CN112472107B - Electroencephalogram artifact removing method and device - Google Patents
Electroencephalogram artifact removing method and device Download PDFInfo
- Publication number
- CN112472107B CN112472107B CN202011449582.3A CN202011449582A CN112472107B CN 112472107 B CN112472107 B CN 112472107B CN 202011449582 A CN202011449582 A CN 202011449582A CN 112472107 B CN112472107 B CN 112472107B
- Authority
- CN
- China
- Prior art keywords
- artifact
- signal
- channel
- network
- electroencephalogram
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- 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
- A61B5/7207—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal of noise induced by motion artifacts
-
- 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)
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Physics & Mathematics (AREA)
- Signal Processing (AREA)
- Biophysics (AREA)
- Animal Behavior & Ethology (AREA)
- Physiology (AREA)
- Psychiatry (AREA)
- Veterinary Medicine (AREA)
- Public Health (AREA)
- General Health & Medical Sciences (AREA)
- Pathology (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention provides an electroencephalogram artifact removing method and device, which are used for removing artifacts from a single-channel electroencephalogram signal by using an artifact removing model obtained by pre-training without additionally adding a reference channel, can remove the artifacts from the single-channel electroencephalogram signal acquired by a single electrode, and reduce the requirement on acquisition equipment. Moreover, the artifact removal model is obtained by training the deep learning network model by utilizing a training set comprising a plurality of segmented signal sets of single-channel electroencephalograms containing at least one artifact in advance, and can be used for removing the artifact from electroencephalograms containing a plurality of different artifacts, so that the artifact removal effect is improved.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to an electroencephalogram artifact removing method and device.
Background
An Electroencephalogram (EEG) technology is a widely used Electroencephalogram physiological signal detection technology, and can be applied to diagnosis of neuropathology, as well as fields of cognitive science research, health monitoring, driving fatigue detection and the like.
Because the electroencephalogram is obtained by attaching the electrodes to the head of a person and amplifying and recording collected electrical signals by a machine, the collected electrical signals are very weak and may be contaminated by various human bodies and externally introduced noise, resulting in artifacts in the collected electroencephalogram.
There is currently a blind source separation algorithm based artifact removal method that can use the best reference electrode placement for artifact removal by selecting the best reference electrode for each set of sources in a series of different reference electrode and recording electrode combinations. However, this method needs to use multi-channel signal acquisition, and has high requirements on acquisition equipment.
Disclosure of Invention
In view of this, the present invention provides an electroencephalogram artifact removing method and apparatus, which implement artifact removal based on a single-channel electroencephalogram signal.
In order to achieve the above purpose, the invention provides the following specific technical scheme:
an electroencephalogram artifact removal method, comprising:
acquiring a segmented signal set of a single-channel electroencephalogram to be processed, wherein the segmented signal set of the single-channel electroencephalogram to be processed comprises a single-channel electroencephalogram signal segment containing an artifact;
extracting a feature vector set of a segmented signal set of the single-channel electroencephalogram to be processed;
inputting the feature vector set into an artifact removal model for artifact removal processing to obtain a segmented signal set of a single-channel electroencephalogram with artifacts removed, wherein the artifact removal model is obtained by training a deep learning network model in advance by using a training set comprising a plurality of segmented signal sets of single-channel electroencephalograms with at least one artifact, the deep learning network model corresponding to the artifact removal model comprises an attention network and a decoder network, and the decoder network comprises an artifact feature removal neural network and a waveform reduction mapping network;
and performing signal reconstruction on the segmented signal set of the artifact-removed single-channel electroencephalogram to obtain the artifact-removed single-channel electroencephalogram signal.
Optionally, the acquiring a segmented signal set of a single-channel electroencephalogram to be processed includes:
acquiring a segmented signal set of a standard single-channel electroencephalogram;
artifact detection is performed on the segmented signal set of the standard single-channel electroencephalogram using an artifact detection model, the artifact detection model is obtained by training a deep learning network model in advance by using a training set comprising a plurality of segmented signal sets of single-channel electroencephalograms containing at least one artifact, the deep learning network model corresponding to the artifact detection model comprises a dimension transformation mapping network, a classification feature extraction GRU network, a feature concentration network and an evaluation network, the dimension transformation mapping network is used for transforming the segmented signal set of the standard single-channel electroencephalogram into a signal vector set with preset dimensions, the classification characteristic extraction network is used for extracting the characteristics of the signal vector set to obtain a characteristic vector set, the feature vector set is weighted by the feature set network, and artifact detection is performed on each feature vector in the feature vector set after weighting by the evaluation network;
and in the case that the signal segment containing the artifact is detected, determining all the signal segments containing the artifact as a segmented signal set of the single-channel electroencephalogram to be processed.
Optionally, the acquiring a segmented signal set of a standard single-channel electroencephalogram includes:
and preprocessing the original single-channel electroencephalogram signal to obtain a segmented signal set of the standard single-channel electroencephalogram, wherein the preprocessing comprises resampling, signal segmentation and standardization processing.
Optionally, the extracting a feature vector set of the segmented signal set of the single-channel electroencephalogram to be processed includes:
inputting the segmented signal set of the single-channel electroencephalogram to be processed into a preset encoder, wherein the preset encoder comprises a dimension transformation mapping network and a noise feature extraction GRU network;
converting the segmented signal set of the single-channel electroencephalogram to be processed into a vector set with preset dimensions by using the dimension conversion mapping network;
and utilizing the noise feature extraction GRU network to encode the vector set to obtain a deep feature vector set.
Optionally, the method further includes:
acquiring a single-channel simulation signal, wherein the single-channel simulation signal comprises: simulating an electroencephalogram signal, simulating a muscle motion artifact, and simulating an eye motion artifact;
acquiring a single-channel semi-simulation signal, wherein the single-channel semi-simulation signal comprises: artificially removing an electroencephalogram signal of an artifact and an electromyographic motion artifact and an eye motion artifact which are separately collected;
acquiring a single-channel electroencephalogram original signal;
preprocessing the single-channel simulation signal and the single-channel semi-simulation signal to obtain the training set, and preprocessing the single-channel electroencephalogram original signal to obtain a verification set;
training a deep learning network model corresponding to the artifact detection model by using the training set, and verifying the trained model by using the verification set to obtain the artifact detection model;
and training a deep learning network model corresponding to the artifact removal model by using the training set, and verifying the trained model by using the verification set to obtain the artifact removal model.
An electroencephalogram artifact removal device, comprising:
the signal acquisition unit is used for acquiring a segmented signal set of the single-channel electroencephalogram to be processed, wherein the segmented signal set of the single-channel electroencephalogram to be processed comprises single-channel electroencephalogram signal segments containing artifacts;
the feature extraction unit is used for extracting a feature vector set of the segmented signal set of the single-channel electroencephalogram to be processed;
the artifact removing unit is used for inputting the feature vector set into an artifact removing model for artifact removing processing to obtain a segmented signal set of a single-channel electroencephalogram with artifacts removed, the artifact removing model is obtained by utilizing a training set comprising a plurality of segmented signal sets of single-channel electroencephalograms containing at least one artifact to train a deep learning network model in advance, the deep learning network model corresponding to the artifact removing model comprises an attention network and a decoder network, and the decoder network comprises an artifact feature removing neural network and a waveform reduction mapping network;
and the signal reconstruction unit is used for performing signal reconstruction on the segmented signal set of the artifact-removed single-channel electroencephalogram to obtain the artifact-removed single-channel electroencephalogram signal.
Optionally, the signal obtaining unit includes:
the signal preprocessing unit is used for acquiring a segmented signal set of a standard single-channel electroencephalogram;
the artifact detection unit is used for utilizing an artifact detection model to perform artifact detection on the segmented signal set of the standard single-channel electroencephalogram, and determining all signal segments containing artifacts as the segmented signal set of the single-channel electroencephalogram to be processed under the condition that the signal segments containing the artifacts are detected; the artifact detection model is obtained by training a deep learning network model in advance by using a training set comprising a plurality of segmented signal sets of single-channel electroencephalograms containing at least one artifact, the deep learning network model corresponding to the artifact detection model comprises a dimension transformation mapping network, a classification feature extraction GRU network, a feature concentration network and an evaluation network, the dimension transformation mapping network is used for transforming the segmented signal sets of the standard single-channel electroencephalograms into signal vector sets with preset dimensions, the classification feature extraction network is used for carrying out feature extraction on the signal vector sets to obtain feature vector sets, the feature concentration network is used for carrying out weighting operation on the feature vector sets, and the evaluation network is used for carrying out artifact detection on each feature vector in the feature vector sets after weighting operation.
Optionally, the signal preprocessing unit is specifically configured to:
and preprocessing the original single-channel electroencephalogram signal to obtain a segmented signal set of the standard single-channel electroencephalogram, wherein the preprocessing comprises resampling, signal segmentation and standardization processing.
Optionally, the feature extraction unit is specifically configured to:
inputting the segmented signal set of the single-channel electroencephalogram to be processed into a preset encoder, wherein the preset encoder comprises a dimension transformation mapping network and a noise feature extraction GRU network;
converting the segmented signal set of the single-channel electroencephalogram to be processed into a vector set with preset dimensions by using the dimension conversion mapping network;
and utilizing the noise feature extraction GRU network to encode the vector set to obtain a deep feature vector set.
Optionally, the apparatus further comprises a signal acquisition unit and a model training unit;
the signal acquisition unit is used for acquiring a single-channel simulation signal, a single-channel semi-simulation signal and a single-channel electroencephalogram original signal, wherein the single-channel simulation signal comprises: simulating an electroencephalogram signal, simulating a muscle motion artifact, and simulating an eye motion artifact, the single-channel semi-simulated signal comprising: artificially removing an electroencephalogram signal of an artifact and an electromyographic motion artifact and an eye motion artifact which are separately collected;
the signal preprocessing unit is further configured to preprocess the single-channel simulation signal and the single-channel semi-simulation signal to obtain the training set, and preprocess the single-channel electroencephalogram original signal to obtain a verification set;
the model training unit is used for training a deep learning network model corresponding to the artifact detection model by using the training set, and verifying the trained model by using the verification set to obtain the artifact detection model; and training a deep learning network model corresponding to the artifact removal model by using the training set, and verifying the trained model by using the verification set to obtain the artifact removal model.
Compared with the prior art, the invention has the following beneficial effects:
according to the electroencephalogram artifact removing method disclosed by the invention, the artifact removing model obtained by pre-training is utilized to perform artifact removing processing on the single-channel electroencephalogram signal, no additional reference channel is needed, the artifact removing can be realized on the single-channel electroencephalogram signal acquired by a single electrode, and the requirement on acquisition equipment is reduced. Moreover, the artifact removal model is obtained by training the deep learning network model by utilizing a training set comprising a plurality of segmented signal sets of single-channel electroencephalograms containing at least one artifact in advance, and can be used for removing the artifact from electroencephalograms containing a plurality of different artifacts, so that the artifact removal effect is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
Fig. 1 is a flowchart illustrating an electroencephalogram artifact removal method according to an embodiment of the present invention;
fig. 2 is a schematic flowchart of artifact detection by using an artifact detection model according to an embodiment of the present invention;
FIG. 3 is a schematic flow chart illustrating the preprocessing of an original single-channel electroencephalogram signal according to the embodiment of the present invention;
FIG. 4 is a schematic flow chart illustrating feature extraction of a segmented signal set of a single-channel electroencephalogram to be processed according to an embodiment of the present invention;
FIG. 5 is a schematic flowchart illustrating artifact removal using an artifact removal model according to an embodiment of the present invention;
fig. 6 is a schematic structural diagram of an electroencephalogram artifact removing device according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The inventor of the invention discovers through research that: the traditional artifact removing method such as a blind source separation algorithm, adaptive filtering and the like has the defects of a large number of required reference channels and complex acquisition equipment. The method is based on the deep learning network, realizes effective removal of artifacts in the single-channel electroencephalogram signal EEG, can remove various artifacts, and has strong practicability. Compared with the traditional filtering algorithm, the method based on learning the artifact characteristics of the measured signal can better inhibit the full-band noise and has stronger artifact removing effect.
Specifically, referring to fig. 1, the electroencephalogram artifact removing method disclosed in this embodiment includes the following steps:
s101: acquiring a segmented signal set of a single-channel electroencephalogram to be processed, wherein the segmented signal set of the single-channel electroencephalogram to be processed comprises single-channel electroencephalogram signal segments containing artifacts.
In order to improve the electroencephalogram artifact removal efficiency and avoid performing artifact removal processing on electroencephalogram signals not containing artifacts, in this embodiment, after a segmented signal set of a standard single-channel electroencephalogram is acquired, artifact detection is performed on the segmented signal set of the standard single-channel electroencephalogram by using an artifact detection model, and when a signal segment containing an artifact is detected, all signal segments containing the artifact are determined as the segmented signal set of the single-channel electroencephalogram to be processed, so that subsequent artifact removal processing is performed.
The artifact detection model is obtained by training a deep learning network model in advance by using a training set comprising a plurality of segmented signal sets of single-channel electroencephalograms containing at least one artifact, the process of detecting the artifact by using the artifact detection model is shown in fig. 2, and the deep learning network model corresponding to the artifact detection model is composed of three layers of bidirectional gated recurrent neural networks (GRUs) and comprises a dimension transformation mapping network, a classification feature extraction GRU network, a feature concentration network and an evaluation network.
The method comprises the steps that a segmented signal set of a standard single-channel electroencephalogram is input into an artifact detection model, a dimension transformation mapping network is used for transforming the segmented signal set of the standard single-channel electroencephalogram into a signal vector set with preset dimensions so as to meet the input requirement of a classification feature extraction GRU network, the classification feature extraction GRU network is used for carrying out feature extraction on the signal vector set to obtain a feature vector set, a feature concentration network is used for carrying out weighting operation on the feature vector set, an evaluation network is used for carrying out artifact detection on each feature vector in the feature vector set after the weighting operation, and the detection result is that the artifact is contained or the artifact is not contained.
It should be further noted that the segmented signal set of the standard single-channel electroencephalogram for artifact detection is obtained by preprocessing an original single-channel electroencephalogram signal, and the segmented signal set of the standard single-channel electroencephalogram obtained after preprocessing is a set of segmented signals with a fixed amplitude range, a fixed sampling frequency and a fixed segmentation interval, so that the segmented signal set meets the input requirement of an artifact detection model.
Referring to fig. 3, the preprocessing of the original single-channel electroencephalogram signal includes resampling, signal segmentation, and normalization. The EEG signals acquired by different devices vary in frequency from a few hundred to over kilohertz, the sampling frequency is determined at a fixed sampling frequency by resampling, and the signal is divided into signal segments for specific time intervals. The length of the segmentation is determined by the system delay and the amount of computational resources allowed in the specific application of the model. The longer the segmentation, the better the artifact removal effect, and the correspondingly longer the computation time. The expression for signal normalization is as follows:
EEG hereinrefThe EEG signal is an EEG signal which is acquired in advance and has no obvious artifact, the value of the normalized EEG signal is between 0 and 1, and sigma is a parameter for adjusting the amplitude range of the finally output EEG signal. After normalization, the output EEG signal cancels out the effects of amplitude and offset in the original signal.
S102: and extracting a feature vector set of the segmented signal set of the single-channel electroencephalogram to be processed.
The feature extraction is to perform deep feature extraction on the EEG signal containing the artifact, so as to be used for separating EEG signal and artifact signal features in the next step. The feature extraction is specifically implemented by an encoder, and referring to fig. 4, the encoder is composed of two layers of bidirectional GRU neural networks, including a dimension transformation mapping network and a noise feature extraction GRU network.
Specifically, a segmented signal set x of the single-channel electroencephalogram to be processed is set as { x ═ xiI 1, 2., n }, which is input to a dimension transformation mapping network and mapped to a vector of fixed dimensions. And inputting the vector set obtained after mapping into a noise feature extraction GRU network, and coding the mapped vector set by the GRU network to obtain deep feature vectors to form a feature vector set for removing the artifact features in the next step. The expression of the whole process is as follows:
{hi|i＝1，，2，...m}＝E(x；θe)
where h isiExtracting a coding mapping function, θ, of the GRU network for the extracted feature vector, E for the noise featureeParameters of the GRU network for encoding are extracted for the features.
S103: and inputting the feature vector set into an artifact removal model for artifact removal to obtain a segmented signal set of the artifact-removed single-channel electroencephalogram, wherein the artifact removal model is obtained by training a deep learning network model by utilizing a training set comprising a plurality of segmented signal sets of the single-channel electroencephalogram containing at least one artifact in advance.
Referring to fig. 5, the deep learning network model includes an attention network and a decoder network, and the decoder network includes an artifact characteristic removing neural network and a waveform reduction mapping network, where the artifact characteristic removing neural network may be a GRU network or an LSTM (Long Short-Term Memory, Long Short-Term Memory artificial neural network), and is not limited herein.
And (3) carrying out attention network feature weighting, artifact feature removal and waveform restoration on a feature vector set comprising EEG signal features not containing artifacts and artifact signal features to obtain a segmented signal set of the artifact-removed single-channel electroencephalogram.
The attention network is composed of full connection layers, the feature vectors are input into the attention network to be subjected to weighting operation, the feature vectors containing main information are emphasized, and finally the weighted feature vectors are obtained. The core of the decoder network is a two-layer unidirectional GRU network, which comprises an artifact characteristic removing GRU network and a waveform restoring mapping network. The attention network inputs the weighted feature vectors into a trained artifact feature removal GRU network, the GRU network removes the features of the artifact, and the artifact-free feature vectors are input into a trained waveform reduction mapping network. And the waveform reduction mapping network maps the feature vectors into the EEG signals without artifacts and outputs the EEG signals. The overall process of artifact removal recovery EEG signals is represented as follows:
where R is a function representing the association of the attention network and the decoding network, θaAs a parameter of the attention network, θdAre parameters of the decoder network. The EEG signal after artifact removal was:
s104: and performing signal reconstruction on the segmented signal set of the artifact-removed single-channel electroencephalogram to obtain the artifact-removed single-channel electroencephalogram signal.
Splicing the segmented signal set of the single-channel electroencephalogram without the artifact after the artifact removal model removes the artifact and the segmented signal of the single-channel electroencephalogram without the artifact after the artifact detection into a complete time sequence signal again, and restoring the signal set into an actual signal. In practical applications, the weak EEG signals also need to be amplified and visually displayed. Firstly, restoring the segmented signals to original amplitude according to a standardized inverse process, then splicing the segmented signals according to a time sequence according to segmented time intervals, and finally outputting the segmented signals to corresponding subsequent processing equipment after back-end processing.
It can be understood that the artifact detection model and the artifact removal model are obtained by training different deep learning network models in advance by using a training set.
Firstly, a training set is required to be obtained, and since a large number of original electroencephalogram signals are difficult to obtain in reality, the training set is obtained by preprocessing the single-channel simulation signal and the single-channel semi-simulation signal by obtaining the single-channel simulation signal and the single-channel semi-simulation signal.
And acquiring a single-channel electroencephalogram original signal, and preprocessing the single-channel electroencephalogram original signal to obtain a verification set.
Wherein, the single-channel simulation signal: the method comprises the steps of simulating an electroencephalogram signal, a muscle movement artifact and an eye movement artifact, wherein the simulated electroencephalogram signal is simulated by pink noise, the simulated muscle movement artifact is simulated by random noise of 20-60 Hz, the eye movement artifact is simulated by a series of sine waves Ai，ωi，φiAmplitude, frequency and phase, respectively. The signal-to-noise ratio of the generated signal is controlled by the ratio of the three signals. The single-channel simulation signal was generated as follows:
the single-channel semi-simulation signal comprises: artificially removing the artifact electroencephalogram signal and separately acquired electromyographic motion artifact and eye motion artifact. The eye motion artifact includes Vertical (VEOG) and Horizontal (HEOG) portions. The resulting signal to noise ratio is controlled by the ratio of the individual signals. The single-channel semi-simulation signal is generated as follows:
EEGcon＝EEGpure+aVEOG+bHEOG+cEMG
the single-channel electroencephalogram original signal is an actual signal obtained by acquisition equipment. The actual signal acquisition environment may include a hospital, a mobile vehicle, outdoors, etc. The actual signal may include a variety of artifacts including non-human production artifacts, muscle artifacts, chewing artifacts, vertical eye movement artifacts, and lateral eye movement artifacts.
And after the training set is obtained, training a deep learning network model corresponding to the artifact detection model by using the training set, and verifying the trained model by using the verification set to obtain the artifact detection model. And training a deep learning network model corresponding to the artifact removal model by using a training set, and verifying the trained model by using a verification set to obtain the artifact removal model.
According to the electroencephalogram artifact removing method disclosed by the embodiment, the artifact removing model obtained through pre-training is used for carrying out artifact removing processing on the single-channel electroencephalogram signal, no additional reference channel is needed, artifact removing can be achieved on the single-channel electroencephalogram signal acquired by the single electrode, and the requirement for acquisition equipment is lowered. Moreover, the artifact removal model is obtained by training the deep learning network model by utilizing a training set comprising a plurality of segmented signal sets of single-channel electroencephalograms containing at least one artifact in advance, and can be used for removing the artifact from electroencephalograms containing a plurality of different artifacts, so that the artifact removal effect is improved.
Based on the electroencephalogram artifact removing method disclosed in the above embodiment, an electroencephalogram artifact removing device is correspondingly disclosed in this embodiment, please refer to fig. 6, and the device includes:
a signal acquisition unit 100, configured to acquire a segmented signal set of a single-channel electroencephalogram to be processed, where the segmented signal set of the single-channel electroencephalogram to be processed includes single-channel electroencephalogram signal segments including artifacts;
a feature extraction unit 200, configured to extract a feature vector set of the segmented signal set of the single-channel electroencephalogram to be processed;
an artifact removing unit 300, configured to input the feature vector set into an artifact removing model for artifact removing processing, so as to obtain a segmented signal set of a single-channel electroencephalogram with artifacts removed, where the artifact removing model is obtained by training a deep learning network model in advance by using a training set including a plurality of segmented signal sets of single-channel electroencephalograms including at least one artifact, the deep learning network model corresponding to the artifact removing model includes an attention network and a decoder network, and the decoder network includes an artifact feature removing neural network and a waveform reduction mapping network;
and a signal reconstruction unit 400, configured to perform signal reconstruction on the segmented signal set of the artifact-removed single-channel electroencephalogram to obtain a single-channel electroencephalogram signal with the artifact removed.
Optionally, the signal obtaining unit 100 includes:
the signal preprocessing unit is used for acquiring a segmented signal set of a standard single-channel electroencephalogram;
the artifact detection unit is used for utilizing an artifact detection model to perform artifact detection on the segmented signal set of the standard single-channel electroencephalogram, and determining all signal segments containing artifacts as the segmented signal set of the single-channel electroencephalogram to be processed under the condition that the signal segments containing the artifacts are detected; the artifact detection model is obtained by training a deep learning network model in advance by using a training set comprising a plurality of segmented signal sets of single-channel electroencephalograms containing at least one artifact, the deep learning network model corresponding to the artifact detection model comprises a dimension transformation mapping network, a classification feature extraction GRU network, a feature concentration network and an evaluation network, the dimension transformation mapping network is used for transforming the segmented signal sets of the standard single-channel electroencephalograms into signal vector sets with preset dimensions, the classification feature extraction network is used for carrying out feature extraction on the signal vector sets to obtain feature vector sets, the feature concentration network is used for carrying out weighting operation on the feature vector sets, and the evaluation network is used for carrying out artifact detection on each feature vector in the feature vector sets after weighting operation.
Optionally, the signal preprocessing unit is specifically configured to:
and preprocessing the original single-channel electroencephalogram signal to obtain a segmented signal set of the standard single-channel electroencephalogram, wherein the preprocessing comprises resampling, signal segmentation and standardization processing.
Optionally, the feature extraction unit 200 is specifically configured to:
inputting the segmented signal set of the single-channel electroencephalogram to be processed into a preset encoder, wherein the preset encoder comprises a dimension transformation mapping network and a noise feature extraction GRU network;
converting the segmented signal set of the single-channel electroencephalogram to be processed into a vector set with preset dimensions by using the dimension conversion mapping network;
and utilizing the noise feature extraction GRU network to encode the vector set to obtain a deep feature vector set.
Optionally, the apparatus further comprises a signal acquisition unit and a model training unit;
the signal acquisition unit is used for acquiring a single-channel simulation signal, a single-channel semi-simulation signal and a single-channel electroencephalogram original signal, wherein the single-channel simulation signal comprises: simulating an electroencephalogram signal, simulating a muscle motion artifact, and simulating an eye motion artifact, the single-channel semi-simulated signal comprising: artificially removing an electroencephalogram signal of an artifact and an electromyographic motion artifact and an eye motion artifact which are separately collected;
the signal preprocessing unit is further configured to preprocess the single-channel simulation signal and the single-channel semi-simulation signal to obtain the training set, and preprocess the single-channel electroencephalogram original signal to obtain a verification set;
the model training unit is used for training a deep learning network model corresponding to the artifact detection model by using the training set, and verifying the trained model by using the verification set to obtain the artifact detection model; and training a deep learning network model corresponding to the artifact removal model by using the training set, and verifying the trained model by using the verification set to obtain the artifact removal model.
According to the electroencephalogram artifact removing device disclosed by the embodiment, the artifact removing model obtained through pre-training is used for carrying out artifact removing processing on the single-channel electroencephalogram signal, no additional reference channel is needed, artifact removing can be achieved on the single-channel electroencephalogram signal acquired by a single electrode, and the requirement for acquisition equipment is lowered. Moreover, the artifact removal model is obtained by training the deep learning network model by utilizing a training set comprising a plurality of segmented signal sets of single-channel electroencephalograms containing at least one artifact in advance, and can be used for removing the artifact from electroencephalograms containing a plurality of different artifacts, so that the artifact removal effect is improved.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.
It is further noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module may reside 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 above embodiments can be combined arbitrarily, and the features described in the embodiments in the present specification can be replaced or combined with each other in the above description of the disclosed embodiments, so that those skilled in the art can implement or use the present application.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (10)
1. An electroencephalogram artifact removal method, comprising:
acquiring a segmented signal set of a single-channel electroencephalogram to be processed, wherein the segmented signal set of the single-channel electroencephalogram to be processed comprises a single-channel electroencephalogram signal segment containing an artifact;
extracting a feature vector set of a segmented signal set of the single-channel electroencephalogram to be processed;
inputting the feature vector set into an artifact removal model for artifact removal processing to obtain a segmented signal set of a single-channel electroencephalogram with artifacts removed, wherein the artifact removal model is obtained by training a deep learning network model in advance by using a training set comprising a plurality of segmented signal sets of single-channel electroencephalograms with at least one artifact, the deep learning network model corresponding to the artifact removal model comprises an attention network and a decoder network, and the decoder network comprises an artifact feature removal neural network and a waveform reduction mapping network;
performing signal reconstruction on the segmented signal set of the artifact-removed single-channel electroencephalogram to obtain artifact-removed single-channel electroencephalogram signals;
inputting the feature vector set into an artifact removal model for artifact removal processing, including:
inputting the feature vector set into the attention network for weighting operation to obtain a weighted feature vector set;
inputting the weighted feature vector set into the artifact feature removing neural network artifact removing feature to obtain an artifact-free feature vector set;
and inputting the artifact-free feature vector set into the waveform reduction mapping network for waveform reduction.
2. The method of claim 1, wherein the acquiring the segmented signal set of the single channel electroencephalogram to be processed comprises:
acquiring a segmented signal set of a standard single-channel electroencephalogram;
artifact detection is performed on the segmented signal set of the standard single-channel electroencephalogram using an artifact detection model, the artifact detection model is obtained by training a deep learning network model in advance by using a training set comprising a plurality of segmented signal sets of single-channel electroencephalograms containing at least one artifact, the deep learning network model corresponding to the artifact detection model comprises a dimension transformation mapping network, a classification feature extraction GRU network, a feature concentration network and an evaluation network, the dimension transformation mapping network is used for transforming the segmented signal set of the standard single-channel electroencephalogram into a signal vector set with preset dimensions, the classification characteristic extraction network is used for extracting the characteristics of the signal vector set to obtain a characteristic vector set, the feature vector set is weighted by the feature set network, and artifact detection is performed on each feature vector in the feature vector set after weighting by the evaluation network;
and in the case that the signal segment containing the artifact is detected, determining all the signal segments containing the artifact as a segmented signal set of the single-channel electroencephalogram to be processed.
3. The method of claim 2, wherein the acquiring a segmented signal set of a standard single-channel electroencephalogram comprises:
and preprocessing the original single-channel electroencephalogram signal to obtain a segmented signal set of the standard single-channel electroencephalogram, wherein the preprocessing comprises resampling, signal segmentation and standardization processing.
4. The method according to claim 1, wherein the extracting a feature vector set of the segmented signal set of the single-channel electroencephalogram to be processed comprises:
inputting the segmented signal set of the single-channel electroencephalogram to be processed into a preset encoder, wherein the preset encoder comprises a dimension transformation mapping network and a noise feature extraction GRU network;
converting the segmented signal set of the single-channel electroencephalogram to be processed into a vector set with preset dimensions by using the dimension conversion mapping network;
and utilizing the noise feature extraction GRU network to encode the vector set to obtain a deep feature vector set.
5. The method according to claim 1 or 3, characterized in that the method further comprises:
acquiring a single-channel simulation signal, wherein the single-channel simulation signal comprises: simulating an electroencephalogram signal, simulating a muscle motion artifact, and simulating an eye motion artifact;
acquiring a single-channel semi-simulation signal, wherein the single-channel semi-simulation signal comprises: artificially removing an electroencephalogram signal of an artifact and an electromyographic motion artifact and an eye motion artifact which are separately collected;
acquiring a single-channel electroencephalogram original signal;
preprocessing the single-channel simulation signal and the single-channel semi-simulation signal to obtain the training set, and preprocessing the single-channel electroencephalogram original signal to obtain a verification set;
training a deep learning network model corresponding to the artifact detection model by using the training set, and verifying the trained model by using the verification set to obtain the artifact detection model;
and training a deep learning network model corresponding to the artifact removal model by using the training set, and verifying the trained model by using the verification set to obtain the artifact removal model.
6. An electroencephalogram artifact removal apparatus, comprising:
the signal acquisition unit is used for acquiring a segmented signal set of the single-channel electroencephalogram to be processed, wherein the segmented signal set of the single-channel electroencephalogram to be processed comprises single-channel electroencephalogram signal segments containing artifacts;
the feature extraction unit is used for extracting a feature vector set of the segmented signal set of the single-channel electroencephalogram to be processed;
the artifact removing unit is used for inputting the feature vector set into an artifact removing model for artifact removing processing to obtain a segmented signal set of a single-channel electroencephalogram with artifacts removed, the artifact removing model is obtained by utilizing a training set comprising a plurality of segmented signal sets of single-channel electroencephalograms containing at least one artifact to train a deep learning network model in advance, the deep learning network model corresponding to the artifact removing model comprises an attention network and a decoder network, and the decoder network comprises an artifact feature removing neural network and a waveform reduction mapping network;
the signal reconstruction unit is used for reconstructing the segmented signal set of the artifact-removed single-channel electroencephalogram to obtain a single-channel electroencephalogram signal with the artifact removed;
inputting the feature vector set into an artifact removal model for artifact removal processing, including:
inputting the feature vector set into the attention network for weighting operation to obtain a weighted feature vector set;
inputting the weighted feature vector set into the artifact feature removing neural network artifact removing feature to obtain an artifact-free feature vector set;
and inputting the artifact-free feature vector set into the waveform reduction mapping network for waveform reduction.
7. The apparatus of claim 6, wherein the signal obtaining unit comprises:
the signal preprocessing unit is used for acquiring a segmented signal set of a standard single-channel electroencephalogram;
the artifact detection unit is used for utilizing an artifact detection model to perform artifact detection on the segmented signal set of the standard single-channel electroencephalogram, and determining all signal segments containing artifacts as the segmented signal set of the single-channel electroencephalogram to be processed under the condition that the signal segments containing the artifacts are detected; the artifact detection model is obtained by training a deep learning network model in advance by using a training set comprising a plurality of segmented signal sets of single-channel electroencephalograms containing at least one artifact, the deep learning network model corresponding to the artifact detection model comprises a dimension transformation mapping network, a classification feature extraction GRU network, a feature concentration network and an evaluation network, the dimension transformation mapping network is used for transforming the segmented signal sets of the standard single-channel electroencephalograms into signal vector sets with preset dimensions, the classification feature extraction network is used for carrying out feature extraction on the signal vector sets to obtain feature vector sets, the feature concentration network is used for carrying out weighting operation on the feature vector sets, and the evaluation network is used for carrying out artifact detection on each feature vector in the feature vector sets after weighting operation.
8. The apparatus according to claim 7, wherein the signal preprocessing unit is specifically configured to:
and preprocessing the original single-channel electroencephalogram signal to obtain a segmented signal set of the standard single-channel electroencephalogram, wherein the preprocessing comprises resampling, signal segmentation and standardization processing.
9. The apparatus according to claim 6, wherein the feature extraction unit is specifically configured to:
inputting the segmented signal set of the single-channel electroencephalogram to be processed into a preset encoder, wherein the preset encoder comprises a dimension transformation mapping network and a noise feature extraction GRU network;
converting the segmented signal set of the single-channel electroencephalogram to be processed into a vector set with preset dimensions by using the dimension conversion mapping network;
and utilizing the noise feature extraction GRU network to encode the vector set to obtain a deep feature vector set.
10. The apparatus according to claim 6 or 8, wherein the apparatus further comprises a signal acquisition unit and a model training unit;
the signal acquisition unit is used for acquiring a single-channel simulation signal, a single-channel semi-simulation signal and a single-channel electroencephalogram original signal, wherein the single-channel simulation signal comprises: simulating an electroencephalogram signal, simulating a muscle motion artifact, and simulating an eye motion artifact, the single-channel semi-simulated signal comprising: artificially removing an electroencephalogram signal of an artifact and an electromyographic motion artifact and an eye motion artifact which are separately collected;
the signal preprocessing unit is further configured to preprocess the single-channel simulation signal and the single-channel semi-simulation signal to obtain the training set, and preprocess the single-channel electroencephalogram original signal to obtain a verification set;
the model training unit is used for training a deep learning network model corresponding to the artifact detection model by using the training set, and verifying the trained model by using the verification set to obtain the artifact detection model; and training a deep learning network model corresponding to the artifact removal model by using the training set, and verifying the trained model by using the verification set to obtain the artifact removal model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011449582.3A CN112472107B (en) | 2020-12-09 | 2020-12-09 | Electroencephalogram artifact removing method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011449582.3A CN112472107B (en) | 2020-12-09 | 2020-12-09 | Electroencephalogram artifact removing method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112472107A CN112472107A (en) | 2021-03-12 |
| CN112472107B true CN112472107B (en) | 2022-04-19 |
Family
ID=74941637
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011449582.3A Active CN112472107B (en) | 2020-12-09 | 2020-12-09 | Electroencephalogram artifact removing method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112472107B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114343670B (en) * | 2022-01-07 | 2023-07-14 | 北京师范大学 | Interpretation information generation method and electronic equipment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110353672A (en) * | 2019-07-15 | 2019-10-22 | 西安邮电大学 | Eye artefact removal system and minimizing technology in a kind of EEG signals |
| CN111543984A (en) * | 2020-04-13 | 2020-08-18 | 重庆邮电大学 | Method for removing ocular artifacts of electroencephalogram signals based on SSDA (steady state data acquisition) |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB0906029D0 (en) * | 2009-04-07 | 2009-05-20 | Nat Univ Ireland Cork | A method of analysing an electroencephalogram (EEG) signal |
| US11576601B2 (en) * | 2019-04-18 | 2023-02-14 | X Development Llc | Artifact identification in EEG measurements |
-
2020
- 2020-12-09 CN CN202011449582.3A patent/CN112472107B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110353672A (en) * | 2019-07-15 | 2019-10-22 | 西安邮电大学 | Eye artefact removal system and minimizing technology in a kind of EEG signals |
| CN111543984A (en) * | 2020-04-13 | 2020-08-18 | 重庆邮电大学 | Method for removing ocular artifacts of electroencephalogram signals based on SSDA (steady state data acquisition) |
Non-Patent Citations (2)
| Title |
|---|
| Quality Assessment of Single-Channel EEG for Wearable Devices;Fanny Grosselin等;《sensor(Basel)》;20190131;第19卷(第3期);第1-17页 * |
| 基于深度学习的P300脑机接口分类算法研究;肖郴杰;《中国优秀硕士学位论文全文数据库 基础科学辑》;20200215(第2期);第13-51页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112472107A (en) | 2021-03-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110353675B (en) | Electroencephalogram signal emotion recognition method and device based on picture generation | |
| CN107945817B (en) | Method for classifying heart-lung sound signals, method for detecting heart-lung sound signals, device, medium and computer equipment | |
| Hosni et al. | Classification of EEG signals using different feature extraction techniques for mental-task BCI | |
| CN105841961A (en) | Bearing fault diagnosis method based on Morlet wavelet transformation and convolutional neural network | |
| KR20190035368A (en) | Method and Apparatus for Recognizing Emotion Based on Image Converted from Brain Signal | |
| WO2017218492A1 (en) | Neural decoding of attentional selection in multi-speaker environments | |
| CN109598222B (en) | EEMD data enhancement-based wavelet neural network motor imagery electroencephalogram classification method | |
| CN115590515A (en) | Emotion recognition method and system based on generative self-supervision learning and electroencephalogram signals | |
| Chuang et al. | IC-U-Net: a U-Net-based denoising autoencoder using mixtures of independent components for automatic EEG artifact removal | |
| Conradsen et al. | Seizure onset detection based on a uni-or multi-modal intelligent seizure acquisition (UISA/MISA) system | |
| Damon et al. | Non-negative matrix factorization for single-channel EEG artifact rejection | |
| CN112472107B (en) | Electroencephalogram artifact removing method and device | |
| CN113349801A (en) | Imaginary speech electroencephalogram signal decoding method based on convolutional neural network | |
| de Oliveira et al. | Compressive sensing of electroencephalogram: a review | |
| Lee et al. | Analysis of attention deficit hyperactivity disorder in EEG using wavelet transform and self organizing maps | |
| CN114601476A (en) | EEG signal emotion recognition method based on video stimulation | |
| CN114662524A (en) | Plug-and-play type domain adaptation method based on electroencephalogram signals | |
| CN114998252A (en) | Image quality evaluation method based on electroencephalogram signals and memory characteristics | |
| CN112790774A (en) | Original electroencephalogram deep learning classification method and application | |
| CN113180663A (en) | Emotion recognition method and system based on convolutional neural network | |
| CN113887339A (en) | Silent voice recognition system and method fusing surface electromyogram signal and lip image | |
| CN113229842A (en) | Heart and lung sound automatic separation method based on complex deep neural network | |
| CN113208622A (en) | Electroencephalogram EEG signal denoising method and system based on deep neural network technology | |
| CN111899713A (en) | Method, device, equipment and storage medium for silencing communication | |
| Ince et al. | ECoG based brain computer interface with subset selection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |