmane et al. ECG based biometric human identification using convolutional neural network in smart health applications
CN111325275B (en) Robust image classification method and device based on low-rank two-dimensional local identification map embedding
Prakash Capsule network for the identification of individuals using quantized ecg signal images
Ye et al. PPG-based biometric identification: Discovering and identifying a new user
Liu et al. LSTM Neural Network for Beat Classification in ECG Identity Recognition.
CN114419657A (en) ECG identity recognition method based on local segment sparse representation
Wu et al. A scalable open-set ECG identification system based on compressed CNNs
Huang et al. Multi-scale deep cascade bi-forest for electrocardiogram biometric recognition
CN112257573B (en) ECG identity recognition method based on t-SNE and Adaboost
Tseng et al. Ecg identification system using neural network with global and local features.
Amritha Varshini et al. Hybrid level fusion schemes for multimodal biometric authentication system based on matcher performance
CN112818315B (en) Electrocardiosignal identity recognition method and system fusing multi-feature information

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination