CN111310578B - Method and device for generating heart beat data sample classification network - Google Patents

Method and device for generating heart beat data sample classification network Download PDF

Info

Publication number
CN111310578B
CN111310578B CN202010057334.8A CN202010057334A CN111310578B CN 111310578 B CN111310578 B CN 111310578B CN 202010057334 A CN202010057334 A CN 202010057334A CN 111310578 B CN111310578 B CN 111310578B
Authority
CN
China
Prior art keywords
heart beat
sample
data
network
sample data
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
Application number
CN202010057334.8A
Other languages
Chinese (zh)
Other versions
CN111310578A (en
Inventor
吴泽剑
曹君
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.)
Shanghai Lepu Yunzhi Technology Co ltd
Original Assignee
Shanghai Lepu Yunzhi Technology Co ltd
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 Shanghai Lepu Yunzhi Technology Co ltd filed Critical Shanghai Lepu Yunzhi Technology Co ltd
Priority to CN202010057334.8A priority Critical patent/CN111310578B/en
Publication of CN111310578A publication Critical patent/CN111310578A/en
Application granted granted Critical
Publication of CN111310578B publication Critical patent/CN111310578B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06F18/24—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
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Abstract

The embodiment of the invention relates to a method and a device for generating a heart beat data sample classification network, wherein the method comprises the following steps: performing large sample screening treatment on the first batch of sample data sequences to generate large sample data sequences, and performing small sample screening treatment on the first batch of sample data sequences to generate small sample data sequences; performing convolutional network training processing on the feature extraction network and the classification network based on the large sample data sequence to generate a large sample feature extraction network and a large sample classification network; performing convolutional network training processing on the large sample feature extraction network based on the small sample data sequence to generate a small sample classification network; and merging the small sample classification network with the large sample classification network to generate a heart beat data sample classification network. The embodiment of the invention also relates to a method and a device for identifying the heart beat data based on the heart beat data sample classification network, which are used for identifying the heart beat category of the real-time heart beat data.

Description

Method and device for generating heart beat data sample classification network
Technical Field
The invention relates to the technical field of electrocardiosignal processing, in particular to a method and a device for generating a heart beat data sample classification network.
Background
The electrocardiograph data is a group of electrical signal data which is collected by the electrocardiograph through the body surface electrode and is related to the cardiac cycle of the heart, and the electrocardiograph analysis is to perform characteristic analysis on the collected electrocardiograph data. The method for carrying out intelligent analysis on the electrocardiographic data by deep learning is to input batch characteristic electrocardiographic data as sample data into a convolutional neural network for network learning training to generate a characteristic classification model, and then to detect and classify the acquired real-time electrocardiographic data by using the classification model. Thus, in the deep learning process, a sufficient amount of data is critical to the training of the network learning. However, in the medical field, insufficient data is very common. In the electrocardiographic data, common heart beat types such as 'normal', 'atrial premature beat', 'ventricular premature beat', and the like can obtain tens of thousands or even millions of heart beat fragments, belong to large sample data, and can meet the requirement of training a deep learning network; unusual heart beat types, such as "ventricular flutter", "ventricular fibrillation", etc., can only collect a few hundred heart beat fragments, belonging to small sample data, far from deep learning.
Disclosure of Invention
The invention aims at overcoming the defects of the prior art, and provides a method and a device for generating a heart beat data sample classification network, which are characterized in that a large sample is utilized to establish a feature extraction and classification network model, small sample data is used as optimization data of the network model, and small sample optimization learning training is continuously carried out on a learning network to finally generate a full sample classification network model. The invention also provides a method and a device for identifying the heart beat data based on the heart beat data sample classification network, namely, the heart beat data sample classification network generated by the method is used for carrying out classification identification on the real-time electrocardio data.
To achieve the above object, a first aspect of an embodiment of the present invention provides a method for generating a classification network for cardiac data samples, which is characterized in that the method includes:
the method comprises the steps that an electrocardio analysis module acquires electrocardio data with a specified batch number from an electrocardio database of an upper computer to generate first batch electrocardio data, and heart beat sample data extraction processing is carried out on the first batch electrocardio data to generate a first batch sample data sequence;
the electrocardiographic analysis module performs large sample screening processing on the first batch of sample data sequences according to the first batch of sample data sequences to generate large sample data sequences, and performs small sample screening processing on the first batch of sample data sequences to generate small sample data sequences;
the electrocardio analysis module carries out convolutional network training treatment on the characteristic extraction network and the classification network based on the large sample data sequence to generate a large sample characteristic extraction network and a large sample classification network;
the electrocardio analysis module carries out convolutional network training treatment on the large sample characteristic extraction network based on the small sample data sequence to generate a small sample classification network;
and the electrocardio analysis module performs classification network merging processing on the small sample classification network and the large sample classification network to generate a heart beat data sample classification network.
Preferably, the electrocardiograph analysis module acquires electrocardiograph data of a designated batch number from an electrocardiograph database of the upper computer to generate first batch electrocardiograph data, and performs heartbeat sample data extraction processing on the first batch electrocardiograph data to generate a first batch sample data sequence, which specifically includes:
the electrocardio analysis module acquires the electrocardio data of the appointed batch number from the electrocardio database of the upper computer to generate the first batch electrocardio data;
the electrocardio analysis module acquires a preset heart beat sample extraction time length and generates a first time length; initializing the first batch of sample data sequences to be empty;
the electrocardio analysis module polls all the electrocardio data in the first batch of electrocardio data, carries out heart beat R point position confirmation processing on the currently polled electrocardio data to generate a current R point position sequence, and extracts the electrocardio data with the first time length forwards and backwards respectively by taking each heart beat R point position in the current R point position sequence as a center to be combined into heart beat sample data;
and the electrocardiographic analysis module performs sample data object adding processing on all the extracted heart beat sample data to the first batch of sample data sequences.
Preferably, the method further comprises:
the large sample data sequence is a group of sample data classification sequences { l } 1 ，l 2 ，...l i ...，l m -a }; the m is the total classification number of the large sample data sequence; the l is i For a group of homogeneous cardiac sample data sequences { h } in said large sample data sequence 1 ，h 2 ，...h s ，...h n -a }; the value range of i is from 1 to m; the value range of s isFrom 1 to n; n is the total number of heart beat sample data of each group of heart beat sample data sequences;
the small sample data sequence is a group of similar heart beat sample data sequences { x } 1 ，x 2 ，...x s ，...x n ，y 1 ，y 2 ，...y s ，...y n }。
Preferably, the electrocardiographic analysis module performs convolutional network training processing on the feature extraction network and the classification network based on the large sample data sequence, and generates a large sample feature extraction network and a large sample classification network, which specifically includes:
the electrocardiographic analysis module is based on the large sample data sequence { l } 1 ，l 2 ，...l i ...，l m Performing convolutional network training processing on the characteristic extraction network to generate the large sample characteristic extraction network; the large sample characteristic extraction network is a convolution layer network F; the large sample feature extraction network F is used for extracting the large sample data sequence { l } 1 ，l 2 ，...l i ...，l m Said l in } i All heart beat sample data { h } 1 ，h 2 ，...h s ，...h n Performing convolution network calculation to generate corresponding heart beat characteristic value sequence { f } 1 ，f 2 ，...f s ，...f n -the heart beat feature value f s ＝F(h s )；
The electrocardiographic analysis module is based on the large sample data sequence { l } 1 ，l 2 ，...l i ...，l m Performing convolutional network training processing on the classification network to generate the large sample classification network; the large sample classification network is a matrix network W m×n ＝(w 1 ，w 2 ，...w i ，...w m ) T The method comprises the steps of carrying out a first treatment on the surface of the The w is i Is a matrix column vector of length n; the T is the matrix transpose operator.
Preferably, the electrocardiographic analysis module performs convolutional network training processing on the large sample feature extraction network based on the small sample data sequence to generate a small sample classification network, and specifically includes:
the electrocardiographic analysis module is used for carrying out the analysis on the small sample data sequence { x } 1 ，x 2 ，...x s ，...x n ，y 1 ，y 2 ，...y s ，...y n Dividing into two parts, and generating a first small sample data sequence and a second small sample data sequence; the first small sample data sequence { x } 1 ，x 2 ，...x s ，...x n Comprises n heart beat sample data, the second small sample data sequence { y } 1 ，y 2 ，...y s ，...y n -comprising n heart beat sample data;
the electrocardiographic analysis module analyzes the first small sample data sequence { x } 1 ，x 2 ，...x s ，...x n Setting to the m+1st class of the large sample data sequence; based on the first small sample data sequence { x } 1 ，x 2 ，...x s ，...x n Performing convolutional network training processing on the large sample feature extraction network F pair to generate a small sample feature generation network
Figure BDA0002371821170000045
Said small sample feature generation network->
Figure BDA0002371821170000046
Is a vector with length of n;
the electrocardiographic analysis module sends the second small sample data sequence { y }, to the computer system 1 ，y 2 ，...y s ，...y n Setting to the m+1st class of the large sample data sequence; using the large sample feature extraction network F for the second small sample data sequence { y } 1 ，y 2 ，...y s ，...y n Feature extraction calculation is carried out to generate a small sample heart beat feature sequence { f' 1 ，f′ 2 ，...f′ s ，...f′ n -a }; the small sample cardiac feature sequence { f' 1 ，f′ 2 ，...f′ s ，...f′ n -comprising n heart beat feature values;
the electrocardiograph is dividedThe analysis module carries out the heart beat characteristic sequence { f 'of the small sample' 1 ，f′ 2 ，...f′ s ，...f′ n Performing eigenvalue average calculation to generate small sample eigenvalue f' Average of The said
Figure BDA0002371821170000041
The electrocardiographic analysis module generates a network according to the small sample characteristics
Figure BDA0002371821170000042
And the small sample feature average f' Average of Generating the small sample classification network w m+1 The method comprises the steps of carrying out a first treatment on the surface of the The w is m+1 Is a vector with length of n, the
Figure BDA0002371821170000043
Said operator->
Figure BDA0002371821170000044
Is the kronecker product.
Preferably, the electrocardiographic analysis module performs classification network merging processing on the small sample classification network and the large sample classification network, and generates a heartbeat data sample classification network, which specifically includes:
the electrocardiographic analysis module classifies the small sample into a network w m+1 And the large sample classification network W m×n Performing classification network merging processing to generate the heart beat data sample classification network W (m+1)×n The method comprises the steps of carrying out a first treatment on the surface of the The W is (m+1)×n ＝(w 1 ，w 2 ，...，w J ...w m ，w m+1 ) T The method comprises the steps of carrying out a first treatment on the surface of the The value of j ranges from 1 to m+1.
Preferably, the method further comprises:
after the heart beat data sample classification network is established, when the electrocardio analysis module acquires a new training small sample data sequence { z }, the data sequence of the training small sample is obtained 1 ，z 2 ，...，z n At the time of } the electrocardiographic analysis module transmits the training small sample data sequence { z }, the electrocardiographic analysis module generates a training small sample data sequence { z }, a training small sample data sequence 1 ，z 2 ，...，z n Setting to the m+1st class of the large sample data sequence;
the electrocardiographic analysis module uses the large sample feature extraction network F to train the small sample data sequence { z } 1 ，z 2 ，...，z n Feature extraction calculation is carried out to generate a training small sample heart beat feature sequence { f } ", and the training small sample heart beat feature sequence { f }", the training small sample heart beat feature sequence { f } 1 ，f″ 2 ，...，f″ i ，...f″ n -a }; the training small sample data sequence { z } 1 ，z 2 ，...，z n -comprising n heart beat sample data; the training small sample heart beat characteristic sequence { f } ", is 1 ，f″ 2 ，...，f″ i ，...f″ n -including n of said heart beat feature values;
the electrocardiographic analysis module performs the training on the heart beat characteristic sequence { f } ", of the small sample 1 ，f″ 2 ，...，f″ i ，...f″ n Performing feature value average calculation processing to generate training small sample feature average value f' and average value, wherein
Figure BDA0002371821170000051
Figure BDA0002371821170000052
The electrocardiographic analysis module generates a network according to the small sample characteristics
Figure BDA0002371821170000053
And the training small sample characteristic average value f Average of Generating an optimized small sample classification network w' m+1 The method comprises the steps of carrying out a first treatment on the surface of the The w' m+1 Is a vector with length of n, the
Figure BDA0002371821170000054
The electrocardiographic analysis module classifies the optimized small sample into a network w' m+1 And the large sample classification network W m×n Performing classification network merging processing to generate an optimized heart beat data sample classification network W' (m+1)×n The method comprises the steps of carrying out a first treatment on the surface of the The W' (m+1)×n ＝(w 1 ，w 2 ，...w j ，...w m ，w′ m+1 ) T 。
According to the method for generating the heart beat data sample classification network, which is provided by the first aspect of the embodiment of the invention, the characteristic extraction or classification network is established by utilizing the large sample, and then the small sample data is used as the optimization data of the network to carry out the small sample optimization training on the learning network, so that the small sample data with small quantity can also obtain good classification network precision through the deep learning of the convolutional neural network.
A second aspect of the embodiments of the present invention provides an apparatus, the apparatus comprising a memory for storing a program and a processor for performing the method of the first aspect and the respective implementation manners of the first aspect.
A third aspect of the embodiments of the present invention provides a computer program product comprising instructions which, when run on a computer, cause the computer to perform the method of the first aspect and implementations of the first aspect.
A fourth aspect of the embodiments of the present invention provides a computer readable storage medium having a computer program stored thereon, which when executed by a processor, implements the method of the first aspect and implementations of the first aspect.
A fifth aspect of the embodiment of the present invention provides a method for identifying cardiac data based on a cardiac data sample classification network, which is characterized in that the method includes:
the method comprises the steps that an electrocardio analysis module acquires electrocardio data with a specified batch number from an electrocardio database of an upper computer to generate first batch electrocardio data, and heart beat sample data extraction processing is carried out on the first batch electrocardio data to generate a first batch sample data sequence;
the electrocardiographic analysis module performs large sample screening processing on the first batch of sample data sequences according to the first batch of sample data sequences to generate large sample data sequences, and performs small sample screening processing on the first batch of sample data sequences to generate small sample data sequences;
the electrocardio analysis module carries out convolutional network training treatment on the characteristic extraction network and the classification network based on the large sample data sequence to generate a large sample characteristic extraction network and a large sample classification network;
The electrocardio analysis module carries out convolutional network training treatment on the large sample characteristic extraction network based on the small sample data sequence to generate a small sample classification network;
the electrocardiographic analysis module performs classification network merging processing on the small sample classification network and the large sample classification network to generate a heart beat data sample classification network W A×B The method comprises the steps of carrying out a first treatment on the surface of the The heart beat data sample classification network W A×B ＝(w 1 ，w 2 ，...w v ，...w A ) T The method comprises the steps of carrying out a first treatment on the surface of the The w is v Is a matrix column vector of length B; the value range of v is from 1 to A; the T is the matrix transpose operator. The heart beat data sample classification network W A×B ＝(w 1 ，w 2 ，...w v ，...w 4 ) T The method comprises the steps of carrying out a first treatment on the surface of the The w is v Is a matrix column vector of length B; the value range of v is from 1 to A; the T is a matrix transpose operator;
the electrocardio analysis module acquires acquired real-time heart beat data C from the electrocardio acquisition module;
the electrocardiographic analysis module uses the heart beat data sample classification network W A×B Classifying and detecting the real-time heart beat data C to generate a real-time heart beat prediction category and a real-time heart beat prediction category probability;
the electrocardio analysis module acquires a preset heart beat probability threshold value of the category according to the real-time heart beat prediction category, when the real-time heart beat prediction category probability is larger than or equal to the heart beat probability threshold value, the real-time heart beat data C are identified to belong to the real-time heart beat prediction category, and when the real-time heart beat prediction category probability is smaller than the heart beat probability threshold value, the real-time heart beat data C are identified to belong to an unknown category.
Preferably, the electrocardiographic analysis module uses the heart beat data sample classification networkW A×B The real-time heartbeat data C is subjected to classification detection to generate a real-time heartbeat prediction category and a real-time heartbeat prediction category probability, and the method specifically comprises the following steps:
the electrocardio analysis module calculates a heart beat characteristic value of the real-time heart beat data C by using the large sample characteristic extraction network to generate a real-time heart beat characteristic value fC;
the electrocardiographic analysis module uses the heart beat data sample classification network W A×B And the real-time heart beat characteristic value fC performs classification and identification processing on the real-time heart beat data C to generate a heart beat data C classification and identification vector { U } 1 ，U 2 ...U V ，...U A -a }; the U is V ＝φ*cos＜w V fC > where phi is a characteristic calculation constant and cos < w V fB > is the w V Cosine value of the included angle with the real-time heart beat feature value fC;
the electrocardiographic analysis module classifies and identifies a vector { U } according to the heartbeat data C 1 ，U 2 ...U V ，...U A Performing heart beat classification probability calculation on the real-time heart beat data C to generate a heart beat data C classification probability sequence { P' 1 ，P′ 2 ...P′ v ，...P′ A -said
Figure BDA0002371821170000071
The electrocardiographic analysis module classifies probability sequences { P 'of the heart beat data C' 1 ，P′ 2 ...P′ v ，...P′ A Maximum statistics to generate heart beat data C maximum probability P' max Acquiring the maximum probability P 'of the heart beat data C' max Generating the real-time heart beat prediction category according to the corresponding heart beat type, and setting the probability of the real-time heart beat prediction category as the maximum probability P 'of the heart beat data C' max 。
According to the heart beat data identification method based on the heart beat data sample classification network, the heart beat data sample classification network is established through combination of large samples and small samples, the heart beat data sample classification network is used for classifying and identifying real-time heart beat data to output real-time heart beat prediction types and real-time heart beat prediction type probabilities, a preset heart beat probability threshold value is called according to the real-time heart beat prediction types and the real-time heart beat prediction type probabilities are compared, and real-time heart beat data meeting the threshold value requirements is identified according to the real-time heart beat prediction types.
A sixth aspect of the embodiments of the present invention provides an apparatus, the apparatus comprising a memory for storing a program and a processor for performing the method of the fifth aspect and each implementation of the fifth aspect.
A seventh aspect of the embodiments of the present invention provides a computer program product comprising instructions which, when run on a computer, cause the computer to perform the method of the fifth aspect and implementations of the fifth aspect.
An eighth aspect of the embodiments of the present invention provides a computer readable storage medium, on which a computer program is stored, which when executed by a processor implements the method in the fifth aspect and each implementation manner of the fifth aspect.
Drawings
Fig. 1 is a schematic diagram of a method for generating a classification network for cardiac data samples according to a first embodiment of the present invention;
fig. 2 is a schematic diagram of a heart beat signal according to an embodiment of the present invention;
fig. 3 is a schematic diagram of an optimization method for a classification network of cardiac data samples according to a second embodiment of the present invention;
fig. 4 is a schematic diagram of a method for identifying cardiac data based on a cardiac data sample classification network according to a third embodiment of the present invention;
fig. 5A is a schematic structural diagram of a generating device of a classification network for cardiac data samples according to a fourth embodiment of the present invention;
fig. 5B is a schematic structural diagram of a cardiac data identification device based on a cardiac data sample classification network according to a fifth embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail below with reference to the accompanying drawings, and it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The artificial intelligence convolutional neural network (Convolutional Neural Network, CNN) model is a deep learning model, which is essentially a multi-level network connection structure that simulates a neural network. Comprising the following steps: a convolutional Layer (Convolution Layer), an activation function (Relu), a Pooling Layer (Pooling Layer), and a fully-connected Layer (Full Connection Layer). The learning model is used, namely, input data is sequentially subjected to convolution layer calculation, activation function calculation and pooling layer calculation to output characteristic values aiming at the input data to be identified, then the characteristic values are subjected to data reconstruction through full-connection layer calculation, and logistic regression is performed by using a Softmax function, so that a classification matrix network aiming at the input data is finally formed. The training of the learning model is to use a large amount of sample data to perform sample training and full-connection network factor superposition on each calculation layer, so that the accuracy of the feature value and the accuracy of the classification matrix obtained by the trained network after calculation can be improved and perfected. The generation method of the heart beat data sample classification network is realized based on a CNN model. In the method of the present invention, the part of the neural network related to the continuous process of convolution layer calculation, activation function calculation and pooling layer calculation is called a feature extraction network, and the matrix network related to full connection layer calculation is called a classification network.
As shown in fig. 1, which is a schematic diagram of a method for generating a classification network for cardiac data samples according to a first embodiment of the present invention, the method mainly includes the following steps:
step 1, an electrocardiograph analysis module acquires electrocardiograph data of a designated batch number from an electrocardiograph database of an upper computer to generate first batch electrocardiograph data, and performs heartbeat sample data extraction processing on the first batch electrocardiograph data to generate a first batch sample data sequence;
the method specifically comprises the following steps: step 11, an electrocardiograph analysis module acquires electrocardiograph data of a designated batch number from an electrocardiograph database of an upper computer to generate first batch electrocardiograph data;
step 12, an electrocardiograph analysis module acquires a preset heart beat sample extraction time length and generates a first time length; initializing a first batch of sample data sequences to be empty;
step 13, the electrocardiograph analysis module polls all electrocardiograph data in the first batch of electrocardiograph data, carries out heart beat R point position confirmation processing on the electrocardiograph data which are polled currently to generate a current R point position sequence, and extracts electrocardiograph data with a first time length forwards and backwards respectively by taking each heart beat R point position in the current R point position sequence as a center to be combined into heart beat sample data;
Here, as shown in fig. 2, which is a schematic diagram of a cardiac signal provided by an embodiment of the present invention, a period of continuous time of cardiac signal data is composed of a plurality of cardiac signal data, each cardiac signal data includes 5 feature points P, Q, R, S, T, and it is also seen that, in these 5 points, the peak value of R point is highest, and the anti-interference capability of R point is strongest compared with P point and T point. Therefore, compared with the traditional 5 snack pulse signal identification method, the identification accuracy of the effective signals is improved when the heart pulse signals are identified through the R point; one heart beat sample data 2 is advanced by delta t from the point of R in the figure 1 Segment electrocardiographic data and take Δt backward 2 Combining the electrocardiographic data, wherein Deltat 1 ＝Δt 2 =first time length;
in step 14, the electrocardiographic analysis module performs sample data object adding processing on all the extracted cardiac sample data to the first batch of sample data sequences.
And 2, performing large sample screening processing on the first batch of sample data sequences by the electrocardiograph analysis module according to the first batch of sample data sequences to generate large sample data sequences, and performing small sample screening processing on the first batch of sample data sequences to generate small sample data sequences.
Here, because the large sample data sequence is used for the electrocardiographic analysis mode The learning network of the block is trained so that the sample data is sample data with a definite classification. Wherein the large sample data sequence is a group of sample data classification sequences { l } 1 ，l 2 ，...l i ...，l m -a }; m is the total number of classifications of the large sample data sequence; l (L) i For a group of homogeneous heart beat sample data sequences { h } in a large sample data sequence 1 ，h 2 ，...h s ，...h n -a }; i has a value ranging from 1 to m; s has a value ranging from 1 to n; n is the total number of heart beat sample data of each group of heart beat sample data sequences; the small sample data sequence is a group of similar heart beat sample data sequences { x } 1 ，x 2 ，...x s ，...x n ，y 1 ，y 2 ，...y s ，...y n }. For example, assume that the first large sample data includes three types of large sample heart beat data: normal heart beat, atrial premature beat and ventricular premature beat, the number of large sample heart beat data in each type is 1000, and then the large sample data sequence is { l } 1 ，l 2 ，l 3 "wherein l 1 Is the normal heart beat data sequence { h } 11 ，h1 2 ，...，h1 1000 }，l 2 Is an atrial premature beat data sequence { h2 1 ，h2 2 ，...，h2 1000 }，l 3 Is a ventricular premature beat data sequence { h3 } 1 ，h3 2 ，...，h3 1000 Here m=3, n=1000.
Here, the small sample data sequence is a group of similar heart beat sample data sequences { x } 1 ，x 2 ，...x s ，...x n ，y 1 ，y 2 ，...y s ，...y n }. . In the subsequent analysis, the method of the invention needs to use the same kind of small sample data to complete the network training and the classification network training, and if the same batch of data is adopted for the two training, the model can not correct errors, so that two groups of different data sequences with the length of n need to be extracted. For example, assuming that a small sample data sequence is used to learn ventricular fibrillation type electrocardiographic data, the small sample data sequence includes 2000 ventricular fibrillation heart beat sample data.
And step 3, the electrocardio analysis module carries out convolutional network training processing on the characteristic extraction network and the classification network based on the large sample data sequence to generate a large sample characteristic extraction network and a large sample classification network.
The method specifically comprises the following steps: step 31, the electrocardiographic analysis module is based on the large sample data sequence { l } 1 ，l 2 ，...l i ...，l m Performing convolutional network training processing on the feature extraction network to generate a large sample feature extraction network;
the large sample characteristic extraction network is a convolution layer network F; the large sample feature extraction network F is used for the large sample data sequence { l } 1 ，l 2 ，...l i ...，l m L in } i All heart beat sample data { h } 1 ，h 2 ，...h s ，...h n Performing convolution network calculation to generate corresponding heart beat characteristic value sequence { f } 1 ，f 2 ，...f s ，...f n Heart beat feature value f s ＝F(h s )；
Step 32, the electrocardiographic analysis module is based on the large sample data sequence { l } 1 ，l 2 ，...l i ...，l m Performing convolutional network training treatment on the classification network to generate a large sample classification network;
wherein the large sample classification network is a matrix network W m×n ＝(w 1 ，w 2 ，...w i ，...w m ) T ；w i Is a matrix column vector of length n; t is the matrix transpose operator.
Here, step 3 is training a heart beat classification model based on the convolutional neural network principle using a large sample data sequence; here, based on the CNN calculation principle, the heart beat classification model can be split into a feature extraction network and a classification network: the feature extraction network is a continuous processing process from convolution layer calculation, activation function calculation to pooling layer calculation in CNN calculation, and the classification network is a matrix network processing process related to full-connection layer calculation.
Here, it is assumed that the first large sample data includes three types of large sample heart beat data: positive directionThe number of the large sample heart beat data of each class is 1000, and the large sample heart beat data is specific to l 1 { h1 of (2) 1 ，h1 2 ，...，h1 1000 Calculated as { f1 } 1 ，f1 2 ，...，f1 1000 -a }; for l 2 { h2 of (2) 1 ，h2 2 ，...，h2 1000 Calculated as { f2 } 1 ，f2 2 ，...，f2 1000 -a }; for l 3 { h3 of (2) 1 ，h3 2 ，...，h3 1000 Calculated as { f3 } 1 ，f3 2 ，...，f3 1000 }. The large sample classification network is a matrix network W m×n Then is W 3×1000 . It is assumed here that the dimensions of the single matrix match the input sample data.
Here, after the large sample feature extraction network and the large sample classification network are trained, the unknown electrocardiographic data can be classified into large samples by using the calculation and classification functions of the large sample feature extraction network and the large sample classification network.
Step 4, the electrocardio analysis module carries out convolutional network training treatment on the large sample characteristic extraction network based on the small sample data sequence to generate a small sample classification network;
the method specifically comprises the following steps: step 41, the electrocardiographic analysis module sequences { x } the small sample data 1 ，x 2 ，...x s ，...x n ，y 1 ，y 2 ，...y s ，...y n Dividing into two parts, and generating a first small sample data sequence and a second small sample data sequence;
wherein the first small sample data sequence { x 1 ，x 2 ，...x s ，...x n The second small sample data sequence { y } comprises n pieces of heart beat sample data 1 ，y 2 ，...y s ，...y n -comprising n heart beat sample data;
Here, assuming that the small sample data sequence is for learning ventricular fibrillation-type electrocardiographic data, the small sample data sequence includes 2000 ventricular fibrillation heart beat sample data, a first small sample data sequence { x } 1 ，x 2 ，...，x 1000 Comprises 1000 chambersData of heart beat sample, second small sample data sequence { y } 1 ，y 2 ，...，y 1000 -1000 chamber flutter sample data;
step 42, the electrocardiographic analysis module compares the first small sample data sequence { x } 1 ，x 2 ，...x s ，...x n Setting to the m+1st class of the large sample data sequence; based on the first small sample data sequence { x } 1 ，x 2 ，...x s ，...x n Performing convolutional network training processing on the large sample feature extraction network F pair to generate a small sample feature generation network
Figure BDA0002371821170000123
Wherein the small sample feature generation network
Figure BDA0002371821170000124
Is a vector with length of n;
step 43, the electrocardiographic analysis module compares the second small sample data sequence { y } 1 ，y 2 ，...y s ，...y n Setting to the m+1st class of the large sample data sequence; second small sample data sequence { y } using large sample feature extraction network F 1 ，y 2 ，...y s ，...y n Feature extraction calculation is carried out to generate a small sample heart beat feature sequence { f' 1 ，f′ 2 ，...f′ s ，...f′ n }；
Wherein, the small sample heart beat feature sequence { f' 1 ，f′ 2 ，...f′ s ，...f′ n -comprising n heart beat feature values;
step 44, the electrocardiographic analysis module performs a small sample cardiac feature sequence { f' 1 ，f′ 2 ，...f′ s ，...f′ n Performing eigenvalue average calculation to generate small sample eigenvalue f' Average of ；
Wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure BDA0002371821170000121
step 45, generating a network by the electrocardiographic analysis module according to the characteristics of the small sample
Figure BDA0002371821170000122
And a small sample feature average f' Average of Generating a small sample classification network w m+1 ；
Wherein, small sample classification network w m+1 Is a vector of length n,
Figure BDA0002371821170000131
operator
Figure BDA0002371821170000132
Is the kronecker product.
Step 5, the electrocardio analysis module carries out classification network merging processing on the small sample classification network and the large sample classification network to generate a heart beat data sample classification network;
the method specifically comprises the following steps: electrocardiogram analysis module classifies small samples into a network w m+1 And large sample classification network W m×n Performing classification network merging processing to generate a heart beat data sample classification network W (m+1)×n ；
Wherein W is (m+1)×n ＝(w 1 ，w 2 ，...，w J ...w m ，w m+1 ) T The method comprises the steps of carrying out a first treatment on the surface of the j ranges from 1 to m+1.
Here, generally, tens of thousands of data of each class of large sample data can satisfy the requirement of deep learning; the small sample is generally hundreds of thousands of data of each classification, if the data is trained alone, the convolutional neural network is difficult to achieve the purpose of deep learning, so that the accuracy of the classification network is affected, and if the data is trained synchronously with the large sample, the data is easy to filter out as noise by mean value. In order to solve the problems, the method of the invention firstly establishes a feature extraction or classification network by using a large sample, and then uses small sample data as optimization data of the network to perform small sample optimization training on a learning network, so that small sample data with small quantity can also obtain good classification network precision through deep learning of a convolutional neural network.
As shown in fig. 3, which is a schematic diagram of an optimization method for a classification network of cardiac data samples according to a second embodiment of the present invention, the method mainly includes the following steps:
step 101, cardiac data sample classification network W (m+1)×n After the establishment, when the electrocardiographic analysis module acquires a new training small sample data sequence { z } 1 ，z 2 ，...，z n At the time of } the electrocardiographic analysis module trains the small sample data sequence { z }, the electrocardiographic analysis module is used for analyzing the small sample data sequence { z }, the electrocardiographic analysis module is 1 ，z 2 ，...，z n Setting to the m+1st class of the large sample data sequence.
Step 102, the electrocardiographic analysis module trains the small sample data sequence { z } using the large sample feature extraction network F 1 ，z 2 ，...，z n Feature extraction calculation is carried out to generate a training small sample heart beat feature sequence { f } ", and the training small sample heart beat feature sequence { f }", the training small sample heart beat feature sequence { f } 1 ，f″ 2 ，...，f″ i ，...f″ n }；
Wherein the training small sample data sequence { z 1 ，z 2 ，...，z n -comprising n heart beat sample data; training small sample cardiac feature sequence { f } " 1 ，f″ 2 ，...，f″ i ,. f "n" includes n heart beat feature values.
Step 103, the electrocardiographic analysis module trains the heart beat feature sequence { f } ", of the small sample 1 ，f″ 2 ，...，f″ i ，...f″ n Performing characteristic value average calculation to generate training small sample characteristic average value f Average of ；
Wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure BDA0002371821170000144
step 104, the electrocardiographic analysis module generates a network according to the characteristics of the small sample
Figure BDA0002371821170000145
And training a small sample feature average f Average of Generating an optimized small sample classification network w' m+1 ；
Wherein, the liquid crystal display device comprises a liquid crystal display device, w′ m+1 Is a vector of length n,
Figure BDA0002371821170000141
step 105, the electrocardiographic analysis module optimizes the small sample classification network w' m+1 And large sample classification network W m×n Performing classification network merging processing to generate an optimized heart beat data sample classification network W' (m+1)×n ；
Wherein W' (m+1)×n ＝(w 1 ，w 2 ，...w j ，...w m ，w′ m+1 ) T 。
Here, after the generation of the heart data sample classification network, the small sample feature generation network of the first embodiment may be used
Figure BDA0002371821170000142
Continuous optimization of the heart beat data sample classification network is continuously performed. The optimization method is to continuously use the generation network meeting the small sample characteristics>
Figure BDA0002371821170000143
Training small sample data sequence of input condition optimally trains the heart beat data sample classification network, and the trained characteristic average value is used for updating the small sample classification network w' m+1 . Each optimization training according to the method is one-time improvement of the small sample identification precision of the heart beat data sample classification network, so that the classification precision with better effect is obtained.
As shown in fig. 4, which is a schematic diagram of a method for identifying cardiac data based on a cardiac data sample classification network according to a third embodiment of the present invention, the method mainly includes the following steps:
step 201, an electrocardiograph analysis module acquires electrocardiograph data of a designated batch number from an electrocardiograph database of an upper computer to generate first batch electrocardiograph data, and performs heartbeat sample data extraction processing on the first batch electrocardiograph data to generate a first batch sample data sequence.
Here, as shown in FIG. 2, the present invention is implementedAs can be seen from the figure, the peak value of R point is highest in the 5 points, and the anti-interference capability of R point is strongest compared with P point and T point. Therefore, compared with the traditional 5 snack pulse signal identification method, the identification accuracy of the effective signals is improved when the heart pulse signals are identified through the R point; one heart beat sample data 2 is advanced by delta t from the point of R in the figure 1 Segment electrocardiographic data and take Δt backward 2 Combining the electrocardiographic data, wherein Deltat 1 ＝Δt 2 =first time length.
In step 202, the electrocardiographic analysis module performs large sample screening processing on the first batch of sample data sequences according to the first batch of sample data sequences to generate large sample data sequences, and performs small sample screening processing on the first batch of sample data sequences to generate small sample data sequences.
Here, because the large sample data sequence is used to train the learning network of the electrocardiographic analysis module, the sample data is all explicitly-classified sample data. Assume that the large sample data sequence is a set of sample data classification sequences { l } 1 ，l 2 ，...l i ...，l m -a }; wherein m is the total number of classifications of the large sample data sequence; l (L) i For a group of homogeneous heart beat sample data sequences { h } in a large sample data sequence 1 ，h 2 ，...h s ，...h n -a }; i has a value ranging from 1 to m; s has a value ranging from 1 to n; n is the total number of heart beat sample data of each group of heart beat sample data sequences; the small sample data sequence is assumed to be a group of similar heart beat sample data sequences { x } 1 ，x 2 ，...x s ，...x n ，y 1 ，y 2 ，...y s ，...y n }. For example, the first large sample data includes three types of large sample heart beat data: normal heart beat, atrial premature beat and ventricular premature beat, the number of large sample heart beat data in each type is 1000, and then the large sample data sequence is { l } 1 ，l 2 ，l 3 "wherein l 1 Is the normal heart beat data sequence { h1 } 1 ，h1 2 ，...，h1 1000 }，l 2 Is an atrial premature beat data sequence { h2 1 ，h2 2 ，...，h2 1000 }，l 3 Is a ventricular premature beat data sequence { h3 } 1 ，h3 2 ，...，h3 1000 Here m=3, n=1000.
Here, the small sample data sequence is a group of similar heart beat sample data sequences { x } 1 ，x 2 ，...x s ，...x n ，y 1 ，y 2 ，...y s ，...y n }. Assuming that the small sample data sequence is for learning ventricular fibrillation-type electrocardiographic data, the small sample data sequence includes 2000 ventricular fibrillation sample data.
In step 203, the electrocardiographic analysis module performs convolutional network training processing on the feature extraction network and the classification network based on the large sample data sequence, so as to generate a large sample feature extraction network and a large sample classification network.
The large sample characteristic extraction network is a convolution layer network F; the large sample feature extraction network F is used for the large sample data sequence { l } 1 ，l 2 ，...l i ...，l m L in } i All heart beat sample data { h } 1 ，h 2 ，...h s ，...h n Performing convolution network calculation to generate corresponding heart beat characteristic value sequence { f } 1 ，f 2 ，...f s ，...f n Heart beat feature value f s ＝F(h s )；
The large sample classification network is a matrix network W m×n ＝(w 1 ，w 2 ，...w i ，...w m ) T ；w i Is a matrix column vector of length n; t is the matrix transpose operator.
Here, it is assumed that the first large sample data includes three types of large sample heart beat data: normal heart beat, atrial premature beat and ventricular premature beat, the number of large sample heart beat data in each class is 1000, and the large sample classification network is a matrix network W m×n Then is W 3×1000 . It is assumed here that the dimension and the output of a single matrixThe input sample data are exactly matched, and a plurality of m multiplied by n matrixes are not required to be distributed for network output.
Here, after the large sample feature extraction network and the large sample classification network are trained, the unknown electrocardiographic data can be classified into large samples by using the calculation and classification functions of the large sample feature extraction network and the large sample classification network.
In step 204, the electrocardiographic analysis module performs convolutional network training processing on the large sample feature extraction network based on the small sample data sequence, and generates a small sample classification network.
Here, the first small sample data sequence { x } 1 ，x 2 ，...x s ，...x n The second small sample data sequence { y } comprises n pieces of heart beat sample data 1 ，y 2 ，...y s ，...y n -comprising n heart beat sample data; assuming that the small sample data sequence is for learning ventricular fibrillation-type electrocardiographic data, the small sample data sequence includes 2000 ventricular fibrillation sample data, a first small sample data sequence { x } 1 ，x 2 ，...，x 1000 The second small sample data sequence { y } comprises 1000 ventricular flutter sample data 1 ，y 2 ，...，y 1000 -1000 chamber flutter sample data;
then, the electrocardiographic analysis module performs a first small sample data sequence { x } 1 ，x 2 ，...x s ，...x n Setting to the m+1st class of the large sample data sequence; based on the first small sample data sequence { x } 1 ，x 2 ，...x s ，...x n Performing convolutional network training processing on the large sample feature extraction network F pair to generate a small sample feature generation network
Figure BDA0002371821170000163
Wherein the small sample feature generation network->
Figure BDA0002371821170000164
Is a vector with length of n;
on the support, the electrocardiographic analysis module carries out the analysis of the second small sample data sequence { y } 1 ，y 2 ，...y s ，...y n Setting to the m+1st class of the large sample data sequence; second small sample data sequence { y } using large sample feature extraction network F 1 ，y 2 ，...y s ，...y n Feature extraction calculation is carried out to generate a small sample heart beat feature sequence { f' 1 ，f′ 2 ，...f′ s ，...f′ n -a }; wherein, the small sample heart beat feature sequence { f' 1 ，f′ 2 ，...f′ s ，...f′ n -comprising n heart beat feature values;
Then, the electrocardiographic analysis module performs the heart beat feature sequence { f 'on the small sample' 1 ，f′ 2 ，...f′ s ，...f′ n Performing eigenvalue average calculation to generate small sample eigenvalue f' Average of The method comprises the steps of carrying out a first treatment on the surface of the Wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure BDA0002371821170000162
finally, the electrocardio analysis module generates a network according to the characteristics of the small sample
Figure BDA0002371821170000161
And a small sample feature average f' Average of Generating a small sample classification network w m+1 The method comprises the steps of carrying out a first treatment on the surface of the Wherein, small sample classification network w m+1 Is a vector of length n,
Figure BDA0002371821170000171
operator->
Figure BDA0002371821170000172
Is the kronecker product.
Step 205, the electrocardiographic analysis module performs classification network merging processing on the small sample classification network and the large sample classification network to generate a heart beat data sample classification network W A×B ；
Wherein, heart beat data sample classification network W A×B ＝(w 1 ，w 2 ，...w v ，...w A ) T ；w v Is a matrix column vector of length B (b=n); v takingValues range from 1 to a (a=m+1); t is the matrix transpose operator. Heart beat data sample classification network W A×B ＝(w 1 ，w 2 ，...w v ，...w A ) T ；w v Is a matrix column vector of length B; v ranges from 1 to A; t is the matrix transpose operator.
Here, the electrocardiographic analysis module classifies the small sample into a network w m+1 And large sample classification network W m×n Performing classification network merging processing to generate a heart beat data sample classification network W (m+1)×n The method comprises the steps of carrying out a first treatment on the surface of the Setting a=m+1, b=n then W (m+1)×n ＝W A×B ；
Here, generally, tens of thousands of data of each class of large sample data can satisfy the requirement of deep learning; the small sample is generally hundreds of thousands of data of each classification, if the data is trained alone, the convolutional neural network is difficult to achieve the purpose of deep learning, so that the accuracy of the classification network is affected, and if the data is trained synchronously with the large sample, the data is easy to filter out as noise by mean value. In order to solve the problems, the method of the invention firstly establishes a feature extraction or classification network by using a large sample, and then uses small sample data as optimization data of the network to perform small sample optimization training on a learning network, so that small sample data with small quantity can also obtain good classification network precision through deep learning of a convolutional neural network.
Here, steps 201 to 205 in the third embodiment are the same processing flow as steps 1 to 5 in the first embodiment.
In step 206, the electrocardiograph analysis module acquires the acquired real-time heartbeat data C from the electrocardiograph acquisition module.
Step 207, the electrocardiographic analysis module classifies the network W using the heart beat data samples A×B Classifying and detecting the real-time heart beat data C to generate a real-time heart beat prediction category and a real-time heart beat prediction category probability;
wherein, heart beat data sample classification network W A×B ＝(w 1 ，w 2 ，...w v ，...w A ) T ；w v Is a matrix column vector of length B; v ranges from 1 to A; t isMatrix transpose operators;
here, a represents the current heart beat data sample classification network W A×B The heart beat types that can be identified are a:
the method specifically comprises the following steps: step 2071, an electrocardiograph analysis module calculates a heart beat characteristic value of the real-time heart beat data C by using a large sample characteristic extraction network to generate a real-time heart beat characteristic value fC;
here, the large sample feature extraction network used for calculation is the generation of a heart beat data sample classification network W A×B A corresponding large sample feature extraction network;
step 2072, the electrocardiographic analysis module classifies the network W with the heart beat data samples A×B And real-time heart beat characteristic value fC, classifying and identifying the real-time heart beat data C to generate heart beat data C classifying and identifying vector { U } 1 ，U 2 ...U V ，...U A }；U V ＝φ*cos＜w V fC >, phi is a characteristic calculation constant, cos < w V fB > is w V Cosine value of included angle with real-time heart beat feature value fC;
here, the real-time heartbeat data C is respectively substituted into the heartbeat data sample classification network W A×B Performing feature calculation in the A-class calculation array, and further performing classification calculation by using the feature calculation result to obtain a heart beat data C classification recognition vector { U } 1 ，U 2 ...U V ，...U A Each vector score in the vector represents: the real-time heart beat data C is in the classification of class A heart beat types, the evaluation value under each classification;
step 2073, the electrocardiographic analysis module classifies the recognition vector { U } according to the heartbeat data C 1 ，U 2 ...U V ，...U A Performing heart beat classification probability calculation on the real-time heart beat data C to generate a heart beat data C classification probability sequence { P' 1 ，P′ 2 ...P′ v ，...P′ A }，
Figure BDA0002371821170000181
Here, the heart beat data C is processed by a normalization processClassification recognition vector { U ] 1 ，U 2 ...U V ，...U A Conversion to a set of normalized probabilities: cardiac data C classification probability sequence { P' 1 ，P′ 2 ...P′ v ，...P′ A ；
Step 2074, the electrocardiographic analysis module classifies the heart beat data C into probability sequence { P' 1 ，P′ 2 ...P′ v ，...P′ A Maximum statistics to generate heart beat data C maximum probability P' max Acquiring maximum probability P 'of heart beat data C' max Generating a real-time heart beat prediction category according to the corresponding heart beat type, and setting the probability of the real-time heart beat prediction category as the maximum probability P 'of heart beat data C' max 。
Here, the maximum probability is taken as a selection principle, and the maximum value is selected as the real-time heart beat prediction class probability of the real-time heart beat data C in the a output results.
In step 208, the electrocardiograph analysis module obtains a preset heart beat probability threshold value of the category according to the real-time heart beat prediction category, identifies that the real-time heart beat data C belongs to the real-time heart beat prediction category when the probability of the real-time heart beat prediction category is greater than or equal to the heart beat probability threshold value, and identifies that the real-time heart beat data C belongs to the unknown category when the probability of the real-time heart beat prediction category is less than the heart beat probability threshold value.
Here, the real-time heart beat prediction class probability of the real-time heart beat data C and the real-time heart beat prediction class to which the probability belongs have been inferred from step 207. Step 208 is to obtain the heart beat probability threshold value of the category from the preset parameters, and compare it with the estimated result (the real-time heart beat prediction category probability): if the calculated result is greater than or equal to the heart beat probability threshold value, the data identification result of the real-time heart beat data C is that: the electrocardiographic data category is a real-time heart beat prediction category; the data identification result of the real-time heartbeat data C is that if the estimated result is smaller than the heartbeat probability threshold value: the electrocardiographic data category is an unknown category.
Fig. 5A is a schematic structural diagram of a generating device of a classification network for cardiac data samples according to a fourth embodiment of the present invention, where the device includes: a processor and a memory. The memory may be coupled to the processor via a bus. The memory may be non-volatile memory, such as a hard disk drive and flash memory, in which software programs and device drivers are stored. The software program can execute various functions of the method provided by the embodiment of the invention; the device driver may be a network and interface driver. The processor is configured to execute a software program, where the software program is executed to implement the method provided by the embodiment of the present invention.
It should be noted that the embodiment of the present invention also provides a computer readable storage medium. The computer readable storage medium stores a computer program, which when executed by a processor, can implement the method provided by the embodiment of the present invention.
Embodiments of the present invention also provide a computer program product comprising instructions. The computer program product, when run on a computer, causes the processor to perform the above method.
Fig. 5B is a schematic structural diagram of a heartbeat data identification device based on a heartbeat data sample classification network according to a fifth embodiment of the present invention, where the device includes: a processor and a memory. The memory may be coupled to the processor via a bus. The memory may be non-volatile memory, such as a hard disk drive and flash memory, in which software programs and device drivers are stored. The software program can execute various functions of the method provided by the embodiment of the invention; the device driver may be a network and interface driver. The processor is configured to execute a software program, where the software program is executed to implement the method provided by the embodiment of the present invention.
It should be noted that the embodiment of the present invention also provides a computer readable storage medium. The computer readable storage medium stores a computer program, which when executed by a processor, can implement the method provided by the embodiment of the present invention.
Embodiments of the present invention also provide a computer program product comprising instructions. The computer program product, when run on a computer, causes the processor to perform the above method.
In summary, according to the method and the device for generating the heart beat data sample classification network provided by the embodiments of the present invention, the feature extraction or classification network is established by using the large sample, and then the small sample data is used as the optimization data of the network to perform the small sample optimization training on the learning network, so that the small sample data with small quantity can also obtain good classification network precision through the deep learning of the convolutional neural network. The full-sample classification network trained by combining the large sample and the small sample provided by the method can support the effective identification of the heart beat signals similar to the large sample data and can also improve the identification precision of the heart beat signals similar to the small sample data. The embodiment of the invention also provides a method and a device for identifying the heart beat data based on the heart beat data sample classification network, and the aim of carrying out classification and identification on the real-time electrocardiographic data is fulfilled by utilizing the heart beat data sample classification network generated by the method.
Those of skill would further appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various illustrative elements and steps are described above generally in terms of function in order to clearly illustrate the interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied in hardware, in a software module executed by a processor, or in a combination of the two. The software modules may be disposed in Random Access Memory (RAM), memory, read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
The foregoing description of the embodiments has been provided for the purpose of illustrating the general principles of the invention, and is not meant to limit the scope of the invention, but to limit the invention to the particular embodiments, and any modifications, equivalents, improvements, etc. that fall within the spirit and principles of the invention are intended to be included within the scope of the invention.

Claims (8)

1. A method of generating a classification network of heart beat data samples, the method comprising:
the method comprises the steps that an electrocardio analysis module acquires electrocardio data with a specified batch number from an electrocardio database of an upper computer to generate first batch electrocardio data, and heart beat sample data extraction processing is carried out on the first batch electrocardio data to generate a first batch sample data sequence;
the electrocardiographic analysis module performs large sample screening processing on the first batch of sample data sequences according to the first batch of sample data sequences to generate large sample data sequences, and performs small sample screening processing on the first batch of sample data sequences to generate small sample data sequences;
the electrocardio analysis module carries out convolutional network training treatment on the characteristic extraction network and the classification network based on the large sample data sequence to generate a large sample characteristic extraction network and a large sample classification network;
the electrocardio analysis module carries out convolutional network training treatment on the large sample characteristic extraction network based on the small sample data sequence to generate a small sample classification network;
the electrocardio analysis module performs classification network merging processing on the small sample classification network and the large sample classification network to generate a heart beat data sample classification network;
Wherein, the liquid crystal display device comprises a liquid crystal display device,
the electrocardio analysis module acquires electrocardio data of a designated batch number from an electrocardio database of an upper computer to generate first batch electrocardio data, and carries out heart beat sample data extraction processing on the first batch electrocardio data to generate a first batch sample data sequence, and the method specifically comprises the following steps:
the electrocardio analysis module acquires the electrocardio data of the appointed batch number from the electrocardio database of the upper computer to generate the first batch electrocardio data;
the electrocardio analysis module acquires a preset heart beat sample extraction time length and generates a first time length; initializing the first batch of sample data sequences to be empty;
the electrocardio analysis module polls all the electrocardio data in the first batch of electrocardio data, carries out heart beat R point position confirmation processing on the currently polled electrocardio data to generate a current R point position sequence, and extracts the electrocardio data with the first time length forwards and backwards respectively by taking each heart beat R point position in the current R point position sequence as a center to be combined into heart beat sample data;
the electrocardiographic analysis module adds all the extracted heart beat sample data to the first batch of sample data sequences to obtain sample data objects;
The method further comprises the steps of:
the large sample data sequence is a group of sample data classification sequences { l } 1 ，l 2 ，...l i ...，l m -a }; the m is the total classification number of the large sample data sequence; the l is i For a group of homogeneous cardiac sample data sequences { h } in said large sample data sequence 1 ，h 2 ，...h s ，...h n -a }; the value range of i is from 1 to m; the value range of s is from 1 to n; n is the total number of heart beat sample data of each group of heart beat sample data sequences;
the small sample data sequence is a group of similar heart beat sample data sequences { x } 1 ，x 2 ，...x s ，...x m ，y 1 ，y 2 ，...y s ，...y n }；
The electrocardiograph analysis module carries out convolution network training processing on the characteristic extraction network and the classification network based on the large sample data sequence to generate a large sample characteristic extraction network and a large sample classification network, and specifically comprises the following steps:
the electrocardiographic analysis module is based on the large sample data sequence { l } 1 ，l 2 ，...l i ...，l m Performing convolutional network training processing on the characteristic extraction network to generate the large sample characteristic extraction network; the large sample characteristic extraction network is a convolution layer network F; the large sample feature extraction network F is used for extracting the large sample data sequence { l } 1 ，l 2 ，...l i ...，l m Said l in } i All heart beat sample data { h } 1 ，h 2 ，...h s ，...h n Performing convolution network calculation to generate corresponding heart beat characteristic value sequence { f } 1 ，f 2 ，...f s ，...f n -the heart beat feature value f s ＝F(h s )；
The electrocardiographic analysis module is based on the large sample data sequence { l } 1 ，l 2 ，...l i ...，l m Performing convolutional network training processing on the classification network to generate the large sample classification network; the large sample classification network is a matrix network W m×n ＝(w 1 ，w 2 ，...w i ，...w m ) T The method comprises the steps of carrying out a first treatment on the surface of the The w is i Is a matrix column vector of length n; the T is a matrix transpose operator;
the electrocardiograph analysis module carries out convolution network training processing on the large sample characteristic extraction network based on the small sample data sequence to generate a small sample classification network, and the method specifically comprises the following steps:
the electrocardiographic analysis module is used for carrying out the analysis on the small sample data sequence { x } 1 ，x 2 ，...x s ，...x m ，y 1 ，y 2 ，...y s ，...y n Dividing into two parts, and generating a first small sample data sequence and a second small sample data sequence; the first small sample data sequence { x } 1 ，x 2 ，...x s ，...x m Comprises n heart beat sample data, the second small sample data sequence { y } 1 ，y 2 ，...y s ，...y n -comprising n heart beat sample data;
the electrocardiographic analysis module analyzes the first small sample data sequence { x } 1 ，x 2 ，...x s ，...x m Setting to the m+1st class of the large sample data sequence; based on the first small sample data sequence { x } 1 ，x 2 ，...x s ，...x m Performing convolutional network training processing on the large sample feature extraction network F to generate a small sample feature generation network
Figure FDA0004083831340000035
Said small sample feature generation network- >
Figure FDA0004083831340000036
Is a vector with length of n;
the electrocardiographic analysis module sends the second small sample data sequence { y }, to the computer system 1 ，y 2 ，...y s ，...y n Setting to the m+1st class of the large sample data sequence; using the large sample feature extraction network F for the second small sample data sequence { y } 1 ,y 2 ，...y s ，...y n Feature extraction calculation is carried out to generate a small sample heart beat feature sequence { f' 1 ，f′ 2 ，...f′ s ，...f′ n -a }; the small sample cardiac feature sequence { f' 1 ，f′ 2 ，...f′ s ，...f′ n -comprising n heart beat feature values;
the electrocardiographic analysis module performs a heart beat feature sequence { f 'on the small sample' 1 ，f′ 2 ，...f′ s ，...f′ n Performing eigenvalue average calculation to generate small sample eigenvalue f' Average of The said
Figure FDA0004083831340000031
The electrocardiographic analysis module generates a network according to the small sample characteristics
Figure FDA0004083831340000032
And the small sample feature average f' Average of Generating the smallSample classification network w m+1 The method comprises the steps of carrying out a first treatment on the surface of the The w is m+1 Is a vector of length n, said +.>
Figure FDA0004083831340000033
Said operator->
Figure FDA0004083831340000034
Is the product of kronecker;
the electrocardiographic analysis module performs classification network merging processing on the small sample classification network and the large sample classification network to generate a heart beat data sample classification network, and specifically comprises the following steps:
the electrocardiographic analysis module classifies the small sample into a network w m+1 And the large sample classification network W m×n Performing classification network merging processing to generate the heart beat data sample classification network W (m+1)×n The method comprises the steps of carrying out a first treatment on the surface of the The W is (m+1)×n ＝(w 1 ，w 2 ，...，w j ...w m ，w m+1 ) T The method comprises the steps of carrying out a first treatment on the surface of the The value of j ranges from 1 to m+1.
2. The method of generating a classification network of cardiac data samples according to claim 1, wherein the method further comprises:
after the heart beat data sample classification network is established, when the electrocardio analysis module acquires a new training small sample data sequence { z }, the data sequence of the training small sample is obtained 1 ，z 2 ，...，z n At the time of } the electrocardiographic analysis module transmits the training small sample data sequence { z }, the electrocardiographic analysis module generates a training small sample data sequence { z }, a training small sample data sequence 1 ，z 2 ，...，z n Setting to the m+1st class of the large sample data sequence;
the electrocardiographic analysis module uses the large sample feature extraction network F to train the small sample data sequence { z } 1 ，z 2 ，...，z n Feature extraction calculation is carried out to generate a training small sample heart beat feature sequence { f } ", and the training small sample heart beat feature sequence { f }", the training small sample heart beat feature sequence { f } 1 ，f″ 2 ，...，f″ i ，...f″ n -a }; the training small sample data sequence { z } 1 ，z 2 ，...，z n -comprising n heart beat sample data; the training small sample heart beat characteristic sequence { f } ", is 1 ，f″ 2 ，...，f″ i ，...f″ n -including n of said heart beat feature values;
the electrocardiographic analysis module performs the training on the heart beat characteristic sequence { f } ", of the small sample 1 ，f″ 2 ，...，f″ i ，...f″ n Performing characteristic value average calculation to generate training small sample characteristic average value f Average of The said
Figure FDA0004083831340000041
Figure FDA0004083831340000042
The electrocardiographic analysis module generates a network according to the small sample characteristics
Figure FDA0004083831340000043
And the training small sample characteristic average value f Average of Generating an optimized small sample classification network w' m+1 The method comprises the steps of carrying out a first treatment on the surface of the The w' m+1 Is a vector with length of n, the
Figure FDA0004083831340000044
The electrocardiographic analysis module classifies the optimized small sample into a network w' m+1 And the large sample classification network W m×n Performing classification network merging processing to generate an optimized heart beat data sample classification network W' (m+1)×n The method comprises the steps of carrying out a first treatment on the surface of the The W' (m+1)×n ＝(w 1 ，w 2 ，...w j ，...w m ，w′ m+1 ) T 。
3. An apparatus comprising a memory for storing a program and a processor for performing the method of any of claims 1 to 2.
4. A computer readable storage medium comprising instructions which, when run on a computer, cause the computer to perform the method of any one of claims 1 to 2.
5. A method of identifying cardiac data based on a cardiac data sample classification network, the method comprising:
the method comprises the steps that an electrocardio analysis module acquires electrocardio data with a specified batch number from an electrocardio database of an upper computer to generate first batch electrocardio data, and heart beat sample data extraction processing is carried out on the first batch electrocardio data to generate a first batch sample data sequence;
the electrocardiographic analysis module performs large sample screening processing on the first batch of sample data sequences according to the first batch of sample data sequences to generate large sample data sequences, and performs small sample screening processing on the first batch of sample data sequences to generate small sample data sequences;
The electrocardio analysis module carries out convolutional network training treatment on the characteristic extraction network and the classification network based on the large sample data sequence to generate a large sample characteristic extraction network and a large sample classification network;
the electrocardio analysis module carries out convolutional network training treatment on the large sample characteristic extraction network based on the small sample data sequence to generate a small sample classification network;
the electrocardiographic analysis module performs classification network merging processing on the small sample classification network and the large sample classification network to generate a heart beat data sample classification network W A×B The method comprises the steps of carrying out a first treatment on the surface of the The heart beat data sample classification network W A×B ＝(w 1 ，w 2 ，...w v ，...w A ) T The method comprises the steps of carrying out a first treatment on the surface of the The w is v Is a matrix column vector of length B; the value range of v is from 1 to A; the T is a matrix transpose operator;
the electrocardio analysis module acquires acquired real-time heart beat data C from the electrocardio acquisition module;
the electrocardiographic analysis module uses the heart beat dataSample classification network W A×B Classifying and detecting the real-time heart beat data C to generate a real-time heart beat prediction category and a real-time heart beat prediction category probability;
the electrocardio analysis module acquires a preset heart beat probability threshold value of the category according to the real-time heart beat prediction category, identifies that the real-time heart beat data C belongs to the real-time heart beat prediction category when the real-time heart beat prediction category probability is larger than or equal to the heart beat probability threshold value, and identifies that the real-time heart beat data C belongs to an unknown category when the real-time heart beat prediction category probability is smaller than the heart beat probability threshold value;
Wherein, the liquid crystal display device comprises a liquid crystal display device,
the electrocardio analysis module acquires electrocardio data of a designated batch number from an electrocardio database of an upper computer to generate first batch electrocardio data, and carries out heart beat sample data extraction processing on the first batch electrocardio data to generate a first batch sample data sequence, and the method specifically comprises the following steps:
the electrocardio analysis module acquires the electrocardio data of the appointed batch number from the electrocardio database of the upper computer to generate the first batch electrocardio data;
the electrocardio analysis module acquires a preset heart beat sample extraction time length and generates a first time length; initializing the first batch of sample data sequences to be empty;
the electrocardio analysis module polls all the electrocardio data in the first batch of electrocardio data, carries out heart beat R point position confirmation processing on the currently polled electrocardio data to generate a current R point position sequence, and extracts the electrocardio data with the first time length forwards and backwards respectively by taking each heart beat R point position in the current R point position sequence as a center to be combined into heart beat sample data;
the electrocardiographic analysis module adds all the extracted heart beat sample data to the first batch of sample data sequences to obtain sample data objects;
The method further comprises the steps of:
the large sample data sequence is a group of sample data classification sequences { l } 1 ，l 2 ，...l i ...，l m -a }; the m is the total classification number of the large sample data sequence; the l is i For a group of homogeneous cardiac sample data sequences { h } in said large sample data sequence 1 ，h 2 ，...h s ，...h n -a }; the value range of i is from 1 to m; the value range of s is from 1 to n; n is the total number of heart beat sample data of each group of heart beat sample data sequences;
the small sample data sequence is a group of similar heart beat sample data sequences { x } 1 ，x 2 ，...x s ，...x n ，y 1 ，y 2 ，...y s ，...y n }；
The electrocardiograph analysis module carries out convolution network training processing on the characteristic extraction network and the classification network based on the large sample data sequence to generate a large sample characteristic extraction network and a large sample classification network, and specifically comprises the following steps:
the electrocardiographic analysis module is based on the large sample data sequence { l } 1 ，l 2 ，...l i ...，l m Performing convolutional network training processing on the characteristic extraction network to generate the large sample characteristic extraction network; the large sample characteristic extraction network is a convolution layer network F; the large sample feature extraction network F is used for extracting the large sample data sequence { l } 1 ，l 2 ，...l i ...，l m Said l in } i All heart beat sample data { h } 1 ，h 2 ，...h s ，...h n Performing convolution network calculation to generate corresponding heart beat characteristic value sequence { f } 1 ，f 2 ，...f s ，...f n -the heart beat feature value f s ＝F(h s )；
The electrocardiographic analysis module is based on the large sample data sequence { l } 1 ，l 2 ，...l i ...，l m Performing convolutional network training processing on the classification network to generate the large sample classification network; the large sample classification network is a matrix network W m×n ＝(w 1 ，w 2 ，...w i ，...w m ) T The method comprises the steps of carrying out a first treatment on the surface of the By a means ofThe w is i Is a matrix column vector of length n; the T is a matrix transpose operator;
the electrocardiograph analysis module carries out convolution network training processing on the large sample characteristic extraction network based on the small sample data sequence to generate a small sample classification network, and the method specifically comprises the following steps:
the electrocardiographic analysis module is used for carrying out the analysis on the small sample data sequence { x } 1 ，x 2 ，...x s ，...x m ，y 1 ,y 2 ，...y s ，...y n Dividing into two parts, and generating a first small sample data sequence and a second small sample data sequence; the first small sample data sequence { x } 1 ，x 2 ，...x s ，...x m Comprises n heart beat sample data, the second small sample data sequence { y } 1 ，y 2 ，...y s ，...y n -comprising n heart beat sample data;
the electrocardiographic analysis module analyzes the first small sample data sequence { x } 1 ，x 2 ，...x s ，...x n Setting to the m+1st class of the large sample data sequence; based on the first small sample data sequence { x } 1 ，x 2 ，...x s ，...x n Performing convolutional network training processing on the large sample feature extraction network F to generate a small sample feature generation network
Figure FDA0004083831340000075
Said small sample feature generation network- >
Figure FDA0004083831340000076
Is a vector with length of n;
the electrocardiographic analysis module sends the second small sample data sequence { y }, to the computer system 1 ，y 2 ，...y s ，...y n Setting to the m+1st class of the large sample data sequence; using the large sample feature extraction network F for the second small sample data sequence { y } 1 ,y 2 ，...y s ，...y n Feature extraction calculation is carried out to generate small samplesHeart beat feature sequence { f' 1 ，f′ 2 ，...f′ s ，...f′ n -a }; the small sample cardiac feature sequence { f' 1 ，f′ 2 ，...f′ s ，...f′ n -comprising n heart beat feature values;
the electrocardiographic analysis module performs a heart beat feature sequence { f 'on the small sample' 1 ，f′ 2 ，...f′ s ，...f′ n Performing eigenvalue average calculation to generate small sample eigenvalue f' Average of The said
Figure FDA0004083831340000071
The electrocardiographic analysis module generates a network according to the small sample characteristics
Figure FDA0004083831340000072
And the small sample feature average f' Average of Generating the small sample classification network w m+1 The method comprises the steps of carrying out a first treatment on the surface of the The w is m+1 Is a vector of length n, said +.>
Figure FDA0004083831340000073
Said operator->
Figure FDA0004083831340000074
Is the product of kronecker;
the electrocardiographic analysis module performs classification network merging processing on the small sample classification network and the large sample classification network to generate a heart beat data sample classification network, and specifically comprises the following steps:
the electrocardiographic analysis module classifies the small sample into a network w m+1 And the large sample classification network W m×n Performing classification network merging processing to generate the heart beat data sample classification network W (m+1)×n The method comprises the steps of carrying out a first treatment on the surface of the The W is (m+1)×n ＝(w 1 ，w 2 ，...，w j ...w m ，w m+1 ) T The method comprises the steps of carrying out a first treatment on the surface of the The value of j ranges from 1 to m+1.
6. The method for identifying cardiac data based on a cardiac data sample classification network of claim 5, wherein the electrocardiographic analysis module uses the cardiac data sample classification network W A×B The real-time heartbeat data C is subjected to classification detection to generate a real-time heartbeat prediction category and a real-time heartbeat prediction category probability, and the method specifically comprises the following steps:
the electrocardio analysis module calculates a heart beat characteristic value of the real-time heart beat data C by using the large sample characteristic extraction network to generate a real-time heart beat characteristic value fC;
the electrocardiographic analysis module uses the heart beat data sample classification network W A×B And the real-time heart beat characteristic value fC performs classification and identification processing on the real-time heart beat data C to generate a real-time heart beat data C classification and identification vector { U } 1 ，U 2 ...U V ，...U A -a }; the U is V ＝φ*cos<w V ，fC>The phi is a characteristic calculation constant, and the cos<w V ，fC>For the w V Cosine value of the included angle with the real-time heart beat feature value fC;
the electrocardio analysis module classifies and identifies a vector { U } according to the real-time heartbeat data C 1 ，U 2 ...U V ，...U A Performing heart beat classification probability calculation on the real-time heart beat data C to generate a real-time heart beat data C classification probability sequence { P' 1 ，P′ 2 ...P′ v ，...P′ A -said
Figure FDA0004083831340000081
The electrocardiographic analysis module classifies probability sequences { P 'of the real-time heartbeat data C' 1 ，P′ 2 ...P′ v ，...P′ A Maximum statistics to generate real-time heart beat data C maximum probability P' max Acquiring the maximum probability P 'of the real-time heartbeat data C' max Generating the real-time heart beat prediction category according to the corresponding heart beat type, and setting the probability of the real-time heart beat prediction category as the real-time heartMaximum probability P 'of beat data C' max 。
7. An apparatus comprising a memory for storing a program and a processor for performing the method of any of claims 5 to 6.
8. A computer readable storage medium comprising instructions which, when run on a computer, cause the computer to perform the method of any one of claims 5 to 6.
CN202010057334.8A 2020-01-17 2020-01-17 Method and device for generating heart beat data sample classification network Active CN111310578B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010057334.8A CN111310578B (en) 2020-01-17 2020-01-17 Method and device for generating heart beat data sample classification network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010057334.8A CN111310578B (en) 2020-01-17 2020-01-17 Method and device for generating heart beat data sample classification network

Publications (2)

Publication Number Publication Date
CN111310578A CN111310578A (en) 2020-06-19
CN111310578B true CN111310578B (en) 2023-05-02

Family

ID=71148290

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010057334.8A Active CN111310578B (en) 2020-01-17 2020-01-17 Method and device for generating heart beat data sample classification network

Country Status (1)

Country Link
CN (1) CN111310578B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107169527A (en) * 2017-06-06 2017-09-15 西北工业大学 Classification method of medical image based on collaboration deep learning
CN109063761A (en) * 2018-07-20 2018-12-21 北京旷视科技有限公司 Diffuser dropping detection method, device and electronic equipment
CN110288028A (en) * 2019-06-27 2019-09-27 北京邮电大学 ECG detecting method, system, equipment and computer readable storage medium
CN110427846A (en) * 2019-07-19 2019-11-08 西安工业大学 It is a kind of using convolutional neural networks to the face identification method of uneven small sample
CN110490802A (en) * 2019-08-06 2019-11-22 北京观微科技有限公司 A kind of satellite image Aircraft Targets type identifier method based on super-resolution

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107169527A (en) * 2017-06-06 2017-09-15 西北工业大学 Classification method of medical image based on collaboration deep learning
CN109063761A (en) * 2018-07-20 2018-12-21 北京旷视科技有限公司 Diffuser dropping detection method, device and electronic equipment
CN110288028A (en) * 2019-06-27 2019-09-27 北京邮电大学 ECG detecting method, system, equipment and computer readable storage medium
CN110427846A (en) * 2019-07-19 2019-11-08 西安工业大学 It is a kind of using convolutional neural networks to the face identification method of uneven small sample
CN110490802A (en) * 2019-08-06 2019-11-22 北京观微科技有限公司 A kind of satellite image Aircraft Targets type identifier method based on super-resolution

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
Guneet S. Dhillon 等.A BASELINE FOR FEW-SHOT IMAGE CLASSIFICATION.arXiv.org.2019,1-19. *

Also Published As

Publication number Publication date
CN111310578A (en) 2020-06-19

Similar Documents

Publication Publication Date Title
CN107822622B (en) Electrocardiogram diagnosis method and system based on deep convolutional neural network
CN111160139B (en) Electrocardiosignal processing method and device and terminal equipment
CN107837082B (en) Automatic electrocardiogram analysis method and device based on artificial intelligence self-learning
Khafaga et al. Optimization of Electrocardiogram Classification Using Dipper Throated Algorithm and Differential Evolution.
CN109948647B (en) Electrocardiogram classification method and system based on depth residual error network
CN109998525B (en) Arrhythmia automatic classification method based on discriminant deep belief network
Haseena et al. Fuzzy clustered probabilistic and multi layered feed forward neural networks for electrocardiogram arrhythmia classification
CN111134662B (en) Electrocardio abnormal signal identification method and device based on transfer learning and confidence degree selection
CN111626114B (en) Electrocardiosignal arrhythmia classification system based on convolutional neural network
Zhang et al. ECG signal classification with deep learning for heart disease identification
WO2021017313A1 (en) Atrial fibrillation detection method and apparatus, computer device, and storage medium
CN110638430B (en) Method for building cascade neural network ECG signal arrhythmia classification model
CN110558975B (en) Electrocardiosignal classification method and system
CN112426160A (en) Electrocardiosignal type identification method and device
Rohmantri et al. Arrhythmia classification using 2D convolutional neural network
CN114052744A (en) Electrocardiosignal classification method based on pulse neural network
Lu et al. A novel deep learning based neural network for heartbeat detection in ballistocardiograph
CN115221926A (en) Heart beat signal classification method based on CNN-GRU network model
Rahuja et al. A deep neural network approach to automatic multi-class classification of electrocardiogram signals
CN111053552A (en) QRS wave detection method based on deep learning
Tao et al. ECG-based expert-knowledge attention network to tachyarrhythmia recognition
Kumari et al. Optimization of multi-layer perceptron neural network using genetic algorithm for arrhythmia classification
CN113288157A (en) Arrhythmia classification method based on depth separable convolution and improved loss function
CN111310578B (en) Method and device for generating heart beat data sample classification network
Song et al. Automatic identification of atrial fibrillation based on the modified Elman neural network with exponential moving average algorithm

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
CB02 Change of applicant information
CB02 Change of applicant information

Address after: 201612 16 / F, block a, no.668, Xinzhuan Road, Songjiang District, Shanghai

Applicant after: Shanghai Lepu Yunzhi Technology Co.,Ltd.

Address before: 201612 16 / F, block a, no.668, Xinzhuan Road, Songjiang District, Shanghai

Applicant before: SHANGHAI YOCALY HEALTH MANAGEMENT Co.,Ltd.

GR01 Patent grant
GR01 Patent grant