CN108470156B - Heart sound signal classification and identification method - Google Patents
Heart sound signal classification and identification method Download PDFInfo
- Publication number
- CN108470156B CN108470156B CN201810183034.7A CN201810183034A CN108470156B CN 108470156 B CN108470156 B CN 108470156B CN 201810183034 A CN201810183034 A CN 201810183034A CN 108470156 B CN108470156 B CN 108470156B
- Authority
- CN
- China
- Prior art keywords
- heart sound
- sound signal
- autocorrelation
- approximate
- detail
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- 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
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Signal Processing (AREA)
- Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)
- Investigating Or Analysing Biological Materials (AREA)
Abstract
The invention discloses a heart sound signal classification and identification method, which comprises the steps of firstly, carrying out discrete wavelet decomposition on a preprocessed heart sound signal to obtain detail wavelet coefficients and approximate wavelet coefficients of different frequency bands; then, sequentially solving normalized average aroma-rich energy envelope and autocorrelation function of the detail wavelet coefficient and the approximate wavelet coefficient to obtain autocorrelation characteristics of the detail wavelet coefficient envelope and autocorrelation characteristics of the approximate wavelet coefficient envelope, then respectively carrying out nonlinear characteristic dimension reduction on the detail autocorrelation characteristics and the approximate autocorrelation characteristics by utilizing a local linear embedding algorithm, and fusing the dimension-reduced detail characteristics and the approximate characteristics to obtain fusion characteristics; finally, the fusion features are used as the input of a support vector machine for classification and identification; the method can avoid the step processing of the heart sound signals, improve the accuracy of the extraction of the heart sound characteristics, and has positive effects on the analysis and the characteristic extraction of pathological heart sounds.
Description
Technical Field
The invention relates to a heart sound signal classification and identification method, in particular to a heart sound classification and identification method based on an autocorrelation function and a local linear embedding algorithm without segmentation.
Background
The generation of heart sounds mainly comes from the opening and closing of heart valves and the turbulent flow of blood, contains physiological information about various parts of the heart, such as atria, ventricles, great vessels, heart vessels and the functional states of various valves, can reflect the mechanical activity and mechanism of the heart, and has biological characteristics of universality, stability, uniqueness, collectability and the like. Auscultation of heart sounds is often used in modern medicine as a means for preliminary diagnosis of heart diseases, however, heart sounds are weak, low-frequency signals, and the auscultation results are easily affected by the hearing limitations of doctors and the subjectivity of doctors.
In the prior art, for the classification and identification of heart sound signals, feature extraction and classification and identification are mainly performed by segmenting heart sounds. The segmentation of the heart sound is to divide the heart sound signal into a series of single periodic signals according to the periodicity of the heart sound, then extract the features in each periodic signal, and perform classification and identification by taking the features as the input of a classifier. For the segmentation of heart sound, the electrocardiosignal is often adopted to assist the segmentation of the heart sound, but the electrocardiosignal and the heart sound signal are required to be recorded and synchronously processed simultaneously by assisting the segmentation with the electrocardiosignal, which is very inconvenient; there are also methods for segmenting heart sounds by using an envelope method, but there are two problems in segmenting by using the method: (1) for pathological heart sounds or background noises, omission of peak points or detection errors inevitably occur; (2) the use of peak points to determine the period of a heart sound signal is based on the premise that the systolic phase of the heart sound signal is shorter than the diastolic phase, however, this condition is not always true, especially for pathological heart sounds. It is therefore important to find a method that does not require segmentation of the heart sound signal.
Disclosure of Invention
The purpose of the invention is as follows: the invention provides the heart sound signal classification and identification method which does not need to carry out segmentation processing on the heart sound signals and can extract the heart sound signals more accurately.
The technical scheme is as follows: the invention relates to a heart sound signal classification and identification method, which comprises the following steps:
(1) preprocessing the heart sound signal x (i);
(2) performing discrete wavelet decomposition on the preprocessed heart sound signals to obtain normalized average aroma concentration energy envelope;
(3) obtaining an autocorrelation function of the normalized average fragrance concentration energy envelope;
(4) taking the first M values of the wavelet coefficient envelope to form the autocorrelation characteristic of the wavelet coefficient envelope, and fusing the autocorrelation characteristic into a characteristic vector;
(5) respectively carrying out nonlinear feature dimensionality reduction on the detail autocorrelation features and the approximate autocorrelation features by using a local linear embedding algorithm, and fusing the dimensionality reduced detail features and the approximate features to obtain fused features;
(6) and taking the fusion features as the input of a support vector machine for classification and identification.
The step (1) comprises the following steps:
(11) reducing the sampling frequency of the heart sound signals x (i) to 2000 HZ;
(12) performing noise reduction by using a Butterworth low-pass filter with zero phase and 0-900HZ frequency band;
(13) normalizing the denoised heart sound signal x (i):
the step (2) comprises the following steps:
(21) performing discrete wavelet decomposition on the preprocessed heart sound signals to obtain detail wavelet coefficients and approximate wavelet coefficients of different frequency bands;
(22) the normalized average fragrance intensity energy envelope is obtained by the following formula:
wherein N ishIn order to be the window length of the sliding window,and the approximate coefficient or detail coefficient corresponding to the nth window length.
The autocorrelation function in the step (3) is:
where p (n + M) is a time-shifted signal of p (n), and M is 0, 1.
And (4) M is larger than the length of a data point contained in one heart sound signal period.
The step (5) comprises the following steps:
(51) respectively calculating sample sets R by adopting a KNN algorithm methoda={ra(1),…,ra(N) } and Rb={rb(1),...,rb(N) k neighbor points for each sample point in the (N) };
(52) calculating a reconstruction weight matrix W of the sample point:
let the minimization objective function be:
the constraint conditions are as follows: if riIs not rjThen W is 0; let the sum of each row of the weight matrix be 1, i.e. sigmajWij=1;
(53) Obtaining low-dimensional embedded L through k adjacent points of each sample point and the reconstruction weight matrix of each sample pointi:
WhereinIf i is equal to j,otherwise isLet ΣkLi0, and for LiCentering treatment is carried out, the influence of unchanged translation is removed, and the constraint of uniform variance is increasedThen for sparse matrices M can be represented by the following formula, namely:
M=(I-W)T(I-W)
sample set Ra={ra(1),...,ra(N)}∈RN×MAnd Rb={rb(1),...,rb(N)}∈RN×MRespectively carrying out nonlinear dimensionality reduction through the LLE algorithm to obtain
R’a={r'a(1),...,r′a(N)}∈RN×D
R′a={r′b(1),...,r′b(N)}∈RN×D
The fused features, namely:
r′ad=[r′a,r′b]∈R2D
has the advantages that: compared with the prior art, the invention has the beneficial effects that: 1. the method can avoid the step of carrying out segmentation pretreatment on the heart sound signals, improve the accuracy of the extraction of the heart sound characteristics, and has positive effects on the analysis and the characteristic extraction of pathological heart sounds; 2. the heart sound signal characteristics are optimized, main characteristic components are highlighted, data processing amount is reduced, and data processing efficiency is improved.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a raw normalized heart sound signal;
FIG. 3 is an average flavor intensity energy envelope of a heart sound signal;
FIG. 4 is an autocorrelation function of the mean richness energy envelope of a heart sound signal;
FIG. 5 is a normalized pathological heart sound signal with systolic murmurs;
FIG. 6 is a fourth layer of approximation coefficients subjected to wavelet decomposition;
FIG. 7 is a second level of detail coefficients subjected to wavelet decomposition;
FIG. 8 is an average fragrance intensity energy envelope for the fourth layer approximation coefficients;
FIG. 9 is an average fragrance intensity energy envelope for the second layer detail coefficients;
FIG. 10 is an autocorrelation function of the average fragrance intensity energy envelope of the fourth layer approximation coefficients;
FIG. 11 is an autocorrelation function of the average fragrance intensity energy envelope of the second layer detail coefficients;
FIG. 12 is a schematic diagram of a local linear embedding algorithm.
The specific implementation mode is as follows:
the invention is described in further detail below with reference to the accompanying drawings:
as shown in FIG. 1, the present invention mainly comprises the following steps
1. Preprocessing the heart sound signal x (i)
In order to avoid the difference between the acquisition environment and the acquisition standard of the heart sound data, the sampling frequency of the heart sound signal x (i) is firstly reduced to 2000HZ, then the noise reduction processing is carried out by using a butterworth low-pass filter with zero phase and 0-900HZ frequency band, and then the normalization processing is carried out on the noise-reduced heart sound signal, then:
2. discrete wavelet decomposition is carried out on the preprocessed heart sound signals to obtain normalized average fragrance concentration energy envelope
Adopting heart sound wavelets to carry out 4-layer wavelet decomposition on the heart sound signals after normalization processing, selecting an approximation coefficient of a fourth layer and a detail coefficient of a second layer to respectively obtain normalized average fragrance concentration energy envelopes due to the morphological similarity of the heart sound signals, and adopting a sliding window with the window length of 20ms and the window length of 10ms to carry out processing, so that the average fragrance concentration energy envelopes can be obtained by calculation through the following formula:
wherein N ishIn order to be the window length of the sliding window,and the approximate coefficient or detail coefficient corresponding to the nth window length.
After the heart sound signal is subjected to wavelet decomposition, the frequency band range of the approximate wavelet coefficient of the fourth layer is 0-125HZ, and the heart sound signal is a narrow-band signal, so that the frequency band contains the main characteristic information of the heart sound signal; the frequency band range of the detail wavelet coefficient at the second layer is 500-1000HZ, and the main pathological information or noise information is contained. Respectively calculating normalized average fragrant energy envelopes of detail wavelet coefficients and approximate wavelet coefficients obtained by heart sound wavelet decomposition by formula (2) to obtain pa(n) and pd(n)。
3. Obtaining an autocorrelation function of a normalized average energy envelope of fragrance concentration
The heart sound signal is a quasi-periodic signal, generally composed of a series of periodic components, and in each period, contains similar structural features, such as S1 and S2, and also contains noise of pathological states. Similar quasi-periodicity also exists in the wavelet coefficients of different frequency bands after wavelet decomposition and the envelope after normalized average fragrance concentration energy processing, and the autocorrelation function of the normalized shannon energy envelope of the wavelet coefficients of different frequency bands can provide periodic translation invariance of a plurality of cardiac cycles. And thus can be characterized as a heart sound signal. Since the autocorrelation function is symmetric, only the portion of it greater than 0 needs to be calculated, i.e.:
wherein p (n + M) is a time shift signal of p (n), and M is 0,1, …, M.
The peak points in the autocorrelation function indicate that there are a large number of periodic structures in the average fragrance intensity energy envelope, and a peak point in the autocorrelation function occurs when the time-shifted signal p (n + m) and the original signal p (n) have similar structural features or components, where m corresponds to the period in the average fragrance intensity energy envelope. Thus, the autocorrelation function may characterize the stationary periodic structure of the heart sound signal. FIG. 2 is a raw normalized heart sound signal; FIG. 3 is an average flavor intensity energy envelope of a heart sound signal; fig. 4 is an autocorrelation function of the average fragrance intensity energy envelope. It is clear from fig. 4 that there are three peak points in the autocorrelation function, where the first peak point represents the average systolic period of the heart sound, since the period of the systolic period is usually shorter than the period of the diastolic period, when the second heart sound S2 in the time-shifted signal p (n + m) coincides with the first heart sound S1 in the original signal p (n), the distance m of the translation is equal to the average systolic period of the heart sound; the second peak point represents the mean diastolic period of the heart sound, when the first heart sound S1 in the time-shifted signal p (n + m) coincides with the second heart sound S2 in the original signal p (n), the distance m of the shift is equal to the mean diastolic period of the heart sound; the third peak point represents the average period of the heart sounds, and when the first heart sound S1 and the second heart sound S2 in the time-shifted signal p (n + m) coincide with the first heart sound S1 and the second heart sound S2 in the original signal p (n), respectively, the distance m of the shift is equal to the average period of the heart sounds. It should be noted that when the systolic phase and the diastolic phase are nearly equal, the first peak point and the second peak point in the autocorrelation function may be very close to each other, or even completely coincident, as is often seen in the heart sounds of adults or children with too fast heart rates.
As can be seen from the above diagram, the periodic characteristics of the heart sound also exist in the autocorrelation function, and the autocorrelation functions of the approximation coefficient envelopes and the detail coefficient envelopes obtained after the wavelet decomposition and the average fragrant energy envelope processing can highlight the periodic characteristics of different frequency bands, and particularly, the autocorrelation functions of the approximation coefficient envelopes and the autocorrelation functions of the detail coefficient envelopes can highlight the characteristics of S1 and S2 and pathological characteristics of the heart sound to different degrees. As shown in fig. 5 to 7, wherein fig. 5 represents a normalized pathological heart sound signal with systolic mur; performing 4-layer wavelet decomposition on the pathological heart sound signal of fig. 5 by using heart sound wavelets to obtain a fourth-layer approximation coefficient shown in fig. 6 and a second-layer detail coefficient shown in fig. 7, and then performing normalized average fragrance concentration energy envelope processing on fig. 6 and 7 respectively to obtain fig. 8 and 9, wherein it can be seen that the normalized average fragrance concentration energy envelope of the detail coefficient and the normalized average fragrance concentration energy envelope of the approximation coefficient both contain periodic components such as S1, S2, pathological components and the like; finally, the signals of fig. 8 and 9 are processed by autocorrelation function, and fig. 10 and 11 are obtained. Clearly, the different periodic components of fig. 8 and 9 are still present in fig. 10 and 11.
4. Autocorrelation characteristics of the envelope of detail wavelet coefficients and autocorrelation characteristics of the envelope of approximation wavelet coefficients
Since the autocorrelation function has the ability to capture the information of the periodic structure of the heart sound signal, the autocorrelation function of the wavelet coefficient envelope can be regarded as a feature of the wavelet coefficient envelope.The first M values of the autocorrelation characteristics of the wavelet coefficient envelope are stacked into a column vector, i.e., r (M) [ [ r (1) ], r (M) ]]T∈RMThen, the self-correlation characteristics of the envelope of the approximate coefficient and the self-correlation characteristics of the envelope of the detail coefficient obtained by sequentially carrying out normalized average aroma-rich energy envelope and self-correlation function on the approximate wavelet coefficient and the detail wavelet coefficient obtained by discrete wavelet decomposition are respectively marked as raAnd rd. It should be noted that M must be larger than the length of the data point included in one heart sound period, so as to ensure that the approximate coefficient envelope autocorrelation characteristic and the detail coefficient envelope autocorrelation characteristic can include enough structural information in the heart sound signal. To make raAnd rdBecome the input of the support vector machine, the two feature vectors r need to be combinedaAnd rdFused into a feature vector. A simple and intuitive method is to connect r directlyaAnd rdNamely:
will r isadSimply referred to as coefficient envelope autocorrelation characteristics.
5. Respectively carrying out nonlinear feature dimensionality reduction on the detail autocorrelation features and the approximate autocorrelation features by utilizing a local linear embedding algorithm, and fusing the detail features and the approximate features after dimensionality reduction to obtain fusion features
Manifold learning maps high-dimensional space data to low-dimensional data space through a certain strategy, and geometric relation and distance measure between data are kept unchanged, so that the problem of dimension disaster in data processing is well solved. The local linear embedding algorithm (LLE) algorithm in manifold learning can well highlight the local geometric structure in a sample space and has the advantages of low calculation complexity, high running speed and the like, so that the invention adopts the LLE algorithm to perform dimension reduction processing on the obtained approximate coefficient envelope autocorrelation characteristics and detail coefficient envelope autocorrelation characteristics. Since the autocorrelation characteristic of the coefficient envelope comprises r after approximate coefficient processingaProcessed r of feature and detail coefficientsdAnd features are subjected to feature dimension reduction respectively when feature dimension reduction is carried out, and the features subjected to dimension reduction are fused to obtain fused features subjected to dimension reduction. Setting an approximate wavelet coefficient obtained by discrete wavelet decomposition of a heart sound data set and a detail wavelet coefficient to be subjected to normalized average aroma-rich energy envelope and autocorrelation function processing in sequence to obtain an approximate coefficient envelope autocorrelation characteristic sample set as Ra={ra(1),…,ra(N)}∈RN×MThe detail coefficient envelope autocorrelation characteristic sample set is Rb={rb(1),…,rb(N)}∈RN×MWhere N represents the number of samples and M represents the dimension of each sample. The local geometric structure of the LLE algorithm is kept unchanged before and after the mapping of the heart sound data, and meanwhile, a low-dimensional data set is obtained: r'a={r′a(1),…,r′a(N)}∈RN×D,R’b={r′b(1),…,r′b(N)}∈RN×DWhere D represents the dimension to which reduction is required, the specific operating steps of the LLE algorithm are therefore as follows:
(1) respectively calculating sample sets R by adopting a KNN algorithm methoda={ra(1),...,ra(N) }, and Rb={rb(1),...,rb(N) K neighboring points of each sample point, such that each sample point can be represented linearly by the K neighboring points.
(2) And (3) calculating a reconstruction weight matrix W of the sample point according to the K adjacent points obtained in the step (1).
First, a minimization objective function is defined as:
ε(W)=∑i||ri-∑jWijrj||2 (5)
when calculating RaWhen the samples are collected, r represents raWhen calculating RbWhen r represents rb. To calculate W, two constraints are added to the minimization objective function: first, since each sample point is reconstructed from its k neighbors, if r isiIs not rjIs close toO, then W ═ 0; secondly, let the sum of each row of the weight matrix be 1, i.e. ΣjWij1. Under the two constraint conditions, a final reconstruction weight matrix W is obtained by solving a least square method problem.
(3) Obtaining low-dimensional embedded L according to k adjacent points of each sample point obtained in the step 1) and the reconstructed weight matrix of each sample point obtained in the step 2)i. Here, a loss function is defined, namely:
thus, LiCan be found by minimizing the loss function, i.e.:
whereinIf i is equal to j,otherwise isLet ΣkLi0, and for LiCentering treatment is carried out, the influence of unchanged translation is removed, and the constraint of uniform variance is increasedThen for sparse matrices M can be represented by the following formula, namely:
M=(I-W)T(I-W) (8)
at this time, the solving problem of the extremum problem is the eigenvector corresponding to the smallest D eigenvalues from 2 nd to D +1 of M, that is:
thus, the sample set Ra={ra(1),...,ra(N)}∈RN×MAnd Rb={rb(1),…,rb(N)}∈RN×MRespectively carrying out nonlinear dimensionality reduction through the LLE algorithm to obtain
R′a={r′a(1),...,r′a(N)}∈RN×D,R’a={r′b(1),...,r′b(N)}∈RN×D. To make r'aAnd r'dBecome the input of the support vector machine, the two feature vectors r need to be combined′aAnd r′dFusing into a feature vector to obtain fused features, namely:
r′ad=[r′a,r′b]∈R2D (10)
the specific flow of the LLE algorithm is shown in fig. 4.
It can be seen from the figure that, before the processing is not performed, the sample set is in the high-dimensional data space, and the 3-step processing shown in fig. 12 is performed, that is, a suitable neighborhood point is selected for each sample point in the sample set, then a reconstruction weight matrix is calculated according to the neighborhood points, finally the low-dimensional embedding is performed by using the sample points and the reconstruction weight matrix of the sample points, and finally the high-dimensional data space is reduced to the two-dimensional space, so that the dimension reduction of the data is realized.
6. And taking the fusion features as the input of a support vector machine for classification and identification.
Claims (5)
1. A heart sound signal classification and identification method is characterized by comprising the following steps:
(1) preprocessing the heart sound signal x (i);
(2) performing discrete wavelet decomposition on the preprocessed heart sound signals to obtain normalized average aroma concentration energy envelope;
(3) obtaining an autocorrelation function of the normalized average fragrance concentration energy envelope;
(4) taking the first M values of the wavelet coefficient envelope to form the autocorrelation characteristic of the wavelet coefficient envelope, and fusing the autocorrelation characteristic into a characteristic vector;
(5) respectively carrying out nonlinear feature dimensionality reduction on the detail autocorrelation features and the approximate autocorrelation features by using a local linear embedding algorithm, and fusing the dimensionality reduced detail features and the approximate features to obtain fused features;
(6) taking the fusion features as the input of a support vector machine for classification and identification;
the step (5) comprises the following steps:
(51) respectively calculating sample sets R by adopting a KNN algorithm methoda={ra(1),…,ra(N) } and Rb={rb(1),…,rb(N) k neighbor points for each sample point in the (N) };
(52) calculating a reconstruction weight matrix W of the sample point:
let the minimization objective function be:
the constraint conditions are as follows: if riIs not rjThen W is 0; let the sum of each row of the weight matrix be 1, i.e. sigmajWij=1;
(53) Obtaining low-dimensional embedded L through k adjacent points of each sample point and the reconstruction weight matrix of each sample pointi:
WhereinIf i is equal to j,otherwise isLet ΣkLi0, and for LiCentering treatment is carried out, the influence of unchanged translation is removed, and the constraint of uniform variance is increasedThen for the sparse matrix B, the following formula can be used:
B=(I-W)T(I-W)
sample set Ra={ra(1),…,ra(N)}∈RN×MAnd Rb={rb(1),…,rb(N)}∈RN×MAfter nonlinear dimensionality reduction is performed through the LLE algorithms of (51), (52) and (53), respectively, the following results are obtained:
R′a={r′a(1),…,r′a(N)}∈RN×D
R′a={r′b(1),…,r′b(N)}∈RN×D
post-fusion characteristic is r'ad=[r′a,r′b]∈R2D。
2. The heart sound signal classification and identification method according to claim 1, wherein the step (1) comprises the steps of:
(11) reducing the sampling frequency of the heart sound signals x (i) to 2000 HZ;
(12) performing noise reduction by using a Butterworth low-pass filter with zero phase and 0-900HZ frequency band;
(13) normalizing the denoised heart sound signal x (i):
3. the method for classifying and identifying a heart sound signal according to claim 1, wherein the step (2) comprises the steps of:
(21) performing discrete wavelet decomposition on the preprocessed heart sound signals to obtain detail wavelet coefficients and approximate wavelet coefficients of different frequency bands;
(22) the normalized average fragrance intensity energy envelope is obtained by the following formula:
5. The method according to claim 1, wherein M in step (4) is greater than the length of the data points included in one heart sound signal period.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810183034.7A CN108470156B (en) | 2018-03-06 | 2018-03-06 | Heart sound signal classification and identification method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810183034.7A CN108470156B (en) | 2018-03-06 | 2018-03-06 | Heart sound signal classification and identification method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108470156A CN108470156A (en) | 2018-08-31 |
CN108470156B true CN108470156B (en) | 2021-09-28 |
Family
ID=63264187
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810183034.7A Active CN108470156B (en) | 2018-03-06 | 2018-03-06 | Heart sound signal classification and identification method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108470156B (en) |
Families Citing this family (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111000564B (en) * | 2019-11-28 | 2023-05-02 | 北京沃东天骏信息技术有限公司 | Wearable device and method for tumble monitoring |
CN111210845B (en) * | 2019-12-20 | 2022-06-21 | 太原理工大学 | Pathological voice detection device based on improved autocorrelation characteristics |
CN111863035A (en) * | 2020-07-21 | 2020-10-30 | 上海宜硕网络科技有限公司 | Method, system and equipment for recognizing heart sound data |
CN112326017B (en) * | 2020-09-28 | 2022-01-04 | 南京航空航天大学 | Weak signal detection method based on improved semi-classical signal analysis |
CN112971839B (en) * | 2021-02-05 | 2023-03-24 | 云南大学 | Heart sound classification method based on feedforward convolution neural network |
CN114010220A (en) * | 2021-10-29 | 2022-02-08 | 平安科技(深圳)有限公司 | Heart sound signal processing method, computer device and storage medium |
CN114027846B (en) * | 2021-11-04 | 2022-08-26 | 湖南万脉医疗科技有限公司 | Intelligent electrocardiosignal processing method |
CN114886404B (en) * | 2022-07-13 | 2022-10-28 | 西南民族大学 | Electronic equipment, device and storage medium |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104473660A (en) * | 2014-12-29 | 2015-04-01 | 哈尔滨师范大学 | Abnormal heart sound recognition method based on sub-band energy envelope autocorrelation characteristics |
CN104887263A (en) * | 2015-05-21 | 2015-09-09 | 东南大学 | Identity recognition algorithm based on heart sound multi-dimension feature extraction and system thereof |
-
2018
- 2018-03-06 CN CN201810183034.7A patent/CN108470156B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104473660A (en) * | 2014-12-29 | 2015-04-01 | 哈尔滨师范大学 | Abnormal heart sound recognition method based on sub-band energy envelope autocorrelation characteristics |
CN104887263A (en) * | 2015-05-21 | 2015-09-09 | 东南大学 | Identity recognition algorithm based on heart sound multi-dimension feature extraction and system thereof |
Also Published As
Publication number | Publication date |
---|---|
CN108470156A (en) | 2018-08-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108470156B (en) | Heart sound signal classification and identification method | |
Li et al. | Feature extraction and classification of heart sound using 1D convolutional neural networks | |
Fahad et al. | Microscopic abnormality classification of cardiac murmurs using ANFIS and HMM | |
CN109998525B (en) | Arrhythmia automatic classification method based on discriminant deep belief network | |
Naseri et al. | Detection and boundary identification of phonocardiogram sounds using an expert frequency-energy based metric | |
Jafarian et al. | Automating detection and localization of myocardial infarction using shallow and end-to-end deep neural networks | |
Alaskar et al. | The implementation of pretrained AlexNet on PCG classification | |
CN104473660B (en) | A kind of abnormal recognition of heart sound method based on sub-belt energy envelope autocorrelation characteristic | |
Ismail et al. | PCG classification through spectrogram using transfer learning | |
Singh et al. | Short unsegmented PCG classification based on ensemble classifier | |
CN115281688A (en) | Cardiac hypertrophy multi-label detection system based on multi-mode deep learning | |
CN113796889A (en) | Auxiliary electronic stethoscope signal discrimination method based on deep learning | |
CN114492519B (en) | Lung ultrasound special sign B line identification and classification method based on ultrasound echo radio frequency signals | |
Maity et al. | Transfer learning based heart valve disease classification from Phonocardiogram signal | |
Zeng et al. | Automatic detection of heart valve disorders using Teager–Kaiser energy operator, rational-dilation wavelet transform and convolutional neural networks with PCG signals | |
CN112869717B (en) | Pulse feature recognition and classification system and method based on BL-CNN | |
CN113116300A (en) | Physiological signal classification method based on model fusion | |
CN113069124A (en) | Electrocardio monitoring method based on CNN-ET model | |
Talal et al. | Machine learning‐based classification of multiple heart disorders from PCG signals | |
Tariq et al. | Automatic multimodal heart disease classification using phonocardiogram signal | |
Cheng et al. | A high recognition rate of feature extraction algorithm without segmentation | |
Malleswari et al. | Cardiac severity classification using pre trained neural networks | |
CN113449636B (en) | Automatic aortic valve stenosis severity classification method based on artificial intelligence | |
WO2023077592A1 (en) | Intelligent electrocardiosignal processing method | |
Quanyu | Pulse signal analysis based on deep learning network |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |