CN113208614A - Electroencephalogram noise reduction method and device and readable storage medium - Google Patents
Electroencephalogram noise reduction method and device and readable storage medium Download PDFInfo
- Publication number
- CN113208614A CN113208614A CN202110488727.9A CN202110488727A CN113208614A CN 113208614 A CN113208614 A CN 113208614A CN 202110488727 A CN202110488727 A CN 202110488727A CN 113208614 A CN113208614 A CN 113208614A
- Authority
- CN
- China
- Prior art keywords
- electroencephalogram
- artifact
- segments
- noise reduction
- noised
- 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
- 230000009467 reduction Effects 0.000 title claims abstract description 123
- 238000000034 method Methods 0.000 title claims abstract description 59
- 230000004424 eye movement Effects 0.000 claims abstract description 124
- 238000012549 training Methods 0.000 claims abstract description 67
- 238000012545 processing Methods 0.000 claims abstract description 54
- 238000002567 electromyography Methods 0.000 claims abstract description 8
- 210000004556 brain Anatomy 0.000 claims abstract description 5
- 230000005611 electricity Effects 0.000 claims abstract description 4
- 238000001914 filtration Methods 0.000 claims description 36
- 230000003183 myoelectrical effect Effects 0.000 claims description 30
- 238000010586 diagram Methods 0.000 claims description 26
- 238000005070 sampling Methods 0.000 claims description 17
- 239000012634 fragment Substances 0.000 claims description 14
- 238000007781 pre-processing Methods 0.000 claims description 10
- 238000012952 Resampling Methods 0.000 claims description 9
- 238000011176 pooling Methods 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 5
- 230000000694 effects Effects 0.000 abstract description 10
- 230000006870 function Effects 0.000 description 15
- 238000003062 neural network model Methods 0.000 description 11
- 230000008569 process Effects 0.000 description 10
- 238000010606 normalization Methods 0.000 description 9
- 238000012360 testing method Methods 0.000 description 8
- 230000033001 locomotion Effects 0.000 description 7
- 238000012795 verification Methods 0.000 description 6
- 238000013527 convolutional neural network Methods 0.000 description 5
- 230000009471 action Effects 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000007787 long-term memory Effects 0.000 description 3
- 238000000926 separation method Methods 0.000 description 3
- 230000006403 short-term memory Effects 0.000 description 3
- 230000007774 longterm Effects 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000006798 recombination Effects 0.000 description 2
- 238000005215 recombination Methods 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 108010076504 Protein Sorting Signals Proteins 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007177 brain activity Effects 0.000 description 1
- 210000004884 grey matter Anatomy 0.000 description 1
- 210000003128 head Anatomy 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012880 independent component analysis Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 210000003205 muscle Anatomy 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 210000004761 scalp Anatomy 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
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
Abstract
The application provides a method and a device for EEG noise reduction and a readable storage medium. The method for reducing noise of brain electricity comprises the following steps: acquiring an electroencephalogram to be denoised; normalizing the electroencephalogram to be subjected to noise reduction to obtain a normalized electroencephalogram to be subjected to noise reduction; inputting the normalized electroencephalogram to be denoised into a pre-trained electroencephalogram denoising model to obtain a denoised electroencephalogram; the pre-trained electroencephalogram noise reduction model comprises the following training data sets: a de-noised electroencephalogram generated based on the de-noised electroencephalogram and noise signals including an eye movement artifact map and an electromyography artifact map; and performing denormalization processing on the electroencephalogram after noise reduction to obtain the denormalized electroencephalogram after noise reduction. The method is used for realizing effective electroencephalogram noise reduction and improving the noise reduction effect.
Description
Technical Field
The application relates to the technical field of image processing, in particular to a method and a device for electroencephalogram noise reduction and a readable storage medium.
Background
Electroencephalograms reflect changes in electrical potentials conducted to the scalp by neurons in gray matter, and are widely used in psychology, neuroscience, and psychiatry, as well as in the direction of brain-computer interfaces.
Electroencephalogram signals include not only brain activity but also noise, such as: if the electroencephalogram is to be applied, the eye movement artifact and the myoelectric artifact need to be subjected to denoising treatment.
In conventional denoising techniques, artifacts are removed by subtracting an estimated noise template signal from electroencephalogram data based on a regression method. Adaptive filter-based methods rely on the input signal to dynamically estimate the filter coefficients. Blind source separation methods decompose the electroencephalogram signal into components, assign them to the neural sources and artifact sources, and reconstruct a clean signal by recombining the neural components.
The methods are mainly used for multi-channel noise reduction and are difficult to be used for single-channel noise reduction.
Therefore, the existing electroencephalogram noise reduction method has limitations and cannot realize effective electroencephalogram noise reduction.
Disclosure of Invention
The embodiment of the application aims to provide an electroencephalogram noise reduction method and device and a readable storage medium, which are used for effectively reducing noise of electroencephalograms and improving noise reduction effects.
In a first aspect, an embodiment of the present application provides a method for reducing noise in electroencephalogram, including: acquiring an electroencephalogram to be denoised; normalizing the electroencephalogram to be subjected to noise reduction to obtain a normalized electroencephalogram to be subjected to noise reduction; inputting the normalized electroencephalogram to be denoised into a pre-trained electroencephalogram denoising model to obtain a denoised electroencephalogram; the pre-trained electroencephalogram noise reduction model comprises the following training data sets: a de-noised electroencephalogram generated based on the de-noised electroencephalogram and noise signals including an eye movement artifact map and an electromyography artifact map; and performing denormalization processing on the electroencephalogram after noise reduction to obtain the denormalized electroencephalogram after noise reduction.
Compared with the prior art, on one hand, the electroencephalogram noise reduction is realized by utilizing the neural network model, and during training, training can be carried out aiming at a single-channel electroencephalogram signal; when the multi-channel electroencephalogram signals are subjected to noise reduction, the noise reduction can be realized only by sequentially inputting the electroencephalogram signals of a plurality of channels into the neural network model one by one; namely, the noise reduction method can be applied to noise reduction of single-channel electroencephalogram signals and noise reduction of multi-channel electroencephalogram signals, the limitation of the prior art is overcome, and effective electroencephalogram noise reduction is realized. On the other hand, the training set of the neural network model comprises a denoised electroencephalogram and a noisy electroencephalogram, the neural network model is standardized denoising network training data, and the electroencephalogram denoising model trained based on the training data has a good denoising effect.
As a possible implementation, before the acquiring the electroencephalogram to be denoised, the method further comprises: acquiring an electroencephalogram, the eye movement artifact diagram and the myoelectricity artifact diagram from a preset database; generating the de-noised electroencephalogram from the electroencephalogram; generating the noisy electroencephalogram from the de-noised electroencephalogram, the eye movement artifact map, and the myoelectric artifact map; and taking the de-noised electroencephalogram and the electroencephalogram with noise as training data sets, and training an initial electroencephalogram noise reduction model to obtain a trained electroencephalogram noise reduction model.
In the embodiment of the application, an electroencephalogram, an eye movement artifact and a myoelectricity artifact are obtained from a preset database, and based on the three data, a de-noised electroencephalogram is generated firstly; a noisy electroencephalogram is then generated, enabling efficient generation of a standardized training data set.
As one possible implementation, the sampling rate of the electroencephalogram is a first sampling rate, and the generating the de-noised electroencephalogram from the electroencephalogram includes: performing band-pass filtering processing on the electroencephalogram to obtain the electroencephalogram after filtering processing; performing notch filtering processing on the electroencephalogram after filtering processing to obtain the electroencephalogram after notch filtering processing; resampling the electroencephalogram after the notch filtering processing based on a second sampling rate to obtain a resampled electroencephalogram; and denoising the resampled electroencephalogram to obtain the denoised electroencephalogram.
In the embodiment of the application, the electroencephalogram is subjected to band-pass filtering, notch filtering, re-adoption and denoising in sequence to obtain the standardized denoised electroencephalogram.
As one possible implementation, the generating the noisy electroencephalogram from the de-noised electroencephalogram, the eye movement artifact and the electromyography artifact comprises: segmenting the de-noised electroencephalogram to obtain a plurality of de-noised electroencephalogram segments; preprocessing the eye movement artifact graph to obtain a plurality of eye movement artifact segments; preprocessing the myoelectricity artifact graph to obtain a plurality of myoelectricity artifact segments; generating a plurality of electroencephalograms with eye movement artifacts from the plurality of de-noised electroencephalogram segments and the plurality of eye movement artifact segments; generating a plurality of electroencephalograms with electromyographic artifacts according to the plurality of de-noised electroencephalogram segments and the plurality of electromyographic artifact segments.
In the embodiment of the application, a plurality of de-noised electroencephalogram segments, a plurality of eye movement artifact segments and a plurality of artifact segments are correspondingly combined to respectively generate a plurality of electroencephalograms with eye movement artifacts and a plurality of electroencephalograms with myoelectric artifacts, so that the effective generation of the noisy electroencephalograms is realized.
As one possible implementation, the generating a plurality of electroencephalograms with eye movement artifacts from the plurality of denoised electroencephalogram segments and the plurality of eye movement artifact segments, comprises: forming the plurality of de-noised electroencephalogram segments and the plurality of eye movement artifact segments into a plurality of data pairs; each data pair comprises a de-noised electroencephalogram segment and an eye movement artifact segment; combining the de-noised electroencephalogram segments and the eye movement artifact segments in the plurality of data pairs according to a preset first linear combination strategy and a preset first combination frequency to obtain a first preset number of electroencephalograms with eye movement artifacts; the first preset number is the product of the number of the eye movement artifact fragments and the preset first combination times, and the signal-to-noise ratios of the electroencephalograms with the eye movement artifacts of the first preset number are uniformly distributed in a preset signal-to-noise ratio range; the preset first linear combination strategy is expressed as: y is1＝x+λ1N; wherein, y1Representing the electroencephalogram with eye movement artifacts, x representing the de-noised electroencephalogram segment, n representing the eye movement artifact segment, λ1Representing the relative contribution, λ, of the eye movement artifact segments1For adjusting the signal-to-noise ratio of the electroencephalogram with eye movement artifacts.
In the embodiment of the application, the de-noised electroencephalogram segments and the eye movement artifact segments are combined in a linear combination mode, so that the electroencephalogram with the eye movement artifacts can be effectively generated.
As one possible implementation, the generating a plurality of electromyographic images with electromyographic artifacts from the plurality of denoised electroencephalogram segments and the plurality of electromyographic artifact segments includes: forming a plurality of data pairs by the plurality of de-noised electroencephalogram segments and the plurality of electromyographic artifact segments; each data pair comprises a de-noised electroencephalogram segment and a myoelectricity artifact segment; combining the de-noised electroencephalogram segments and the electromyogram artifact segments in the plurality of data pairs according to a preset second linear combination strategy and a preset second combination frequency to obtain a second preset number of electroencephalograms with electromyograms; the second preset number is the product of the number of the myoelectric artifact fragments and the preset second combination times, and the signal-to-noise ratios of the electroencephalograms with the myoelectric artifacts of the second preset number are uniformly distributed within the range of the preset signal-to-noise ratio; the preset second linear combination strategy is expressed as: y is2＝x+λ2M; wherein, y2Representing the electroencephalogram with electromyographic artifacts, x representing the de-noised electroencephalogram segment, m representing an electromyographic artifact segment, λ2Representing the relative contribution of the electromyographic artifact segments, said λ2For adjusting the signal-to-noise ratio of the electroencephalogram with electromyographic artifacts.
In the embodiment of the application, the plurality of de-noised electroencephalogram segments and the plurality of electromyographic artifact segments are combined in a linear combination mode, so that the electroencephalogram with the electromyographic artifacts is effectively generated.
As a possible implementation manner, the pre-trained electroencephalogram noise reduction model includes: the system comprises a plurality of first modules, a second module and a full-connection layer, wherein the first modules are connected in sequence, the second module is connected with the last module in the first modules, and the full-connection layer is connected with the second module; wherein, the first module comprises: the system comprises a plurality of preset blocks and a one-dimensional average pooling layer, wherein the preset blocks are connected in sequence, and the one-dimensional average pooling layer is connected with the next preset block in the preset blocks and has the size of 2; each preset block comprises: 1 x 3 one-dimensional convolution layer of kernel, ReLU activation function; the number of the characteristic graphs of the one-dimensional convolutional layers follows a preset exponential function; the second module comprises: the device comprises a plurality of preset blocks and a tiled layer, wherein the preset blocks are connected in sequence, and the tiled layer is connected with the next preset block in the preset blocks.
In the embodiment of the application, overfitting can be prevented through the novel convolutional neural network, the sampling rate can be gradually reduced through the average pooling layer, and the electroencephalogram noise reduction effect is improved.
As a possible implementation manner, the loss function of the pre-trained electroencephalogram noise reduction model is as follows:
wherein f isi(y) is the output signal of the EEG noise reduction model, xiFor the de-noised electroencephalogram, N is the data point of the input electroencephalogram.
In the embodiment of the application, the noise reduction effect of the electroencephalogram noise reduction model is improved through the loss function based on the sum of squares.
In a second aspect, an embodiment of the present application provides an electroencephalogram noise reduction apparatus, which includes functional modules for implementing the method for reducing noise in an electroencephalogram in the first aspect and any one of the possible implementation manners of the first aspect.
In a third aspect, an embodiment of the present application provides a readable storage medium, where a computer program is stored on the readable storage medium, and when the computer program is executed by a computer, the method for reducing noise in brain electricity as described in the first aspect and any one of the possible implementation manners of the first aspect is performed.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are required to be used in the embodiments of the present application will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and that those skilled in the art can also obtain other related drawings based on the drawings without inventive efforts.
Fig. 1 is a flowchart of a method for electroencephalogram noise reduction provided in an embodiment of the present application;
FIG. 2 is a schematic diagram of an electroencephalogram fragment provided according to an embodiment of the present application;
FIG. 3 is a first schematic diagram of an eye movement artifact segment as provided by an embodiment of the present application;
FIG. 4 is a second schematic diagram of an eye movement artifact segment as provided by an embodiment of the present application;
FIG. 5 is a schematic diagram of an electromyographic artifact segment provided by an embodiment of the present application;
FIG. 6 is a schematic structural diagram of an electroencephalogram noise reduction model provided in an embodiment of the present application;
fig. 7 is a schematic diagram of an electroencephalogram noise reduction device provided in the embodiment of the present application.
Icon: 700-means for electroencephalogram noise reduction; 710-an acquisition module; 720-processing module.
Detailed Description
The technical solutions in the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application.
The electroencephalogram noise reduction method provided by the embodiment of the application can be applied to various application scenes needing electroencephalograms, including but not limited to: brain-computer interface, human-computer interaction, biomedical treatment, etc. Such as: application scenes of psychology, neuroscience, psychiatry, research on brain-computer interface direction and the like.
The hardware environment of the electroencephalogram noise reduction method includes but is not limited to: an electroencephalogram processing apparatus, a server of an electroencephalogram processing system, and the like. Taking an electroencephalogram processing device as an example, an electroencephalogram is input into the electroencephalogram processing device, and the electroencephalogram processing device performs noise reduction processing on the input electroencephalogram by using the electroencephalogram noise reduction method and then outputs the processed electroencephalogram. Taking a server of an electroencephalogram processing system as an example, the electroencephalogram processing system may include a front end and a server, the front end sends an electroencephalogram to the server, the server performs noise reduction processing on an input electroencephalogram by using the electroencephalogram noise reduction method and then feeds back the processed electroencephalogram to the front end, and the front end performs corresponding feedback.
Before the method for reducing noise of electroencephalogram, a simple description is made on electroencephalogram.
In electroencephalogram acquisition, an electroencephalogram signal may be acquired using one electrode, or an electroencephalogram signal may be acquired using a plurality of electrodes. If one electrode acquires an electroencephalogram signal, the acquired electroencephalogram signal is a single-channel electroencephalogram; if multiple electrodes acquire electroencephalogram signals, the acquired electroencephalogram signals are multichannel electroencephalograms, such as: with 64 electrodes acquiring electroencephalogram signals, a 64-channel electroencephalogram can be obtained.
The single-electrode or multi-electrode acquisition method usually corresponds to a sampling rate. As an alternative embodiment, the sampling rate of the acquisition is 512Hz by adopting the international 10-10 system.
In addition, the electroencephalogram may also be an electroencephalogram corresponding to a motor task, which includes a motor imagery task and a real motion. For example, when the collected object imagines the movement of the left hand and the right hand, the collected electroencephalogram is the electroencephalogram corresponding to the imagined movement of the left hand and the right hand; when the collected object does real left and right hand movement, the collected electroencephalogram is the electroencephalogram corresponding to the real left and right hand movement.
The noise included in the electroencephalogram may be: eye movement artifacts and myoelectrical artifacts. The eye movement artifact is a noise signal generated by vibration of eyes when an electroencephalogram is acquired. Myoelectric artifacts are noise signals generated by the vibration of the muscles of the head and neck when electroencephalograms are acquired.
These two artifact signals can be acquired separately, such as: the electrode is arranged on the eye, so that an eye artifact picture can be obtained; the myoelectricity artifact image can be obtained by arranging the electrodes on the face. Similarly, the eye artifact graph and the myoelectricity artifact graph can also adopt a single-electrode or multi-electrode acquisition mode, for example, the eye artifact graph can be acquired through 3-4 electrodes, and signals of 3-4 channels are correspondingly acquired; the electromyogram artifact graph can be acquired through 1-2 electrodes, and signals of 1-2 channels are correspondingly acquired.
Based on the introduction of the application scenario, referring to fig. 1, a flowchart of a method for reducing noise in an electroencephalogram provided in an embodiment of the present application is shown, where the method includes:
step 110: and acquiring the electroencephalogram to be denoised.
Step 120: and carrying out normalization processing on the electroencephalogram to be subjected to noise reduction to obtain the normalized electroencephalogram to be subjected to noise reduction.
Step 130: and inputting the normalized electroencephalogram to be denoised into a pre-trained electroencephalogram denoising model to obtain the electroencephalogram after denoising. The pre-trained electroencephalogram noise reduction model comprises the following training data in a concentrated mode: a de-noised electroencephalogram generated based on the de-noised electroencephalogram and noise signals including an eye movement artifact map and a myoelectric artifact map.
Step 140: and performing denormalization processing on the electroencephalogram subjected to noise reduction to obtain the denormalized electroencephalogram subjected to noise reduction.
Compared with the prior art, on one hand, the electroencephalogram noise reduction is realized by utilizing the neural network model, and during training, training can be carried out aiming at a single-channel electroencephalogram signal; when the multi-channel electroencephalogram signals are subjected to noise reduction, the noise reduction can be realized only by sequentially inputting the electroencephalogram signals of a plurality of channels into the neural network model one by one; namely, the noise reduction method can be applied to noise reduction of single-channel electroencephalogram signals and noise reduction of multi-channel electroencephalogram signals, the limitation of the prior art is overcome, and effective electroencephalogram noise reduction is realized. On the other hand, the training set of the neural network model comprises a denoised electroencephalogram and a noisy electroencephalogram, the neural network model is standardized denoising network training data, and the electroencephalogram denoising model trained based on the training data has a good denoising effect.
Next, a detailed embodiment of the method for reducing noise of brain electricity will be described.
In step 110, the electroencephalogram to be denoised may have different acquisition modes in different application scenarios. Such as: the electroencephalogram to be denoised may be input into an electroencephalogram processing apparatus for a user (such as a physician); the electroencephalogram can also be acquired in real time by the electroencephalogram acquisition device and then transmitted to the electroencephalogram processing device, or a server of the electroencephalogram processing system.
The electroencephalogram to be denoised may be a single channel electroencephalogram or may be a multi-channel electroencephalogram. And the electroencephalogram to be reduced may be one electroencephalogram or a plurality of electroencephalograms, which is not limited in the embodiment of the present application.
In step 120, the normalization processing method may be: the data is divided by the standard deviation of the noisy electroencephalogram signal. The processing mode can be expressed as:
wherein σyThe standard deviation of the electroencephalogram signal with noise is known parameter; x is the noisy electroencephalogram and y is the de-noised electroencephalogram.
By normalizing the electroencephalogram to be denoised, the data input into the electroencephalogram denoising model can be standardized, and the denoising effect is improved.
After the normalization process in step 120 is completed, in step 130, the normalized electroencephalogram to be denoised is input into the pre-trained electroencephalogram denoising model, and the electroencephalogram denoising model can output the denoised electroencephalogram. In order to facilitate understanding of the technical scheme of the embodiment of the present application, a training mode of the electroencephalogram noise reduction model is introduced first.
In the embodiment of the present application, the training of the electroencephalogram noise reduction model depends on a standardized training data set, and in the standardized training data set, the method includes: a de-noised electroencephalogram and a noisy electroencephalogram, wherein the noisy electroencephalogram is generated based on the de-noised electroencephalogram and a noise signal, the noise signal comprising: eye movement trace diagrams and myoelectrical artifact diagrams.
As an optional implementation mode, the training process of the electroencephalogram noise reduction model comprises the following steps: acquiring an electroencephalogram, an eye movement artifact diagram and a myoelectricity artifact diagram from a preset database; generating a de-noised electroencephalogram from the electroencephalogram; generating a noisy electroencephalogram from the de-noised electroencephalogram, the eye movement artifact diagram and the myoelectric artifact diagram; and taking the de-noised electroencephalogram and the electroencephalogram with noise as training data sets, and training the initial electroencephalogram noise reduction model to obtain a trained electroencephalogram noise reduction model.
In the embodiment, the electroencephalogram, the eye movement artifact diagram and the electromyogram artifact diagram are acquired from a preset database, and based on the three data, the de-noised electroencephalogram is generated firstly; a noisy electroencephalogram is then generated, enabling efficient generation of a standardized training data set.
The preset database may be an existing publicly available database in which there are not only electroencephalograms, but also eye movement artifacts and myoelectricity artifacts. As an example, the website corresponding to the preset database may be: http:// gigadb. org/dataset/10295; http; // u4ag2kanosrl. blogspot. jp/; http; fi/URN; NBN is fi; tty, and the like. When the data is obtained, the three kinds of data can be obtained from multiple channels of data so as to ensure the number of samples in the training data set.
After the three data sets are obtained, the electroencephalogram may be processed to generate a de-noised electroencephalogram, and the sampling rate of the unprocessed electroencephalogram is a first sampling rate, such as 512Hz in the previous embodiment. As an optional implementation, the corresponding processing procedure includes: performing band-pass filtering processing on the electroencephalogram to obtain the electroencephalogram after filtering processing; performing notch filtering on the electroencephalogram after filtering to obtain the electroencephalogram after notch filtering; resampling the electroencephalogram after notch filtering based on a second sampling rate to obtain a resampled electroencephalogram; and denoising the resampled electroencephalogram to obtain a denoised electroencephalogram.
In this embodiment, the electroencephalogram is subjected to band-pass filtering, notch filtering, re-adoption, and de-noising in sequence to obtain a standardized de-noised electroencephalogram.
The frequency range of the band-pass filtering can be 1-80Hz, and signals outside the frequency range obviously do not belong to standard electroencephalogram signals, so that the band-pass filtering can initially play a role in removing more obvious noise.
Notch filtering, which is used for removing power frequency interference, wherein corresponding power frequency interference frequencies are different in different application scenarios (such as different countries or regions). By way of example, the power frequency interference frequency may be 50 Hz.
The second sampling rate may be 256Hz at the time of resampling, which is intended to facilitate subsequent segmentation of the electroencephalogram into electroencephalogram segments.
When denoising the resampled electroencephalogram, a denoising method for a multi-channel electroencephalogram may be adopted, such as: blind source separation methods (e.g. independent component analysis). Specifically, the electrical signals of a plurality of channels are mapped to a space with a plurality of dimensions through some statistical methods, then a plurality of dimensions are judged as noise and eliminated, and the rest dimensions are restored to obtain a clean electroencephalogram signal.
Of course, other practicable denoising processing modes besides the blind source separation method may also be adopted, and the embodiment of the present application is not limited thereto.
After obtaining the de-noised electroencephalogram, a noisy electroencephalogram is generated based on the de-noised electroencephalogram, the eye movement artifact and the myoelectricity artifact. As an alternative implementation, the process includes: segmenting the de-noised electroencephalogram to obtain a plurality of de-noised electroencephalogram segments; preprocessing the eye movement artifact graph to obtain a plurality of eye movement artifact segments; preprocessing the myoelectricity artifact graph to obtain a plurality of myoelectricity artifact segments; generating a plurality of electroencephalograms with eye movement artifacts from the plurality of de-noised electroencephalogram segments and the plurality of eye movement artifact segments; and generating a plurality of electroencephalograms with electromyographic artifacts according to the plurality of de-noised electroencephalogram segments and the plurality of electromyographic artifact segments.
In this embodiment, a plurality of denoised electroencephalogram segments, a plurality of eye movement artifact segments and a plurality of artifact segments are combined to generate a plurality of electroencephalograms with eye movement artifacts and a plurality of electroencephalograms with myoelectric artifacts, respectively, thereby achieving efficient generation of a noisy electroencephalogram.
When the denoised electroencephalogram is divided, the denoised electroencephalogram can be divided into 2s segments, and then a plurality of electroencephalogram segments with the time length of 2s can be obtained. By way of example, reference is made to fig. 2, which is an exemplary diagram of the finally obtained 2s electroencephalogram fragment.
The pre-processing of the eye movement artifact map, similar to the processing of the de-noised electroencephalogram, may include: band pass filtering, notch filtering, resampling, signal splitting.
Wherein the band-pass filtering may be in the frequency range of 0.3-10 Hz; the sampling rate of the resampling is the same as that of the electroencephalogram. The signal is divided in the same way as the electroencephalogram, dividing the resampled eye movement artifact map into a plurality of 2s eye movement artifact segments. By way of example, reference is made to fig. 3 and 4 for two exemplary diagrams of a 2s eye movement artifact fragment that is finally obtained.
The preprocessing of the electromyographic artifact map, similar to the processing of the eye movement artifact map, may include: band pass filtering, notch filtering, resampling, signal splitting.
Wherein the band-pass filtering may be in the frequency range of 1-120 Hz; the sampling rate of the resampling is the same as that of the electroencephalogram. The signal segmentation is performed in the same way as the electroencephalogram, by dividing the resampled electromyogram into a plurality of electromyogram segments of 2 s. By way of example, please refer to fig. 5, which is an exemplary diagram of a finally obtained myoelectric artifact segment of 2 s.
After the three signal segments are obtained, the three signal segments can be fed back to an expert, the expert can visually check the three signal segments, and after the expert confirms that the three signal segments have no problems (such as all being clean signals), the three signal segments can be finally applied. According to the actual application result, the number of the finally obtained electroencephalogram segments can be 4514, the number of the eye movement artifact segments can be 3400, and the number of the myoelectric artifact segments can be 5598.
Further, in generating a noisy electroencephalogram based on the three signal segments, on the one hand, a plurality of electroencephalograms with eye movement artifacts, i.e., a first noisy electroencephalogram, are generated from the plurality of denoised electroencephalogram segments and the plurality of eye movement artifact segments; on the other hand, a plurality of electroencephalograms with electromyographic artifacts, i.e. second noisy electroencephalograms, are generated from the plurality of denoised electroencephalogram segments and the plurality of electromyographic artifact segments.
As an alternative embodiment, the process of generating an electroencephalogram with eye movement artifacts includes: forming a plurality of data pairs by the plurality of de-noised electroencephalogram segments and the plurality of eye movement artifact segments; each data pair comprises a de-noised electroencephalogram segment and an eye movement artifact segment; combining the de-noised electroencephalogram segments and the eye movement artifact segments in the plurality of data pairs according to a preset first linear combination strategy and a preset first combination frequency to obtain a first preset number of electroencephalograms with eye movement artifacts; the first preset number is the product of the number of the eye movement artifact fragments and the preset first combination times, and the signal-to-noise ratios of the electroencephalograms with the eye movement artifacts of the first preset number are uniformly distributed in the preset signal-to-noise ratio range; the preset first linear combination strategy is expressed as: y is1＝x+λ1N; wherein, y1Representing an electroencephalogram with eye movement artifacts, x representing a de-noised electroencephalogram segment, n representing an eye movement artifact segment, λ1Representing the relative contribution, λ, of eye movement artifact segments1For adjusting the signal-to-noise ratio of an electroencephalogram with ocular movement artifacts.
In such an embodiment, the plurality of denoised electroencephalogram segments and the plurality of eye movement artifact segments are combined in a linear combination manner, enabling efficient generation of an electroencephalogram with eye movement artifacts.
Wherein, when composing the data pairs, the composition can be performed according to the number of the eye movement artifact segments. Such as: assuming that the number of electroencephalogram segments is 4514 and the number of eye movement artifact segments is 3400, 3400 electroencephalogram segments may be randomly selected from the 4514 electroencephalogram segments, and then the 3400 electroencephalogram segments and the 3400 eye movement artifact segments are combined one by one to obtain 3400 data pairs.
The obtained data pairs can be divided into three parts, one part is used for generating a training data set, one part is used for generating a verification set, and the other part is used for generating a test set. Such as: eighty percent is used to generate the training set, ten percent is used to generate the validation set, and ten percent is used to generate the test set.
Further, when recombination is performed again based on the data pairs, only a corresponding number of the data pairs are combined for the training data set.
A first linear combination strategy, represented as: y is1＝x+λ1N; wherein, y1Representing an electroencephalogram with eye movement artifacts, x representing a de-noised electroencephalogram segment, n representing an eye movement artifact segment, λ1Representing the relative contribution, λ, of eye movement artifact segments1For adjusting the signal-to-noise ratio of an electroencephalogram with ocular movement artifacts. In specific implementation, for the generation of an electroencephalogram with eye movement artifacts, a de-noised electroencephalogram segment can be randomly selected from a data pair, then an eye movement artifact segment is randomly selected from the data pair, then the two segments are mixed, in the mixing process, the signal-to-noise ratio is adjusted, and finally an electroencephalogram with eye movement artifacts is generated.
Wherein λ is1The relative contribution representing the eye movement artifact fragment is a known parameter. The predetermined snr range may be-7 dB to 2dB, and within the snr range, a total of 10 different snrs, and correspondingly, the first combination number may be 10. In generating the training data set, a random number within the signal-to-noise ratio range may be automatically generated and then assigned to the corresponding combination of electroencephalogram segments and eye movement artifact segments.
Furthermore, it will be appreciated that each combination requires coverage of all data pairs, i.e. that after each combination the same number of ocularly artifact segmented electroencephalograms may be obtained. Furthermore, the number of electroencephalograms with eye movement artifacts finally obtained is the product of the number of eye movement artifact segments and the preset first combined number. For example, assuming that eighty percent of the data pairs (i.e., 2700 data pairs) selected from 3400 pairs are combined 10 times to generate a training data set, the number of electroencephalograms with eye movement artifacts finally obtained is: 27000. in this way, an efficient extension of the training data set can be achieved.
As an alternative implementationIn another aspect, a process for generating an electroencephalogram with electromyographic artifacts includes: forming a plurality of data pairs by the plurality of de-noised electroencephalogram segments and the plurality of electromyographic artifact segments; each data pair comprises a de-noised electroencephalogram segment and a myoelectricity artifact segment; combining the de-noised electroencephalogram segments and the electromyogram artifact segments in the plurality of data pairs according to a preset second linear combination strategy and a preset second combination frequency to obtain a second preset number of electroencephalograms with electromyograms; the second preset number is the product of the number of the myoelectricity artifact fragments and the preset second combination times, and the signal-to-noise ratios of the electroencephalograms with the myoelectricity artifacts in the second preset number are uniformly distributed in the preset signal-to-noise ratio range; the preset second linear combination strategy is expressed as: y is2＝x+λ2M; wherein, y2Representing an electroencephalogram with myoelectric artifacts, x representing a de-noised electroencephalogram segment, m representing a myoelectric artifact segment, λ2Representing the relative contribution, λ, of the electromyographic artifact segments2For adjusting the signal-to-noise ratio of the electroencephalogram with electromyographic artifacts.
In this embodiment, the plurality of de-noised electroencephalogram segments and the plurality of electromyographic artifact segments are combined in a linear combination manner, so that the electroencephalogram with the electromyographic artifacts is generated efficiently.
When data pairs are formed, the data pairs can be formed according to the number of the myoelectric artifact segments. Such as: assuming that the number of electroencephalogram segments is 4514 and the number of electromyogram artifact segments is 5598, 5598-4514-1084 data pairs may be randomly selected from the 4514 electroencephalogram segments as reused data pairs, that is, 5598 electroencephalogram segments are obtained in total, wherein 1084 electroencephalogram segments are reused, and the same electroencephalogram segment may exist in the 1084 electroencephalogram segments, or all the electroencephalogram segments may be different electroencephalogram segments. The 5598 electroencephalogram segments are then combined one-by-one with the 5598 electromyography artifact segments to obtain 5598 data pairs.
The obtained data pairs can be divided into three parts, one part is used for generating a training data set, one part is used for generating a verification set, and the other part is used for generating a test set. Such as: eighty percent is used to generate the training set, ten percent is used to generate the validation set, and ten percent is used to generate the test set.
Further, when recombination is performed again based on the data pairs, only a corresponding number of the data pairs are combined for the training data set.
The second linear combination strategy is represented as: y is2＝x+λ2M; wherein, y2Representing an electroencephalogram with myoelectric artifacts, x representing a de-noised electroencephalogram segment, m representing a myoelectric artifact segment, λ2Representing the relative contribution of the myoelectric artifact segments, λ2For adjusting the signal-to-noise ratio of the electroencephalogram with electromyographic artifacts.
In specific implementation, for the generation of an electroencephalogram with myoelectric artifacts, a de-noised electroencephalogram segment can be randomly selected from a data pair, then a myoelectric artifact segment is randomly selected from the data pair, then the two segments are mixed, and in the mixing process, the signal-to-noise ratio is adjusted to finally generate an electroencephalogram with myoelectric artifacts.
Wherein λ is2The relative contribution representing the eye movement artifact fragment is a known parameter. The predetermined snr range may be-7 dB to 2dB, and within the snr range, a total of 10 different snrs, and correspondingly, the first combination number may be 10. In generating the training data set, a random number within the signal-to-noise ratio range may be automatically generated and then assigned to the corresponding combination of electroencephalogram segments and eye movement artifact segments.
Furthermore, it will be understood that each combination requires coverage of all the data pairs, i.e. that after each combination the same number of electromyographic artefacts with electromyographic artefacts can be obtained. And then, the number of the finally obtained electroencephalograms with the electromyographic artifacts is the product of the number of the electromyographic artifact fragments and a preset second combination frequency. For example, assuming that eighty percent of the data pairs (i.e., 4478 pairs) selected from 5598 pairs generate a training data set, the number of combinations is 10, and the number of electroencephalograms with eye movement artifacts finally obtained is: 44780. in this way, an efficient extension of the training data set can be achieved.
The electroencephalogram with the eye movement artifact and the electroencephalogram with the myoelectricity artifact are used as the electroencephalogram with noise, and a training data set can be formed by the electroencephalogram with noise removal. Similarly, the verification set and the test set may also be generated in a training data set manner, and the data amount of the verification set and the number of the test set are different from the number of the training set.
Based on the training data set, an initial electroencephalogram noise reduction model can be trained. Before inputting the training data set into the initial electroencephalogram noise reduction model, normalization processing is performed on the data in the training set, and the embodiment of the normalization processing refers to the description in the foregoing embodiment, and will not be described again.
In the process of training the initial electroencephalogram noise reduction model, the accuracy of the model can be verified and tested by utilizing the verification set and the test set, and the electroencephalogram noise reduction model is improved based on the verification result and the test result so as to ensure the accuracy of the finally trained electroencephalogram noise reduction model.
The electroencephalogram noise reduction model is a neural network model adopted in the embodiment of the application, the neural network model can be in various implementation modes, and the structure of the electroencephalogram noise reduction model is introduced next. It should be noted that, no matter the model is a trained electroencephalogram noise reduction model or an initial electroencephalogram noise reduction model, the model structure is not changed.
Referring to fig. 6, a schematic diagram of a first model structure of an electroencephalogram noise reduction model is shown, in fig. 6, the electroencephalogram noise reduction model includes 6 first modules connected in sequence, a second module connected to a last module of the 6 first modules, and a full connection layer connected to the second module.
Wherein, the first module includes: the system comprises 2 preset blocks which are connected in sequence and 1 one-dimensional average pooling layer with the size of 2, wherein the 1 one-dimensional average pooling layer is connected with the next preset block in the 2 preset blocks; each preset block comprises: 1 × 3 kernel one-dimensional convolution layer, 1 ReLU (linear rectification function) activation function; the number of feature maps of the one-dimensional convolutional layers follows a predetermined exponential function. The second module comprises: the device comprises 2 preset blocks and a tiled layer, wherein the 2 preset blocks are connected in sequence, and the tiled layer is connected with the next preset block in the 2 preset blocks.
In the embodiment, overfitting can be prevented through the novel convolutional neural network, the sampling rate can be gradually reduced through the average pooling layer, and the electroencephalogram noise reduction effect is improved.
It should be noted that the number of the preset blocks, the number of the first modules, the number of the one-dimensional average pooling layers, and the number of the flat layers in the model structure shown in fig. 6 may be changed according to the actual application scenario, and the number defined in fig. 6 does not constitute the definition of the convolutional neural network structure.
The predetermined exponential function may be 16 x 2n. The numerical value of the feature map number (a parameter) of the convolutional neural network layer obeys an exponential distribution through the exponential function, and further the feature can be gradually extracted and the feature dimension can be increased through the convolutional neural network layer with the feature map number gradually increased.
The second model structure of the electroencephalogram noise reduction model can be a four-layer fully-connected network, and a ReLu activation function layer and a Dropout (regularization) layer for preventing overfitting are connected behind each fully-connected layer. The number of neurons in each fully connected layer is equal to the number of samples of the input signal. The noisy electroencephalogram is input from the first layer, and then the denoised electroencephalogram is output from the last layer.
The third model structure of the electroencephalogram noise reduction model can be a simple convolution network. The network contains 4 1-dimensional convolutional layers, each convolutional layer having a convolutional kernel size of 1 × 3, a step size of 1, and a feature map number of 64(k3n64s 1). Each 1-dimensional convolutional layer is followed by a batch normalization layer and a ReLu activation function layer. To reconstruct the signal, the final convolutional layer is followed by a fully-connected layer from which the de-noised electroencephalogram is output.
The fourth model structure of the electroencephalogram noise reduction model can be a long-term and short-term memory network. The long-term and short-term memory network can learn long-term dependence, and is beneficial to distinguishing long-term characteristics of noise and electroencephalogram signals. Each electroencephalogram data is sequentially input to the long-short term memory network unit and a series of signal sequences of equal length is output. The long-term and short-term memory network layer is followed by three layers of full-connection networks, and a ReLu activation function layer and a regularization layer are connected behind each full-connection layer.
In practical application, one model structure can be determined from the four model structures according to specific requirements in an application scene, and then the model structure is applied.
In this embodiment of the present application, the loss function of the electroencephalogram noise reduction model may be:
wherein f isi(y) is the output signal of the EEG noise reduction model, xiFor de-noised electroencephalograms, N is the data point for the input electroencephalogram.
For example, assuming that the input electroencephalogram consists of 512 data points, then N is 512, and then i refers to the data point at the ith position in the series of electroencephalogram data points, ranging from 1 to 512.
The loss function simply calculates the sum of squares between the output of the electroencephalogram noise reduction model and the reference clean signal (i.e., the de-noised electroencephalogram) as the error between the output and the reference clean signal.
In the embodiment of the application, the noise reduction effect of the electroencephalogram noise reduction model is improved through the loss function based on the sum of squares.
According to the electroencephalogram noise reduction model training method and the electroencephalogram noise reduction model training device, the electroencephalogram noise reduction model training method and the electroencephalogram noise reduction model training system can be seen, the standardized electroencephalogram noise reduction training data set is constructed based on the electroencephalogram data set, the electrooculogram data set and the electromyogram data set, the neural network model supervision learning method is adopted, effective training of the electroencephalogram noise reduction model can be achieved, and the noise reduction effect of the electroencephalogram noise reduction model is guaranteed.
And further, inputting the normalized electroencephalogram to be denoised into the trained electroencephalogram denoising model based on the trained electroencephalogram denoising model, and outputting the denoised electroencephalogram by the electroencephalogram denoising model.
In step 140, since the normalized data is input to the model and the normalized data is output from the model, the denoised electroencephalogram needs to be denormalised to obtain a standard denoised electroencephalogram.
Referring to the normalization process, in the normalization process, the amplitude of the electroencephalogram signal after noise reduction is recovered by multiplying the electroencephalogram after noise reduction by the corresponding standard deviation, so that the normalization can be realized.
In the embodiment of the application, a training data set is generated mainly based on an electroencephalogram, an eye movement artifact and a myoelectricity artifact, and the eye movement artifact and the myoelectricity artifact can be removed simultaneously by an electroencephalogram noise reduction model obtained by training the training data set.
In actual application, other training data set generation modes can be adopted according to the requirements of application scenarios. Such as: generating a first training data set based on an electroencephalogram and an eye movement artifact diagram, and generating a second training data set based on the electroencephalogram and an electromyography artifact diagram; then training an electroencephalogram noise reduction model for removing eye movement artifacts through a first training data set; and training an electroencephalogram noise reduction model for removing myoelectric artifacts through a second training data set.
Based on the same inventive concept, please refer to fig. 7, an embodiment of the present application further provides an apparatus 700 for electroencephalogram noise reduction, which includes an obtaining module 710 and a processing module 720.
An obtaining module 710, configured to obtain an electroencephalogram to be denoised. A processing module 720 for: inputting the normalized electroencephalogram to be denoised into a pre-trained electroencephalogram denoising model to obtain a denoised electroencephalogram; the pre-trained electroencephalogram noise reduction model comprises the following training data sets: a de-noised electroencephalogram generated based on the de-noised electroencephalogram and noise signals including an eye movement artifact map and an electromyography artifact map; and performing denormalization processing on the electroencephalogram after noise reduction to obtain the denormalized electroencephalogram after noise reduction.
In this embodiment of the application, the obtaining module 710 is further configured to obtain an electroencephalogram, the eye movement artifact diagram, and the myoelectricity artifact diagram from a preset database; the processing module 720 is further configured to generate the de-noised electroencephalogram from the electroencephalogram; generating the noisy electroencephalogram from the de-noised electroencephalogram, the eye movement artifact map, and the myoelectric artifact map; and taking the de-noised electroencephalogram and the electroencephalogram with noise as training data sets, and training an initial electroencephalogram noise reduction model to obtain a trained electroencephalogram noise reduction model.
In this embodiment of the application, the processing module 720 is specifically configured to perform band-pass filtering on the electroencephalogram to obtain a filtered electroencephalogram; performing notch filtering processing on the electroencephalogram after filtering processing to obtain the electroencephalogram after notch filtering processing; resampling the electroencephalogram after the notch filtering processing based on a second sampling rate to obtain a resampled electroencephalogram; and denoising the resampled electroencephalogram to obtain the denoised electroencephalogram.
In this embodiment of the application, the processing module 720 is further specifically configured to: segmenting the de-noised electroencephalogram to obtain a plurality of de-noised electroencephalogram segments; preprocessing the eye movement artifact graph to obtain a plurality of eye movement artifact segments; preprocessing the myoelectricity artifact graph to obtain a plurality of myoelectricity artifact segments; generating a plurality of electroencephalograms with eye movement artifacts from the plurality of de-noised electroencephalogram segments and the plurality of eye movement artifact segments; generating a plurality of electroencephalograms with electromyographic artifacts according to the plurality of de-noised electroencephalogram segments and the plurality of electromyographic artifact segments.
In this embodiment of the application, the processing module 720 is further specifically configured to: forming the plurality of de-noised electroencephalogram segments and the plurality of eye movement artifact segments into a plurality of data pairs; each data pair comprises a de-noised electroencephalogram segment and an eye movement artifact segment; combining the de-noised electroencephalogram segments and the eye movement artifact segments in the plurality of data pairs according to a preset first linear combination strategy and a preset first combination frequency to obtain a first preset number of electroencephalograms with eye movement artifacts; the first preset number is the product of the number of the eye movement artifact fragments and the preset first combination times, and the signal-to-noise ratios of the electroencephalograms with the eye movement artifacts of the first preset number are uniformly distributed in a preset signal-to-noise ratio range; the preset first linear combination strategy is expressed as: y is1＝x+λ1N; wherein, y1Representing the electroencephalogram with eye movement artifacts, x representing the de-noised electroencephalogram segment, n representing the eye movement artifact segment, λ1Representing the relative contribution, λ, of the eye movement artifact segments1For adjusting the signal-to-noise ratio of the electroencephalogram with eye movement artifacts.
In this embodiment of the application, the processing module 720 is further specifically configured to: forming a plurality of data pairs by the plurality of de-noised electroencephalogram segments and the plurality of electromyographic artifact segments; each data pair comprises a de-noised electroencephalogram segment and a myoelectricity artifact segment; combining the de-noised electroencephalogram segments and the electromyogram artifact segments in the plurality of data pairs according to a preset second linear combination strategy and a preset second combination frequency to obtain a second preset number of electroencephalograms with electromyograms; the second preset number is the product of the number of the myoelectric artifact fragments and the preset second combination times, and the signal-to-noise ratios of the electroencephalograms with the myoelectric artifacts of the second preset number are uniformly distributed within the range of the preset signal-to-noise ratio; the preset second linear combination strategy is expressed as: y is2＝x+λ2M; wherein, y2Representing the electroencephalogram with electromyographic artifacts, x representing the de-noised electroencephalogram segment, m representing an electromyographic artifact segment, λ2Representing the relative contribution of the electromyographic artifact segments, said λ2For adjusting the signal-to-noise ratio of the electroencephalogram with electromyographic artifacts.
The electroencephalogram noise reduction device 700 corresponds to the electroencephalogram noise reduction method in the foregoing embodiment, and each module corresponds to each step of the electroencephalogram noise reduction method, so that the implementation of each module refers to the implementation of each step of the electroencephalogram noise reduction method, and no repeated description is provided here.
Based on the same inventive concept, an embodiment of the present application further provides a readable storage medium, where a computer program is stored on the readable storage medium, and when the computer program is executed by a computer, the method for reducing noise in electroencephalogram described in the foregoing embodiment is executed.
In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. The above-described embodiments of the apparatus are merely illustrative, and for example, the division of the units is only one logical division, and there may be other divisions when actually implemented, and for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection of devices or units through some communication interfaces, and may be in an electrical, mechanical or other form.
In addition, units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
Furthermore, the functional modules in the embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.
In this document, 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.
The above description is only an example of the present application and is not intended to limit the scope of the present application, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims (10)
1. A method for electroencephalogram noise reduction, comprising:
acquiring an electroencephalogram to be denoised;
normalizing the electroencephalogram to be subjected to noise reduction to obtain a normalized electroencephalogram to be subjected to noise reduction;
inputting the normalized electroencephalogram to be denoised into a pre-trained electroencephalogram denoising model to obtain a denoised electroencephalogram; the pre-trained electroencephalogram noise reduction model comprises the following training data sets: a de-noised electroencephalogram generated based on the de-noised electroencephalogram and noise signals including an eye movement artifact map and an electromyography artifact map;
and performing denormalization processing on the electroencephalogram after noise reduction to obtain the denormalized electroencephalogram after noise reduction.
2. The method according to claim 1, wherein prior to said acquiring the electroencephalogram to be denoised, the method further comprises:
acquiring an electroencephalogram, the eye movement artifact diagram and the myoelectricity artifact diagram from a preset database;
generating the de-noised electroencephalogram from the electroencephalogram;
generating the noisy electroencephalogram from the de-noised electroencephalogram, the eye movement artifact map, and the myoelectric artifact map;
and taking the de-noised electroencephalogram and the electroencephalogram with noise as training data sets, and training an initial electroencephalogram noise reduction model to obtain a trained electroencephalogram noise reduction model.
3. The method of claim 2, wherein the electroencephalogram has a sampling rate of a first sampling rate, the generating the de-noised electroencephalogram from the electroencephalogram comprising:
performing band-pass filtering processing on the electroencephalogram to obtain the electroencephalogram after filtering processing;
performing notch filtering processing on the electroencephalogram after filtering processing to obtain the electroencephalogram after notch filtering processing;
resampling the electroencephalogram after the notch filtering processing based on a second sampling rate to obtain a resampled electroencephalogram;
and denoising the resampled electroencephalogram to obtain the denoised electroencephalogram.
4. The method according to claim 3, wherein said generating the noisy electroencephalogram from the de-noised electroencephalogram, the eye movement artifact map, and the myoelectric artifact map comprises:
segmenting the de-noised electroencephalogram to obtain a plurality of de-noised electroencephalogram segments;
preprocessing the eye movement artifact graph to obtain a plurality of eye movement artifact segments;
preprocessing the myoelectricity artifact graph to obtain a plurality of myoelectricity artifact segments;
generating a plurality of electroencephalograms with eye movement artifacts from the plurality of de-noised electroencephalogram segments and the plurality of eye movement artifact segments;
generating a plurality of electroencephalograms with electromyographic artifacts according to the plurality of de-noised electroencephalogram segments and the plurality of electromyographic artifact segments.
5. The method according to claim 4, wherein said generating a plurality of electroencephalograms with eye movement artifacts from said plurality of de-noised electroencephalogram segments and said plurality of eye movement artifact segments comprises:
forming the plurality of de-noised electroencephalogram segments and the plurality of eye movement artifact segments into a plurality of data pairs; each data pair comprises a de-noised electroencephalogram segment and an eye movement artifact segment;
combining the de-noised electroencephalogram segments and the eye movement artifact segments in the plurality of data pairs according to a preset first linear combination strategy and a preset first combination frequency to obtain a first preset number of electroencephalograms with eye movement artifacts; the first preset number is the product of the number of the eye movement artifact fragments and the preset first combination times, and the signal-to-noise ratios of the electroencephalograms with the eye movement artifacts of the first preset number are uniformly distributed in a preset signal-to-noise ratio range;
the above-mentionedThe preset first linear combination strategy is expressed as: y is1＝x+λ1N; wherein, y1Representing the electroencephalogram with eye movement artifacts, x representing the de-noised electroencephalogram segment, n representing the eye movement artifact segment, λ1Representing the relative contribution, λ, of the eye movement artifact segments1For adjusting the signal-to-noise ratio of the electroencephalogram with eye movement artifacts.
6. The method according to claim 4, wherein said generating a plurality of electromyographic artifacts based on said plurality of de-noised electroencephalogram segments and said plurality of electromyographic artifact segments comprises:
forming a plurality of data pairs by the plurality of de-noised electroencephalogram segments and the plurality of electromyographic artifact segments; each data pair comprises a de-noised electroencephalogram segment and a myoelectricity artifact segment;
combining the de-noised electroencephalogram segments and the electromyogram artifact segments in the plurality of data pairs according to a preset second linear combination strategy and a preset second combination frequency to obtain a second preset number of electroencephalograms with electromyograms; the second preset number is the product of the number of the myoelectric artifact fragments and the preset second combination times, and the signal-to-noise ratios of the electroencephalograms with the myoelectric artifacts of the second preset number are uniformly distributed within the range of the preset signal-to-noise ratio;
the preset second linear combination strategy is expressed as: y is2＝x+λ2M; wherein, y2Representing the electroencephalogram with electromyographic artifacts, x representing the de-noised electroencephalogram segment, m representing an electromyographic artifact segment, λ2Representing the relative contribution of the electromyographic artifact segments, said λ2For adjusting the signal-to-noise ratio of the electroencephalogram with electromyographic artifacts.
7. The method of claim 1, wherein the pre-trained brain electrical noise reduction model comprises: the system comprises a plurality of first modules, a second module and a full-connection layer, wherein the first modules are connected in sequence, the second module is connected with the last module in the first modules, and the full-connection layer is connected with the second module;
wherein, the first module comprises: the system comprises a plurality of preset blocks and a one-dimensional average pooling layer, wherein the preset blocks are connected in sequence, and the one-dimensional average pooling layer is connected with the next preset block in the preset blocks and has the size of 2; each preset block comprises: 1 x 3 one-dimensional convolution layer of kernel, ReLU activation function; the number of the characteristic graphs of the one-dimensional convolutional layers follows a preset exponential function;
the second module comprises: the device comprises a plurality of preset blocks and a tiled layer, wherein the preset blocks are connected in sequence, and the tiled layer is connected with the next preset block in the preset blocks.
8. The method of claim 1, wherein the loss function of the pre-trained electroencephalogram noise reduction model is:
wherein f isi(y) is the output signal of the EEG noise reduction model, xiFor the de-noised electroencephalogram, N is the data point of the input electroencephalogram.
9. An apparatus for reducing noise in brain electricity, comprising:
the acquisition module is used for acquiring an electroencephalogram to be denoised;
a processing module to:
inputting the normalized electroencephalogram to be denoised into a pre-trained electroencephalogram denoising model to obtain a denoised electroencephalogram; the pre-trained electroencephalogram noise reduction model comprises the following training data sets: a de-noised electroencephalogram generated based on the de-noised electroencephalogram and noise signals including an eye movement artifact map and an electromyography artifact map;
and performing denormalization processing on the electroencephalogram after noise reduction to obtain the denormalized electroencephalogram after noise reduction.
10. A readable storage medium, having stored thereon a computer program which, when executed by a computer, performs the method of electroencephalogram noise reduction according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110488727.9A CN113208614A (en) | 2021-04-30 | 2021-04-30 | Electroencephalogram noise reduction method and device and readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110488727.9A CN113208614A (en) | 2021-04-30 | 2021-04-30 | Electroencephalogram noise reduction method and device and readable storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113208614A true CN113208614A (en) | 2021-08-06 |
Family
ID=77091067
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110488727.9A Pending CN113208614A (en) | 2021-04-30 | 2021-04-30 | Electroencephalogram noise reduction method and device and readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113208614A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113484913A (en) * | 2021-08-16 | 2021-10-08 | 成都理工大学 | Seismic data denoising method with multi-granularity feature fusion convolution neural network |
| CN114190953A (en) * | 2021-12-09 | 2022-03-18 | 四川新源生物电子科技有限公司 | Training method and system of electroencephalogram signal noise reduction model for electroencephalogram acquisition equipment |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018053968A1 (en) * | 2016-09-21 | 2018-03-29 | 广州视源电子科技股份有限公司 | Device for removing electrooculograph artifact in sleep condition analysis |
| US20190015052A1 (en) * | 2016-01-15 | 2019-01-17 | The Regents Of The University Of California | Machine-learning-based denoising of doppler ultrasound blood flow and intracranial pressure signal |
| CN109784242A (en) * | 2018-12-31 | 2019-05-21 | 陕西师范大学 | EEG Noise Cancellation based on one-dimensional residual error convolutional neural networks |
| WO2020128134A1 (en) * | 2018-12-18 | 2020-06-25 | Nokia Technologies Oy | Data denoising based on machine learning |
| CN112244876A (en) * | 2020-09-30 | 2021-01-22 | 北京工业大学 | Electroencephalogram signal time sequence random noise removing method of adaptive filter based on variable forgetting factor |
-
2021
- 2021-04-30 CN CN202110488727.9A patent/CN113208614A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190015052A1 (en) * | 2016-01-15 | 2019-01-17 | The Regents Of The University Of California | Machine-learning-based denoising of doppler ultrasound blood flow and intracranial pressure signal |
| WO2018053968A1 (en) * | 2016-09-21 | 2018-03-29 | 广州视源电子科技股份有限公司 | Device for removing electrooculograph artifact in sleep condition analysis |
| WO2020128134A1 (en) * | 2018-12-18 | 2020-06-25 | Nokia Technologies Oy | Data denoising based on machine learning |
| CN109784242A (en) * | 2018-12-31 | 2019-05-21 | 陕西师范大学 | EEG Noise Cancellation based on one-dimensional residual error convolutional neural networks |
| CN112244876A (en) * | 2020-09-30 | 2021-01-22 | 北京工业大学 | Electroencephalogram signal time sequence random noise removing method of adaptive filter based on variable forgetting factor |
Non-Patent Citations (3)
| Title |
|---|
| JONATHAN R. WOLPAW 等: "《个体建构视角下的刻板印象激活》", 西安交通大学出版社, pages: 1549 * |
| 吉安卡洛•扎克尼 等: "《TensorFlow深度学习》", 31 March 2020, 机械工业出版社, pages: 1 - 2 * |
| 陈云霁 等: "《智能计算系统》", 30 April 2020, 机械工业出版社, pages: 41 - 68 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113484913A (en) * | 2021-08-16 | 2021-10-08 | 成都理工大学 | Seismic data denoising method with multi-granularity feature fusion convolution neural network |
| CN113484913B (en) * | 2021-08-16 | 2023-06-16 | 成都理工大学 | Seismic data denoising method for multi-granularity feature fusion convolutional neural network |
| CN114190953A (en) * | 2021-12-09 | 2022-03-18 | 四川新源生物电子科技有限公司 | Training method and system of electroencephalogram signal noise reduction model for electroencephalogram acquisition equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110765920B (en) | Motor imagery classification method based on convolutional neural network | |
| Sun et al. | A novel end-to-end 1D-ResCNN model to remove artifact from EEG signals | |
| Leite et al. | Deep convolutional autoencoder for EEG noise filtering | |
| CN110353672B (en) | System and method for removing eye artifacts in electroencephalogram signals | |
| Wu et al. | One-versus-the-rest (OVR) algorithm: An extension of common spatial patterns (CSP) algorithm to multi-class case | |
| Chan et al. | The removal of ocular artifacts from EEG signals using adaptive filters based on ocular source components | |
| CN113208614A (en) | Electroencephalogram noise reduction method and device and readable storage medium | |
| CN114224360B (en) | EEG signal processing method, equipment and storage medium based on improved EMD-ICA | |
| CN112890827B (en) | Electroencephalogram identification method and system based on graph convolution and gate control circulation unit | |
| CN111543984B (en) | Method for removing ocular artifacts of electroencephalogram signals based on SSDA | |
| CN114947883A (en) | Time-frequency domain information fusion deep learning electroencephalogram noise reduction method | |
| CN113723171A (en) | Electroencephalogram signal denoising method based on residual error generation countermeasure network | |
| CN114601476A (en) | EEG signal emotion recognition method based on video stimulation | |
| CN114648048A (en) | Electrocardiosignal noise reduction method based on variational self-coding and PixelCNN model | |
| CN114081503A (en) | Method for removing ocular artifacts in electroencephalogram signals | |
| CN113723557A (en) | Depression electroencephalogram classification system based on multiband time-space convolution network | |
| Dobrea et al. | The selection of proper discriminative cognitive tasks—A necessary prerequisite in high-quality BCI applications | |
| CN113180659A (en) | Electroencephalogram emotion recognition system based on three-dimensional features and cavity full convolution network | |
| Khan et al. | Wrist movement discrimination in single-trial EEG for Brain–Computer Interface using band powers | |
| CN115462803A (en) | BG-Attention-based electroencephalogram signal denoising method, device and storage medium | |
| CN111493864B (en) | EEG signal mixed noise processing method, equipment and storage medium | |
| CN114795247A (en) | Electroencephalogram signal analysis method and device, electronic equipment and storage medium | |
| Kordmahale et al. | A novel artifact removal method for the SSVEP signal using hybrid CCA-DWT and comparative analysis for feature selection and classification in the P300 signal | |
| Rekrut et al. | Improving Silent Speech BCI Training Procedures Through Transfer from Overt to Silent Speech | |
| CN116602691B (en) | Denoising method and device for electroencephalogram signals, electronic equipment and storage medium |
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 |
Application publication date: 20210806 |
|
| RJ01 | Rejection of invention patent application after publication |