CN112120691A - Signal identification method and device based on deep learning and computer equipment - Google Patents
Signal identification method and device based on deep learning and computer equipment Download PDFInfo
- Publication number
- CN112120691A CN112120691A CN202011000576.XA CN202011000576A CN112120691A CN 112120691 A CN112120691 A CN 112120691A CN 202011000576 A CN202011000576 A CN 202011000576A CN 112120691 A CN112120691 A CN 112120691A
- Authority
- CN
- China
- Prior art keywords
- deep learning
- groups
- electrocardiogram
- inputting
- submodel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7225—Details of analog processing, e.g. isolation amplifier, gain or sensitivity adjustment, filtering, baseline or drift compensation
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7253—Details of waveform analysis characterised by using transforms
- A61B5/7257—Details of waveform analysis characterised by using transforms using Fourier transforms
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7253—Details of waveform analysis characterised by using transforms
- A61B5/726—Details of waveform analysis characterised by using transforms using Wavelet transforms
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Physiology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- Pathology (AREA)
- Veterinary Medicine (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Psychiatry (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Power Engineering (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The application relates to a signal identification method, a signal identification device and computer equipment based on deep learning, wherein the signal identification method based on deep learning comprises the following steps: compared with the related art, the signal identification method based on deep learning provided by the embodiment of the application obtains the electrocardiogram signals, preprocesses the electrocardiogram signals to obtain multiple groups of preprocessing information, inputs the multiple groups of preprocessing information into the first deep learning model to obtain multiple groups of initial deep characteristics, wherein the first deep learning model comprises a plurality of deep learning submodels, the multiple groups of preprocessing information correspond to the plurality of deep learning submodels, and inputs the multiple groups of initial deep characteristics into the second deep learning model to obtain a final identification result, so that the problem that the classification input characteristics used in the related art are too extensive, the identified information precision is low is solved, and the identified information precision is improved.
Description
Technical Field
The present application relates to the field of signal processing technologies, and in particular, to a method and an apparatus for signal recognition based on deep learning, and a computer device.
Background
Deep learning algorithms in signal processing, the two most commonly used: long Short-term Memory (LSTM) and Convolutional Neural Networks (CNN). The LSTM is a special recurrent neural network and is the algorithm which is used for processing the time sequence signal longest, theoretically, an Electrocardiogram (ECG) signal is a non-stable, non-linear and strong-randomness time sequence signal, and the LSTM is used for ECG time sequence feature extraction and is very suitable, but the LSTM network is not suitable for processing overlong time sequences due to the problem of long-term dependence; CNN is a feedforward neural network containing convolution calculation and having a deep structure, is constructed by simulating a visual mechanism of a living being, and enables the convolutional neural network to learn with a small calculation amount due to parameter sharing of convolution kernels in hidden layers and sparsity of connection among layers.
In the related art, there are several methods for identifying Atrial Fibrillation data (AF) based on an electrocardiogram signal: the first approach, using ECGQT interval, QRS slope, etc. features widely used by clinicians, as well as RR interval variability, HRV variability, statistical features, etc. for identification of AF, has the disadvantage of: 1. the used characteristics need to be extracted by a separately designed algorithm, and hundreds of different characteristics are extracted, so that the algorithm is too complex; 2. the automatic measurement algorithm for detecting P waves and QT intervals has large errors, is easily influenced by noise and unreliable in many cases, and the characteristics easily cause the problems of overfitting, incapability of convergence of the algorithm and the like; 3. although hundreds of different features are used, the characteristics of signals can be described comprehensively and are difficult to ensure; 4. many of the used features are one-dimensional features, which are too simple to describe the characteristics of the signal more comprehensively. In the second mode, the ECG signal or the instantaneous frequency spectrum of the ECG is used as a time sequence signal, and the LSTM network is used to identify AF, so that the ECG signal is used as an input, and the algorithm effect is poor due to the long-term dependence problem. Although the classification accuracy is improved by classifying AF by using the instantaneous frequency spectrum and the frequency spectrum entropy of an ECG signal as input (the input data volume is greatly reduced), the AF heart rhythm is only identified from the perspective of the ECG instantaneous frequency, so that the AF heart rhythm is too complete, and partial diagnosis rules of doctors cannot be used for reference. In the third mode, CNN is used for carrying out convolution feature extraction on an original electrocardiogram signal, an LSTM network is used for carrying out feature extraction on a representative heart beat waveform 'centerwave' in a section of the original electrocardiogram signal, and finally, feature combinations are classified by using XGBOST; the features used are still too comprehensive, ignoring the time-frequency information of the signal, instantaneous frequency variations, and the timing features of the ECG signal for great help in AF identification. Therefore, it is very difficult to accurately recognize AF due to the complexity of the ECG signal, and there is a problem that the accuracy of the recognized information is low because the input features are too broad in many cases.
At present, aiming at the problem that the classification input features used in the related technology are too one-sided and the accuracy of the identified information is low, an effective solution is not provided.
Disclosure of Invention
The embodiment of the application provides a signal identification method and device based on deep learning and computer equipment, and aims to at least solve the problems that classification input features used in the related technology are too extensive and low in identification information precision.
In a first aspect, an embodiment of the present application provides a signal identification method based on deep learning, where the method includes: acquiring an electrocardiogram signal;
preprocessing the electrocardiogram signals to obtain a plurality of groups of preprocessing information;
inputting a plurality of groups of preprocessing information into a first deep learning model to obtain a plurality of groups of initial deep features, wherein the first deep learning model comprises a plurality of deep learning submodels, and the plurality of groups of preprocessing information correspond to the plurality of deep learning submodels;
and inputting the plurality of groups of initial depth features into a second deep learning model to obtain a final recognition result.
In some embodiments, the electrocardiogram signal is preprocessed to obtain a plurality of groups of preprocessed information; inputting a plurality of groups of the preprocessing information into a first deep learning model, and obtaining a plurality of groups of initial depth features comprises:
extracting medical features from the electrocardiogram signals;
inputting the medical features into a first deep learning submodel corresponding to the medical features to obtain initial deep features corresponding to the medical features, wherein the first deep learning submodel comprises a plurality of layers of fully-connected feedforward neural networks, regularization terms are added to the fully-connected feedforward neural networks, and a learning rate attenuation strategy is adopted in the process of training the first deep learning submodel.
In some embodiments, the electrocardiogram signal is preprocessed to obtain a plurality of groups of preprocessed information; inputting a plurality of groups of the preprocessing information into a first deep learning model, and obtaining a plurality of groups of initial depth features comprises:
performing wavelet transform or short-time Fourier transform on the electrocardiogram signal to obtain a time-frequency spectrogram;
and inputting the time-frequency spectrogram into a second deep learning submodel corresponding to the time-frequency spectrogram to obtain initial depth features corresponding to the time-frequency spectrogram, wherein the second deep learning submodel adopts a migration learning mode based on a convolutional neural network and adds a classification layer in the trained convolutional neural network.
In some embodiments, the electrocardiogram signal is preprocessed to obtain a plurality of groups of preprocessed information; inputting a plurality of groups of the preprocessing information into a first deep learning model, and obtaining a plurality of groups of initial depth features comprises:
carrying out instantaneous frequency spectrum and frequency spectrum entropy analysis on the electrocardiogram signal to obtain an instantaneous frequency spectrum and a frequency spectrum entropy;
inputting the instantaneous frequency spectrum and the frequency spectrum entropy into a corresponding third deep learning submodel to obtain corresponding initial depth features, wherein the third deep learning submodel adopts a bidirectional long-time and short-time memory network, the bidirectional long-time and short-time memory network is added with an attention strategy and a regularization item, and a learning rate attenuation strategy is adopted in the process of training the third deep learning submodel.
In some embodiments, the electrocardiogram signal is preprocessed to obtain a plurality of groups of preprocessed information; inputting a plurality of groups of the preprocessing information into a first deep learning model, and obtaining a plurality of groups of initial depth features comprises:
performing down-sampling processing on the electrocardiogram signals to obtain down-sampled electrocardiogram sequences;
cutting the down-sampled electrocardiogram sequence to obtain a cut-off electrocardiogram sequence;
and inputting the truncated electrocardiogram sequence into a corresponding fourth deep learning submodel to obtain corresponding initial depth features, wherein the fourth deep learning submodel adopts a bidirectional long-time and short-time memory network, the bidirectional long-time and short-time memory network is added with an attention strategy and a regularization item, and a learning rate attenuation strategy is adopted in the process of training the fourth deep learning submodel.
In some embodiments, the electrocardiogram signal is preprocessed to obtain a plurality of groups of preprocessed information; inputting a plurality of groups of the preprocessing information into a first deep learning model, and obtaining a plurality of groups of initial depth features comprises:
carrying out normalization processing on the electrocardiogram signals to obtain an electrocardiogram sequence after normalization processing;
inputting the electrocardiogram sequence into a corresponding fifth deep learning submodel to obtain corresponding initial depth features, wherein the fifth deep learning submodel adopts a cavity convolution and a convolution function to perform one-dimensional convolution filtering and one-dimensional convolution, a regularization item is added to the fifth deep learning submodel, and a learning rate attenuation strategy is adopted in the process of training the fifth deep learning submodel.
In some of these embodiments, for multi-lead electrocardiographic signals, the method further comprises:
acquiring final recognition results of a plurality of single leads, and counting the number of the same final recognition results;
and taking the final recognition result with the largest number as the final recognition result of the multi-lead electrocardiogram signals.
In a second aspect, an embodiment of the present application provides a signal identification apparatus based on deep learning, where the apparatus includes: the device comprises an acquisition module, a preprocessing module, a first learning module and a second learning module;
the acquisition module is used for acquiring electrocardiogram signals;
the preprocessing module is used for preprocessing the electrocardiogram signals to obtain a plurality of groups of preprocessing information;
the first learning module is used for inputting a plurality of groups of preprocessing information into a first deep learning model to obtain a plurality of groups of initial deep features, wherein the first deep learning model comprises a plurality of deep learning submodels, and the plurality of groups of preprocessing information correspond to the plurality of deep learning submodels;
and the second learning module is used for inputting the plurality of groups of initial depth features into a second deep learning model to obtain a final recognition result.
In a third aspect, an embodiment of the present application provides a computer device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the computer program, the processor implements the deep learning based signal identification method according to the first aspect.
In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, which when executed by a processor implements the deep learning-based signal identification method according to the first aspect.
Compared with the related art, the signal identification method, the signal identification device and the computer equipment based on deep learning provided by the embodiment of the application have the advantages that through acquiring an electrocardiogram signal, the electrocardiogram signal is preprocessed to obtain multiple groups of preprocessed information, the multiple groups of preprocessed information are input into the first deep learning model to obtain multiple groups of initial deep features, the first deep learning model comprises multiple deep learning submodels, the multiple groups of preprocessed information correspond to the multiple deep learning submodels, the multiple groups of initial deep features are input into the second deep learning model to obtain a final identification result, the problems that the classification input features used in the related art are too extensive, the identified information precision is low are solved, and the identified information precision is improved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the application and together with the description serve to explain the application and not to limit the application. In the drawings:
fig. 1 is a flow chart one of a signal identification method based on deep learning according to an embodiment of the present application;
FIG. 2 is a flow chart of a method of obtaining an initial depth feature corresponding to a medical feature according to an embodiment of the present application;
FIG. 3 is a flowchart of a method for obtaining initial depth features corresponding to a time-frequency spectrogram according to an embodiment of the present disclosure;
FIG. 4 is a flow chart of a method of obtaining initial depth features corresponding to temporal spectrum and spectral entropy according to an embodiment of the present application;
FIG. 5 is a flow chart of a method of obtaining initial depth features corresponding to truncated electrocardiographic sequences according to an embodiment of the present application;
FIG. 6 is a flow chart of a method of obtaining initial depth features corresponding to an original electrocardiogram sequence according to an embodiment of the present application;
FIG. 7 is a flowchart II of a signal identification method based on deep learning according to an embodiment of the present application;
FIG. 8 is a recognition framework based on single-lead ECG signals according to an embodiment of the present application;
FIG. 9 is a multi-lead ECG signal based recognition framework according to an embodiment of the present application;
fig. 10 is a block diagram of a deep learning based signal recognition apparatus according to an embodiment of the present application;
fig. 11 is a schematic diagram of an internal structure of a computer device according to an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments provided in the present application without any inventive step are within the scope of protection of the present application.
It is obvious that the drawings in the following description are only examples or embodiments of the present application, and that it is also possible for a person skilled in the art to apply the present application to other similar contexts on the basis of these drawings without inventive effort. Moreover, it should be appreciated that in the development of any such actual implementation, as in any engineering or design project, numerous implementation-specific decisions must be made to achieve the developers' specific goals, such as compliance with system-related and business-related constraints, which may vary from one implementation to another.
Reference in the specification to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the specification. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those of ordinary skill in the art will explicitly and implicitly appreciate that the embodiments described herein may be combined with other embodiments without conflict.
Unless defined otherwise, technical or scientific terms referred to herein shall have the ordinary meaning as understood by those of ordinary skill in the art to which this application belongs. Reference to "a," "an," "the," and similar words throughout this application are not to be construed as limiting in number, and may refer to the singular or the plural. The present application is directed to the use of the terms "including," "comprising," "having," and any variations thereof, which are intended to cover non-exclusive inclusions; for example, a process, method, system, article, or apparatus that comprises a list of steps or modules (elements) is not limited to the listed steps or elements, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus. Reference to "connected," "coupled," and the like in this application is not intended to be limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The term "plurality" as referred to herein means two or more. "and/or" describes an association relationship of associated objects, meaning that three relationships may exist, for example, "A and/or B" may mean: a exists alone, A and B exist simultaneously, and B exists alone. The character "/" generally indicates that the former and latter associated objects are in an "or" relationship. Reference herein to the terms "first," "second," "third," and the like, are merely to distinguish similar objects and do not denote a particular ordering for the objects.
The application provides a signal identification method based on deep learning, it is based on the rhythm of the heart discernment to the electrograph signal (ECG signal), the signal identification method based on deep learning that this application provided, carry out the preliminary treatment to the electrograph signal and obtain multiunit preliminary treatment information, input multiunit preliminary treatment information into corresponding a plurality of deep learning submodels, obtain multiunit initial depth characteristic, input multiunit initial depth characteristic into the second deep learning model, obtain the final recognition result based on the electrocardiograph signal, this application is based on the electrocardiogram signal, and combine first deep learning module and the second deep learning module that contain a plurality of deep learning submodels, from different angles more comprehensive discernment electrocardiogram signal, in order to improve the precision of recognition result.
The present embodiment provides a signal identification method based on deep learning, and fig. 1 is a flow chart i of a signal identification method based on deep learning according to an embodiment of the present application, as shown in fig. 1, the method includes the following steps:
step S101, acquiring an electrocardiogram signal; the electrocardiogram signal may be based on a single-lead electrocardiogram signal, and the electrocardiogram signal is a curve pattern of the variation of the electrical activity generated by each cardiac cycle of the heart recorded from the body surface by using an electrocardiograph.
Step S102, preprocessing the electrocardiogram signals to obtain a plurality of groups of preprocessing information; it should be noted that, before preprocessing the electrocardiograph signal to obtain a plurality of groups of preprocessed information, denoising the electrocardiograph signal may be performed, mainly to remove noise such as baseline drift and power frequency interference, so as to improve the authenticity of the plurality of groups of preprocessed information.
Step S103, inputting multiple groups of preprocessing information into a first deep learning model to obtain multiple groups of initial depth features, wherein the first deep learning model comprises multiple deep learning submodels, and the multiple groups of preprocessing information correspond to the multiple deep learning submodels;
it should be noted that, the deep learning technology is capable of automatically extracting features required by the deep learning technology from complex and multi-modal high-dimensional data with high efficiency, and the deep learning applied to electrocardiogram signal recognition can describe classification targets more comprehensively compared with the traditional machine learning, which is beneficial to improving the precision of a classifier.
Step S104, inputting multiple groups of initial depth features into a second deep learning model to obtain a final recognition result;
it should be noted that the multiple deep learning submodels include a multilayer neural network, so that it can be understood that the output of the multiple deep learning submodels is a high-level feature obtained by performing multilayer mapping on the low-level features of the electrocardiogram signal, and the actual high-level feature has a low dimension and contains rich heart rhythm classification information; for the second deep learning model for processing advanced features, a plurality of layers of fully connected layers are not required, that is, the second deep learning model fuses a plurality of groups of initial depth features, a plurality of layers of fully connected networks are used for screening, transforming and dimension reduction, and finally a softmax classifier is used for classifying the heart rhythm information to obtain a final recognition result, for example, whether the single-lead electrocardiogram signal is of an atrial fibrillation heart rhythm type or other heart rhythms can be judged according to the final recognition result.
Through the steps from S101 to S104, the electrocardiogram signals are preprocessed to obtain a plurality of groups of preprocessed information, the plurality of groups of preprocessed information are respectively input into corresponding deep learning submodels to obtain a plurality of groups of initial depth features based on the electrocardiogram signals, the plurality of groups of initial depth features are input into a second deep learning model, the second deep learning model fuses a plurality of groups of recognition results based on the electrocardiogram signals, a multilayer full-connection network is used for screening, transforming and dimension reduction, and a final recognition result based on the electrocardiogram signals is output.
In some embodiments, the present embodiment provides a signal identification method based on deep learning, and fig. 2 is a flowchart of a method for acquiring an initial depth feature corresponding to a medical feature according to an embodiment of the present application; as shown in fig. 2, the method comprises the steps of:
step S201, extracting medical characteristics according to electrocardiogram signals;
it should be noted that, since it is almost difficult to accurately locate the start point and the end point of the P/T wave (P wave refers to atrial depolarization wave and T wave refers to rapid ventricular repolarization period) in the single-lead electrocardiogram signal, in order to avoid the influence of the measurement error, the embodiments of the present application do not consider the characteristics easily affected by the measurement error, such as PR interval (from the start point of P wave to the start point of QRS complex, which represents the time when the atrium starts to depolarize, where QRS complex represents the point location change of ventricular muscle depolarization), ST slope (ST segment refers to the initial stage of ventricular repolarization), etc., and only select the parameters less affected by noise and measurement error, such as heart rate variability characteristics SDNN, QRS amplitude variability characteristics, etc.; in addition, the embodiment of the application does not use the characteristics with too complex calculation, so that the algorithm is prevented from being too complex due to various characteristic extraction algorithms, and the medical characteristics comprise the following aspects:
the morphological characteristics of the electrocardiogram signals are as follows: QRS wave width, QRS wave amplitude standard deviation, QRS wave amplitude variability index, R wave energy standard deviation and the like; wherein, QRS complex represents the point position change of ventricular muscle depolarization, and R wave is ventricular end diastole;
RR interval variability feature (RR interval refers to heart beat cycle) and HRV variability feature (HRV refers to heart rate variability): RR interval variability such as approximate entropy and sample entropy of RR intervals, delta RR intervals, etc.; HRV characteristics include SDNN (sinus beat to beat standard deviation), SDSD (an index in heart rate variability characteristics), etc., and normalized power in the frequency range of 0.04Hz, 0.04-0.15Hz, and 0.15-0.5 Hz;
including the statistical features of mean, median, variance, peak and variance values of RR intervals, estimates of RR interval probability density, and in addition, the number of peaks in RR intervals and Delta RR intervals, and the energy variation between RR peaks as features.
Step S202, inputting medical features into a first deep learning submodel corresponding to the medical features to obtain initial deep features corresponding to the medical features, wherein the first deep learning submodel comprises a plurality of layers of fully-connected feedforward neural networks, regularization terms are added to the fully-connected feedforward neural networks, and a learning rate attenuation strategy is adopted in the process of training the first deep learning submodel;
for a group where medical features in multiple groups of preprocessed information are located, the medical features serve as an input deep learning submodel, because the dimensionality of the input features is not high (a large number of features which are susceptible to noise and too complex are removed), a simple fully-connected feedforward neural network comprising several layers is only needed, and in order to prevent algorithm overfitting, L2 regularization items and Dropout are added to each layer of the network when a first deep learning submodel corresponding to the medical features is trained; considering accelerating the deep learning submodel, adding Batch Normalization (Batch standardization, abbreviated as BN) when training the first deep learning submodel; in order to prevent the learning rate from being too large, the learning rate is swung back and forth when the global optimal point is converged, so that the learning rate is required to be continuously reduced along with the number of training rounds, and a learning rate attenuation strategy is adopted in the process of training the first deep learning submodel.
Through steps S201 to S202, medical features less affected by noise and measurement errors are selected based on the electrocardiogram signal, on one hand, the output accuracy of the deep learning submodel is improved, and on the other hand, the complexity of an algorithm for extracting features is reduced.
In some embodiments, this embodiment provides a signal identification method based on deep learning, and fig. 3 is a flowchart of a method for obtaining an initial depth feature corresponding to a time-frequency spectrogram according to an embodiment of the present application; as shown in fig. 3, the method comprises the steps of:
step S301, performing wavelet transformation or short-time Fourier transformation on the electrocardiogram signal to obtain a time-frequency spectrogram;
it should be noted that, a time-frequency spectrogram can be obtained from an ECG signal according to wavelet transform or short-time fourier transform, and after the time-frequency spectrogram is obtained, the time-frequency spectrogram is converted into a pixel scale of 224 × 224 in the embodiment of the present application, and since the time-frequency spectrogram is a color picture, the size of data actually input to the corresponding deep learning sub-model is 224 × 224 × 3.
Step S302, inputting the time-frequency spectrogram into a second deep learning submodel corresponding to the time-frequency spectrogram to obtain initial depth features corresponding to the time-frequency spectrogram, wherein the second deep learning submodel adopts a migration learning mode based on a convolutional neural network, and the front half part of a CNN network trained by million-level picture information is used for constructing the second deep learning submodel;
for the group of time-frequency spectrograms in the multi-group preprocessing information, the time-frequency spectrogram is used as an input second deep learning submodel, and as the length of input data is 150528(224 multiplied by 3), the input data is too large and is not suitable for adopting a fully-connected network, and a CNN network model can be selected; because the cost required for reconstructing a large-scale network is very high, in the embodiment of the application, transfer learning is adopted, a time-frequency spectrogram in the embodiment of the application is used as an input second deep learning submodel by taking the first half part of a CNN network GoogleLeNet trained by million-level picture information, and a plurality of full connection layers and a softmax layer, and the 224 multiplied by 3 time-frequency spectrum data obtained by wavelet transformation is utilized for training and fine tuning the second deep learning submodel by electrocardiogram signals in an ECGAF classification database.
Among them, google lenet is a pre-trained model that has been trained on a subset of the ImageNet database that is used for the ImageNet large-scale visual recognition challenge race (ILSVRC). The model was trained over one million images, with 144 layers, and images could be classified into 1000 object classes (e.g., keyboard, mouse, pencil, and many animals).
In some embodiments, the present embodiment provides a signal identification method based on deep learning, and fig. 4 is a flowchart of a method for obtaining initial depth features corresponding to transient spectrum and spectrum entropy according to an embodiment of the present application; as shown in fig. 4, the method includes the steps of:
step S401, instantaneous frequency spectrum and frequency spectrum entropy analysis are carried out on the electrocardiogram signals to obtain instantaneous frequency spectrum and frequency spectrum entropy;
it should be noted that the instantaneous spectrum analysis and the spectrum entropy analysis result of a section of the electrocardiogram signal form a two-dimensional time sequence, and the instantaneous spectrum and the spectrum entropy can be determined by using instfreq (instantaneous spectrum) and pendopy (spectrum entropy) functions in a matlab kit, wherein the matlab kit is mathematical software used in the fields of data analysis, wireless communication, deep learning, image processing and computer vision, signal processing, quantitative finance and risk management, robots, control systems and the like.
It should be further noted that, for example, a length of 30s electrocardiogram data, in the case of a sampling rate of 300Hz (total 9000 sample points), the length of the obtained spectrum curve is 255 points (default settings of instfreq and pendorpy) through instantaneous spectrum analysis and spectrum entropy analysis, so that the input data of the deep learning submodel corresponding to the instantaneous spectrum and the spectrum entropy is 255 × 2, and compared with directly inputting the original data, the input data amount is equivalent to being compressed by 17.6 times, thereby reducing the complexity of data processing.
Step S402, inputting the instantaneous frequency spectrum and the frequency spectrum entropy into a corresponding third deep learning submodel to obtain a corresponding initial depth characteristic, wherein the third deep learning submodel adopts a bidirectional long-time and short-time memory network, the bidirectional long-time and short-time memory network is added with an attention strategy and a regularization item, and a learning rate attenuation strategy is adopted in the process of training the third deep learning submodel;
it should be noted that, for a group where an instantaneous spectrum and a spectrum entropy in a plurality of groups of preprocessed information are located, a third deep learning sub-model taking the instantaneous spectrum and the spectrum entropy as inputs, because the instantaneous spectrum and the spectrum entropy are time sequence signals and the scale of input data is small, a small-scale layer LSTM network + fully-connected feedforward neural network can achieve a relatively good classification effect, in the embodiment of the application, a bidirectional LSTM + Attention is used, and the third deep learning sub-model further comprises a plurality of layers of fully-connected layers and 1 softmax classification layer; compared with the prior art that one BilSTM layer and one softmax classification layer are used, the number of model layers is too small, the whole classification result is not ideal, and the deep learning submodel in the embodiment of the application obviously improves the heart rhythm type identification precision based on the electrocardiogram signals by enlarging the number of the layers of the network; in order to improve the accuracy of the algorithm, prevent overfitting and the like, L2 regularization, Dropout, BatchNormalization and the like are added, a learning rate attenuation strategy is adopted in the training process, and the Attention strategy is particularly added in an LSTM network.
In some embodiments, the present embodiment provides a signal identification method based on deep learning, fig. 5 is a flowchart of a method for acquiring initial depth features corresponding to truncated electrocardiogram sequences according to an embodiment of the present application; as shown in fig. 5, the method includes the steps of:
step S501, down-sampling processing is carried out on electrocardiogram signals to obtain down-sampled electrocardiogram sequences;
wherein, a section of electrocardiogram signal is directly used as LSTM network input to extract time sequence information and identify the heart rhythm, and the test precision of the algorithm is relatively low due to the problem of long-term dependence; in the embodiment of the present application, the sampling rate of the electrocardiogram signal is reduced, for example, from 300Hz to 60Hz, and the electrocardiogram signal is down-sampled, so that although more high-frequency information is lost, more useful information can be provided for the classification of the heart rhythm.
Step S502, the down-sampled electrocardiogram sequence is cut off to obtain a cut-off electrocardiogram sequence;
in the embodiment of the present application, the down-sampled ecg sequence is truncated into several smaller ecg sequences, for example, when one segment is 30s and the sampling rate is 300Hz (9000 points), 1800 points (the sampling rate is reduced to 60Hz) are down-sampled and then truncated into 4 segments, the final input is changed to a scale of 450 × 4, and although the down-sampling loses high-frequency information and truncates partial sequence information, the accuracy of the classification algorithm can be greatly improved.
Step S503, inputting the truncated electrocardiogram sequence into a corresponding fourth deep learning submodel to obtain corresponding initial depth features, wherein the fourth deep learning submodel adopts a bidirectional long-time and short-time memory network, the bidirectional long-time and short-time memory network is added with an attention strategy and a regularization item, and a learning rate attenuation strategy is adopted in the process of training the fourth deep learning submodel;
for a group where a truncated electrocardiogram sequence in a plurality of groups of preprocessed information is located, a fourth deep learning submodel taking the truncated electrocardiogram sequence as input is similar to the deep learning submodel taking an instantaneous spectrum and a spectrum entropy as input, a bidirectional LSTM + Attention is used, and the deep learning submodel further comprises a plurality of full connection layers and 1 softmax classification layer; similarly, for the problems of testing accuracy of the algorithm, preventing overfitting and the like, adding L2 regularization, Dropout and Batch Normalization, adopting a learning rate attenuation strategy in the training process and the like, particularly adding an Attention strategy in an LSTM network, and improving the accuracy of the initial depth feature.
In some embodiments, the present embodiment provides a signal identification method based on deep learning, and fig. 6 is a flowchart of a method for acquiring an initial depth feature corresponding to an original electrocardiogram sequence according to an embodiment of the present application; as shown in fig. 6, the method includes the steps of:
step S601, normalization processing is carried out on the electrocardiogram signals to obtain an electrocardiogram sequence after normalization processing.
Step S602, an electrocardiogram sequence is input into a corresponding fifth deep learning submodel to obtain corresponding initial depth features, wherein the fifth deep learning submodel adopts a cavity convolution and a convolution function to perform one-dimensional convolution filtering and one-dimensional convolution, a regularization item is added to the fifth deep learning submodel, and a learning rate attenuation strategy is adopted in the process of training the fifth deep learning submodel;
after a section of electrocardiogram signals can be directly subjected to Normalization preprocessing, a corresponding fifth deep learning submodel is input to be used for extracting one-dimensional Convolution characteristics of the electrocardiogram signals, the fifth deep learning submodel can use the Attaus Convolution1D (Attaus Convolution, namely Convolution with holes and sparse Convolution kernel) and the Convolution1D function of an open source artificial neural network library to carry out 1-dimensional Convolution filtering and 1-dimensional Convolution, a maximum pooling layer is used, in the embodiment of the application, a plurality of Convolution layers and a plurality of full connection layers are used in the fifth deep learning submodel, and similarly, in order to prevent overfitting, each layer is measured by Dropout, Batch Normalization, L2 Normalization and the like, a learning rate attenuation strategy is adopted in the training process.
In some embodiments, the present embodiment provides a signal identification method based on deep learning, and fig. 7 is a flowchart ii of the signal identification method based on deep learning according to the embodiment of the present application; as shown in fig. 7, for a multi-lead ecg signal, the method further comprises the steps of:
step S701, obtaining final recognition results of a plurality of single leads, and counting the number of the same final recognition results; fig. 8 is a recognition framework based on single-lead ecg signals according to an embodiment of the present application, and as shown in fig. 8, the single-lead ecg signals are preprocessed to obtain multiple sets of preprocessed information; and respectively inputting multiple groups of preprocessing information into corresponding deep learning submodels to obtain multiple groups of initial depth features, inputting the multiple groups of initial depth features into the full-connection layer, and processing the multiple groups of initial depth features through the Softmax classification layer to finally obtain a final recognition result of the single lead.
Step S702, using the final identification result with the largest quantity as the final identification result of the multi-lead electrocardiogram signals; fig. 9 is a recognition frame based on multi-lead ecg signals according to an embodiment of the present application, and as shown in fig. 9, for n single-lead ecg signals, the ecg signals with lead 1, 2, and 3 … … n are respectively recognized according to the recognition frame of fig. 8 to obtain final recognition results of a plurality of single leads, and finally, the number of the same final recognition results is counted, and the final recognition result with the largest number is used as the final recognition result of the ecg signals with multiple leads.
Through steps S701 to S702, the final recognition results of a plurality of single leads are obtained for the multi-lead electrocardiogram signals, the number of the same final recognition results is counted, and the final recognition result with the largest number is used as the final recognition result of the multi-lead electrocardiogram signals, so as to improve the accuracy of the final recognition result of the multi-lead electrocardiogram signals.
In some embodiments, in the training process of training the multiple deep learning submodels, because the multiple deep learning submodels are different models, in order to ensure that each deep learning submodel does not interfere with the training of other models in the training process, a softmax layer may be added to each deep learning submodel, each model is trained separately, heart rhythm recognition of atrial fibrillation type is completed, and finally the softmax classification layer is removed from the trained deep learning submodels.
It should be noted that, in the present application, the first deep learning model and the second deep learning model may be finally combined together to form a deep learning model, for example, in the training process, a softmax layer may be added to each deep learning sub-model, each deep learning sub-model is trained separately to complete the heart rhythm recognition of the atrial fibrillation type, the softmax classification layer is removed from each trained deep learning sub-model, the rest parts are combined together, a total deep learning model is formed through several full connection layers and one softmax layer, and finally, the training samples are used to fine tune the total deep learning model parameters and train the total deep learning model.
It should be noted that the steps illustrated in the above-described flow diagrams or in the flow diagrams of the figures may be performed in a computer system, such as a set of computer-executable instructions, and that, although a logical order is illustrated in the flow diagrams, in some cases, the steps illustrated or described may be performed in an order different than here.
The present embodiment further provides a signal identification apparatus based on deep learning, which is used to implement the foregoing embodiments and preferred embodiments, and the description of the apparatus is omitted here. As used hereinafter, the terms "module," "unit," "subunit," and the like may implement a combination of software and/or hardware for a predetermined function. Although the means described in the embodiments below are preferably implemented in software, an implementation in hardware, or a combination of software and hardware is also possible and contemplated.
Fig. 10 is a block diagram of a deep learning based signal identification apparatus according to an embodiment of the present application, and as shown in fig. 10, the apparatus includes: the learning system comprises an acquisition module 100, a preprocessing module 101, a first learning module 102 and a second learning module 103;
an acquisition module 100 for acquiring an electrocardiogram signal;
the preprocessing module 101 is used for preprocessing the electrocardiogram signals to obtain a plurality of groups of preprocessing information;
the first learning module 102 is configured to input multiple sets of preprocessing information into a first deep learning model to obtain multiple sets of initial deep features, where the first deep learning model includes multiple deep learning submodels, and the multiple sets of preprocessing information correspond to the multiple deep learning submodels;
and the second learning module 103 is configured to input multiple sets of initial depth features into the second deep learning model to obtain a final recognition result.
In some embodiments, the preprocessing module 101, the first learning module 102, and the second learning module 103 are further configured to implement steps in the signal identification method based on deep learning provided in the foregoing embodiments, and details are not repeated here.
The above modules may be functional modules or program modules, and may be implemented by software or hardware. For a module implemented by hardware, the modules may be located in the same processor; or the modules can be respectively positioned in different processors in any combination.
In one embodiment, a computer device is provided, which may be a terminal. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a method of signal recognition based on deep learning. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on the shell of the computer equipment, an external keyboard, a touch pad or a mouse and the like.
In an embodiment, fig. 11 is a schematic internal structural diagram of a computer device according to an embodiment of the present application, and as shown in fig. 11, a computer device is provided, where the computer device may be a server, and its internal structural diagram may be as shown in fig. 11. The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The database of the computer device is used for storing data. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a method of signal recognition based on deep learning.
Those skilled in the art will appreciate that the architecture shown in fig. 11 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, a computer device is provided, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the computer program to implement the steps of the deep learning based signal identification method provided by the foregoing embodiments.
In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and the computer program, when executed by a processor, implements the steps in the deep learning based signal identification method provided by the above embodiments.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
The technical features of the embodiments described above may be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the embodiments described above are not described, but should be considered as being within the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.
Claims (10)
1. A method for signal recognition based on deep learning, the method comprising:
acquiring an electrocardiogram signal;
preprocessing the electrocardiogram signals to obtain a plurality of groups of preprocessing information;
inputting a plurality of groups of preprocessing information into a first deep learning model to obtain a plurality of groups of initial deep features, wherein the first deep learning model comprises a plurality of deep learning submodels, and the plurality of groups of preprocessing information correspond to the plurality of deep learning submodels;
and inputting the plurality of groups of initial depth features into a second deep learning model to obtain a final recognition result.
2. The signal identification method of claim 1, wherein the electrocardiogram signals are preprocessed to obtain a plurality of groups of preprocessed information; inputting a plurality of groups of the preprocessing information into a first deep learning model, and obtaining a plurality of groups of initial depth features comprises:
extracting medical features from the electrocardiogram signals;
inputting the medical features into a first deep learning submodel corresponding to the medical features to obtain initial deep features corresponding to the medical features, wherein the first deep learning submodel comprises a plurality of layers of fully-connected feedforward neural networks, regularization terms are added to the fully-connected feedforward neural networks, and a learning rate attenuation strategy is adopted in the process of training the first deep learning submodel.
3. The signal identification method of claim 1, wherein the electrocardiogram signals are preprocessed to obtain a plurality of groups of preprocessed information; inputting a plurality of groups of the preprocessing information into a first deep learning model, and obtaining a plurality of groups of initial depth features comprises:
performing wavelet transform or short-time Fourier transform on the electrocardiogram signal to obtain a time-frequency spectrogram;
and inputting the time-frequency spectrogram into a second deep learning submodel corresponding to the time-frequency spectrogram to obtain initial depth features corresponding to the time-frequency spectrogram, wherein the second deep learning submodel adopts a migration learning mode based on a convolutional neural network and adds a classification layer in the trained convolutional neural network.
4. The signal identification method of claim 1, wherein the electrocardiogram signals are preprocessed to obtain a plurality of groups of preprocessed information; inputting a plurality of groups of the preprocessing information into a first deep learning model, and obtaining a plurality of groups of initial depth features comprises:
carrying out instantaneous frequency spectrum and frequency spectrum entropy analysis on the electrocardiogram signal to obtain an instantaneous frequency spectrum and a frequency spectrum entropy;
inputting the instantaneous frequency spectrum and the frequency spectrum entropy into a corresponding third deep learning submodel to obtain corresponding initial depth features, wherein the third deep learning submodel adopts a bidirectional long-time and short-time memory network, the bidirectional long-time and short-time memory network is added with an attention strategy and a regularization item, and a learning rate attenuation strategy is adopted in the process of training the third deep learning submodel.
5. The signal identification method of claim 1, wherein the electrocardiogram signals are preprocessed to obtain a plurality of groups of preprocessed information; inputting a plurality of groups of the preprocessing information into a first deep learning model, and obtaining a plurality of groups of initial depth features comprises:
performing down-sampling processing on the electrocardiogram signals to obtain down-sampled electrocardiogram sequences;
cutting the down-sampled electrocardiogram sequence to obtain a cut-off electrocardiogram sequence;
and inputting the truncated electrocardiogram sequence into a corresponding fourth deep learning submodel to obtain corresponding initial depth features, wherein the fourth deep learning submodel adopts a bidirectional long-time and short-time memory network, the bidirectional long-time and short-time memory network is added with an attention strategy and a regularization item, and a learning rate attenuation strategy is adopted in the process of training the fourth deep learning submodel.
6. The signal identification method of claim 1, wherein the electrocardiogram signals are preprocessed to obtain a plurality of groups of preprocessed information; inputting a plurality of groups of the preprocessing information into a first deep learning model, and obtaining a plurality of groups of initial depth features comprises:
carrying out normalization processing on the electrocardiogram signals to obtain an electrocardiogram sequence after normalization processing;
inputting the electrocardiogram sequence into a corresponding fifth deep learning submodel to obtain corresponding initial depth features, wherein the fifth deep learning submodel adopts a cavity convolution and a convolution function to perform one-dimensional convolution filtering and one-dimensional convolution, a regularization item is added to the fifth deep learning submodel, and a learning rate attenuation strategy is adopted in the process of training the fifth deep learning submodel.
7. The signal recognition method of claim 1, wherein for a multi-lead electrocardiogram signal, the method further comprises:
acquiring final recognition results of a plurality of single leads, and counting the number of the same final recognition results;
and taking the final recognition result with the largest number as the final recognition result of the multi-lead electrocardiogram signals.
8. An apparatus for signal recognition based on deep learning, the apparatus comprising: the device comprises an acquisition module, a preprocessing module, a first learning module and a second learning module;
the acquisition module is used for acquiring electrocardiogram signals;
the preprocessing module is used for preprocessing the electrocardiogram signals to obtain a plurality of groups of preprocessing information;
the first learning module is used for inputting a plurality of groups of preprocessing information into a first deep learning model to obtain a plurality of groups of initial deep features, wherein the first deep learning model comprises a plurality of deep learning submodels, and the plurality of groups of preprocessing information correspond to the plurality of deep learning submodels;
and the second learning module is used for inputting the plurality of groups of initial depth features into a second deep learning model to obtain a final recognition result.
9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the deep learning based signal recognition method according to any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, implements the deep learning based signal recognition method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011000576.XA CN112120691A (en) | 2020-09-22 | 2020-09-22 | Signal identification method and device based on deep learning and computer equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011000576.XA CN112120691A (en) | 2020-09-22 | 2020-09-22 | Signal identification method and device based on deep learning and computer equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112120691A true CN112120691A (en) | 2020-12-25 |
Family
ID=73842165
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011000576.XA Pending CN112120691A (en) | 2020-09-22 | 2020-09-22 | Signal identification method and device based on deep learning and computer equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112120691A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112971798A (en) * | 2021-02-10 | 2021-06-18 | 武汉心络科技有限公司 | Electrocardiosignal processing method, electrocardiosignal processing device, electrocardiosignal processing terminal and storage medium |
| CN113116300A (en) * | 2021-03-12 | 2021-07-16 | 复旦大学 | Physiological signal classification method based on model fusion |
| CN113349796A (en) * | 2021-06-16 | 2021-09-07 | 珠海中科先进技术研究院有限公司 | Sleep monitoring method and device based on multi-source signals and storage medium |
| CN113729709A (en) * | 2021-09-23 | 2021-12-03 | 中国科学技术大学先进技术研究院 | Neurofeedback apparatus, neurofeedback method, and computer-readable storage medium |
| KR20230047613A (en) * | 2021-10-01 | 2023-04-10 | 조선대학교산학협력단 | Personal identification method using bio signal |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106214123A (en) * | 2016-07-20 | 2016-12-14 | 杨平 | A kind of electrocardiogram compressive classification method based on degree of depth learning algorithm |
| CN106901723A (en) * | 2017-04-20 | 2017-06-30 | 济南浪潮高新科技投资发展有限公司 | A kind of electrocardiographic abnormality automatic diagnosis method |
| CN108186011A (en) * | 2017-12-13 | 2018-06-22 | 深圳竹信科技有限公司 | Atrial fibrillation detection method, device and readable storage medium storing program for executing |
| CN110584649A (en) * | 2019-09-09 | 2019-12-20 | 武汉大学 | Method and system for identifying abnormal electrocardiosignals in dynamic single-lead electrocardiogram |
| CN110840402A (en) * | 2019-11-19 | 2020-02-28 | 山东大学 | Atrial fibrillation signal identification method and system based on machine learning |
| CN111053552A (en) * | 2019-12-03 | 2020-04-24 | 上海理工大学 | QRS wave detection method based on deep learning |
| CN111329469A (en) * | 2020-03-05 | 2020-06-26 | 广州天嵌计算机科技有限公司 | Arrhythmia prediction method |
| CN111557660A (en) * | 2020-06-08 | 2020-08-21 | 东北大学 | Arrhythmia identification method under sub-population deep learning framework |
-
2020
- 2020-09-22 CN CN202011000576.XA patent/CN112120691A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106214123A (en) * | 2016-07-20 | 2016-12-14 | 杨平 | A kind of electrocardiogram compressive classification method based on degree of depth learning algorithm |
| CN106901723A (en) * | 2017-04-20 | 2017-06-30 | 济南浪潮高新科技投资发展有限公司 | A kind of electrocardiographic abnormality automatic diagnosis method |
| CN108186011A (en) * | 2017-12-13 | 2018-06-22 | 深圳竹信科技有限公司 | Atrial fibrillation detection method, device and readable storage medium storing program for executing |
| CN110584649A (en) * | 2019-09-09 | 2019-12-20 | 武汉大学 | Method and system for identifying abnormal electrocardiosignals in dynamic single-lead electrocardiogram |
| CN110840402A (en) * | 2019-11-19 | 2020-02-28 | 山东大学 | Atrial fibrillation signal identification method and system based on machine learning |
| CN111053552A (en) * | 2019-12-03 | 2020-04-24 | 上海理工大学 | QRS wave detection method based on deep learning |
| CN111329469A (en) * | 2020-03-05 | 2020-06-26 | 广州天嵌计算机科技有限公司 | Arrhythmia prediction method |
| CN111557660A (en) * | 2020-06-08 | 2020-08-21 | 东北大学 | Arrhythmia identification method under sub-population deep learning framework |
Non-Patent Citations (1)
| Title |
|---|
| 方圆圆编: "《人脸识别与美颜算法实战 基于Python、机器学习与深度学习》", 30 April 2020, 机械工业出版社 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112971798A (en) * | 2021-02-10 | 2021-06-18 | 武汉心络科技有限公司 | Electrocardiosignal processing method, electrocardiosignal processing device, electrocardiosignal processing terminal and storage medium |
| CN113116300A (en) * | 2021-03-12 | 2021-07-16 | 复旦大学 | Physiological signal classification method based on model fusion |
| CN113349796A (en) * | 2021-06-16 | 2021-09-07 | 珠海中科先进技术研究院有限公司 | Sleep monitoring method and device based on multi-source signals and storage medium |
| CN113349796B (en) * | 2021-06-16 | 2024-05-10 | 珠海中科先进技术研究院有限公司 | Sleep monitoring method, equipment and storage medium based on multi-source signals |
| CN113729709A (en) * | 2021-09-23 | 2021-12-03 | 中国科学技术大学先进技术研究院 | Neurofeedback apparatus, neurofeedback method, and computer-readable storage medium |
| CN113729709B (en) * | 2021-09-23 | 2023-08-11 | 中科效隆(深圳)科技有限公司 | Nerve feedback device, nerve feedback method, and computer-readable storage medium |
| KR20230047613A (en) * | 2021-10-01 | 2023-04-10 | 조선대학교산학협력단 | Personal identification method using bio signal |
| KR102679575B1 (en) * | 2021-10-01 | 2024-06-27 | 조선대학교산학협력단 | Personal identification method using bio signal |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | Patient-specific ECG classification by deeper CNN from generic to dedicated | |
| CN112120691A (en) | Signal identification method and device based on deep learning and computer equipment | |
| Wang et al. | Arrhythmia classification algorithm based on multi-head self-attention mechanism | |
| Oh et al. | Automated diagnosis of arrhythmia using combination of CNN and LSTM techniques with variable length heart beats | |
| Zhao et al. | ECG authentication system design incorporating a convolutional neural network and generalized S-Transformation | |
| Uwaechia et al. | A comprehensive survey on ECG signals as new biometric modality for human authentication: Recent advances and future challenges | |
| Srivastva et al. | PlexNet: A fast and robust ECG biometric system for human recognition | |
| CN107516075B (en) | Electrocardiosignal detection method and device and electronic equipment | |
| Camara et al. | Real-time electrocardiogram streams for continuous authentication | |
| CN111209885A (en) | Gesture information processing method and device, electronic equipment and storage medium | |
| Sepahvand et al. | A novel multi-lead ECG personal recognition based on signals functional and structural dependencies using time-frequency representation and evolutionary morphological CNN | |
| CN110309758B (en) | Electrocardiosignal feature extraction method and device, computer equipment and storage medium | |
| CN111657925A (en) | Electrocardiosignal classification method, system, terminal and storage medium based on machine learning | |
| CN109124620B (en) | Atrial fibrillation detection method, device and equipment | |
| Fatimah et al. | Biometric identification from ECG signals using Fourier decomposition and machine learning | |
| Li et al. | Signal processing and feature selection preprocessing for classification in noisy healthcare data | |
| CN109497986A (en) | Electrocardiogram intelligent analysis method, device, computer equipment and system based on deep neural network | |
| Ullah et al. | An End‐to‐End Cardiac Arrhythmia Recognition Method with an Effective DenseNet Model on Imbalanced Datasets Using ECG Signal | |
| El Boujnouni et al. | A wavelet-based capsule neural network for ECG biometric identification | |
| Ullah et al. | [Retracted] An Effective and Lightweight Deep Electrocardiography Arrhythmia Recognition Model Using Novel Special and Native Structural Regularization Techniques on Cardiac Signal | |
| Tseng et al. | Sliding large kernel of deep learning algorithm for mobile electrocardiogram diagnosis | |
| Ukil et al. | AFSense-ECG: Atrial fibrillation condition sensing from single lead electrocardiogram (ECG) signals | |
| Huang et al. | Robust multi-feature collective non-negative matrix factorization for ECG biometrics | |
| CN113317804A (en) | Rehabilitation efficiency prediction method, training device and computer equipment | |
| Rahuja et al. | A deep neural network approach to automatic multi-class classification of electrocardiogram signals |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20201225 |