CN115054270A - Sleep staging method and system for extracting sleep spectrogram features based on GCN - Google Patents
Sleep staging method and system for extracting sleep spectrogram features based on GCN Download PDFInfo
- Publication number
- CN115054270A CN115054270A CN202210687782.5A CN202210687782A CN115054270A CN 115054270 A CN115054270 A CN 115054270A CN 202210687782 A CN202210687782 A CN 202210687782A CN 115054270 A CN115054270 A CN 115054270A
- Authority
- CN
- China
- Prior art keywords
- sleep
- vector sequence
- electroencephalogram
- sequence
- gcn
- 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
- 238000000034 method Methods 0.000 title claims abstract description 46
- 239000013598 vector Substances 0.000 claims abstract description 88
- 238000012545 processing Methods 0.000 claims abstract description 29
- 238000001228 spectrum Methods 0.000 claims description 18
- 238000013528 artificial neural network Methods 0.000 claims description 14
- 230000008667 sleep stage Effects 0.000 claims description 12
- 230000002457 bidirectional effect Effects 0.000 claims description 9
- 230000007246 mechanism Effects 0.000 claims description 4
- 230000000306 recurrent effect Effects 0.000 claims description 4
- 230000001131 transforming effect Effects 0.000 claims description 3
- 230000004461 rapid eye movement Effects 0.000 claims description 2
- 238000000605 extraction Methods 0.000 claims 1
- 239000011159 matrix material Substances 0.000 description 34
- 238000004364 calculation method Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 7
- 230000004913 activation Effects 0.000 description 6
- 238000012544 monitoring process Methods 0.000 description 4
- 125000004122 cyclic group Chemical group 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 208000019116 sleep disease Diseases 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000003595 spectral effect Effects 0.000 description 2
- 208000010340 Sleep Deprivation Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000005180 public health Effects 0.000 description 1
- 208000020685 sleep-wake disease Diseases 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 230000002618 waking effect Effects 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
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/369—Electroencephalography [EEG]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/48—Other medical applications
- A61B5/4806—Sleep evaluation
- A61B5/4812—Detecting sleep stages or cycles
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7225—Details of analog processing, e.g. isolation amplifier, gain or sensitivity adjustment, filtering, baseline or drift compensation
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7253—Details of waveform analysis characterised by using transforms
- A61B5/7257—Details of waveform analysis characterised by using transforms using Fourier transforms
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Signal Processing (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Biophysics (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Pathology (AREA)
- Artificial Intelligence (AREA)
- Psychiatry (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physiology (AREA)
- Mathematical Physics (AREA)
- Fuzzy Systems (AREA)
- Evolutionary Computation (AREA)
- Power Engineering (AREA)
- Psychology (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses a sleep staging method and a sleep staging system for extracting sleep spectrogram characteristics based on GCN, wherein the method comprises the following steps: s1, acquiring an electroencephalogram signal, and processing the electroencephalogram signal to obtain an electroencephalogram time-frequency image; s2, based on the time-frequency image, carrying out position coding processing to obtain a characteristic vector sequence; s3, modeling based on the characteristic vector sequence to obtain an output vector sequence; and S4, classifying the output vector sequences based on the output vector sequences to obtain the stage result of the output vector sequences. The system comprises: the device comprises a signal acquisition unit, a coding processing unit, a modeling unit and a classification unit; the invention provides interpretability for the model by coding the self-attention score into a heat map based on a Transformer backbone. At the sequence processing level, a novel multi-branch strategy is provided, the final feature vector is calculated, and the output is classified by using a softmax layer, so that the staging result of the sequence is obtained.
Description
Technical Field
The invention belongs to the field of sleep monitoring, and particularly relates to a sleep staging method and a sleep staging system for extracting sleep spectrogram features based on GCN.
Background
Sleep problems such as sleep disorder and sleep deprivation bring huge burden to public health. How to analyze the sleep condition and better diagnose the sleep problem has important significance on the global sleep health. Sleep staging is the first step in sleep diagnosis and assessment and is also a critical step. In recent years, thanks to the development of deep learning technology, researchers have tried to apply deep learning methods to the field of automatic sleep staging, and early attempts have mainly used relatively popular deep network frameworks such as convolutional neural networks, cyclic neural networks, automatic encoders, etc., and have achieved certain results.
However, in clinical application, sleep evaluation and monitoring still depend heavily on manual evaluation of sleep experts and there is no automatic sleep staging scheme that can be effectively applied to the fields of home monitoring and the like. This is mainly because the existing automatic sleep staging methods have two main drawbacks: firstly, the structural stacking or independent structures of different neural networks do not well combine the characteristics of sleep electroencephalogram signals, so that the classification result obtained by simple probability calculation of the finally obtained characteristic vector is low in accuracy; and secondly, the 'black box suspicion' of deep learning also limits sleep evaluation in application scenes such as clinical and accurate home monitoring to a certain extent.
Disclosure of Invention
The invention provides a sleep stage method and a sleep stage system for extracting sleep spectrogram characteristics based on GCN. At the sequence processing level, a novel multi-branch strategy is provided, and finally, a final feature vector is calculated, and the output is classified by using a softmax layer, so that the stage result of the sequence is obtained.
In order to achieve the purpose, the invention provides the following scheme:
the sleep staging method for extracting the sleep spectrogram characteristics based on the GCN comprises the following steps:
s1, acquiring an electroencephalogram signal, and processing the electroencephalogram signal to obtain an electroencephalogram time-frequency image;
s2, performing position coding processing based on the electroencephalogram time-frequency image to obtain a characteristic vector sequence;
s3, modeling based on the characteristic vector sequence to obtain an output vector sequence;
and S4, classifying the output vector sequence based on the output vector sequence to obtain the stage result of the output vector sequence.
Preferably, the S1 method for processing the electroencephalogram signal includes:
s11, carrying out short-time Fourier transform on the electroencephalogram signal to obtain a power spectrum;
s12, transforming the power spectrum by adopting a logarithmic scaling method to obtain a logarithmic power spectrum; and obtaining the electroencephalogram time-frequency image based on the logarithmic power spectrum.
Preferably, the position encoding method in S2 includes: and (3) coding the electroencephalogram signals by adopting a Transformer as a main structure and combining an attention mechanism.
Preferably, the modeling method in S3 includes: and modeling the characteristic vector sequence by adopting a graph convolution network and a bidirectional recurrent neural network.
Preferably, the output vector sequence classification in S4 adopts a softmax function.
Preferably, the staging result in S4 corresponds to five stages of sleep staging: awake phase, N1 phase, N2 phase, N3 phase and rapid eye movement phase.
The invention also provides a sleep staging system for extracting sleep spectrogram characteristics based on the GCN, which comprises the following steps: the device comprises a signal acquisition unit, a coding processing unit, a modeling unit and a classification unit;
the signal acquisition unit is connected with the coding processing unit and is used for acquiring electroencephalogram signals and processing the electroencephalogram signals to obtain electroencephalogram time-frequency images;
the encoding processing unit is also connected with the modeling unit and is used for encoding the electroencephalogram time-frequency image to obtain a characteristic vector sequence;
the modeling unit is also connected with the classification unit and is used for modeling the characteristic vector sequence to obtain an output vector sequence;
the classification unit is used for classifying the output vector sequence to obtain a stage result of the output vector sequence.
Preferably, the signal acquisition unit adopts a head-wearing electrode electroencephalogram signal acquisition device.
The invention has the beneficial effects that:
at the epoch processing level, the present invention encodes the self-attention scores into heatmaps based on the transform backbone, providing interpretability for the model. At the sequence processing level, the present invention proposes a novel multi-branch strategy. The convolution network is used as a main branch to deeply mine the potential characteristics of the electroencephalogram signal, and compared with the traditional neural network such as the convolution neural network, the convolution network can better process data with topological structures such as the electroencephalogram signal. On the auxiliary branch, the invention adopts a bidirectional circulation neural network to extract the characteristics. And finally, fusing the outputs of the two branches by using a fusion strategy, calculating a final feature vector, and classifying the outputs by using a softmax layer to obtain a staging result of the sequence.
Experimental results on Sleep-EDF public data sets show that the Sleep staging method is superior to the existing Sleep staging method, and the overall accuracy is as high as 90%.
Drawings
In order to more clearly illustrate the technical solution of the present invention, the drawings needed to be used in the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without inventive labor.
Fig. 1 is a schematic flowchart of a sleep staging method for extracting sleep spectrogram features based on a GCN according to a first embodiment of the present invention;
FIG. 2 is a schematic diagram of a neural network structure of a sleep staging method for extracting sleep spectrogram features based on GCN according to the present invention;
fig. 3 is a schematic structural diagram of a sleep staging system for extracting sleep spectrogram features based on a GCN according to a third 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.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Example one
As shown in fig. 1, a schematic flow chart of the sleep staging method for extracting sleep spectrogram features based on GCN of the present invention includes the following steps:
s1, acquiring an electroencephalogram signal, and processing the electroencephalogram signal to obtain an electroencephalogram time-frequency image;
in the first embodiment, the acquired electroencephalogram signals are divided into L electroencephalogram signal segments with the duration of 30s, and the power spectrum of each electroencephalogram signal is calculated through short-time Fourier transform with the time window of 2 s; then, converting the power spectrum into a logarithmic power spectrum by a logarithmic scaling method; further obtaining a two-dimensional time-frequency image S, wherein S belongs to R F×T (ii) a Wherein F represents the total number of frequency intervals between all samples in the frequency domain, and T represents a spectrum column; in the first embodiment, F is 129, and T is 29.
S2, based on the time-frequency image, carrying out position coding processing to obtain a characteristic vector sequence;
processing the time-frequency image S into a sequence of T spectrum columns, and coding through sine and cosine functions; in this embodiment, the ith row, the 2j column and the (2j +1) th column are taken as examples:
wherein, P represents a coding matrix; f denotes the total number of frequency intervals between all samples in the frequency domain.
Then, the position information coding is carried out on the time frequency image S:
in the formula,
representing the time-frequency image after encoding the position information, P ep ∈R T×F Representing a position-coding matrix.
N Using Epochtransformer E The converter is used for coding, and the coding method comprises the following steps:
X (i) ＝EpochTransformer(X (i-1) ),X (i) ∈R T×F ，1≤i≤N E ，
wherein,
X (i) show that
The sequence of spectral columns of (a) is encoded.
To reduce the output of all transformers to a compact eigenvector representing epoch, the sequence is reduced
Obtaining the sequence (X) by weighted combination through an attention mechanism 1 ，…，X L )；The calculation method is as follows:
wherein X ∈ R F A derived feature vector representing the input epoch;
is the output sequence of Epochtransformer
One of (a), α t Is the attention weight learned by the softmax layer.
Wherein, the attention weight calculation formula is as follows:
a e ＝tanh(W a X t +b a ),W a ∈R A×F ,b a ∈R A ；
wherein, W a ∈R A×F And b a ∈R A Respectively learnable weight matrix and bias, a e ∈R A Is the trained epoch context vector. A represents the attention size.
S3, modeling based on the characteristic vector sequence to obtain an output vector sequence;
as shown in fig. 2, the present invention proposes a novel multi-branch strategy: the method adopts a multilayer graph convolution network as a main branch to deeply mine the potential characteristics of the electroencephalogram signal, and adopts a bidirectional circulation neural network as an auxiliary branch to extract the characteristics. The specific method comprises the following steps:
in the multi-layer graph convolution network part, attention feature vector sequence (X) 1 ，…，X L ) Is encoded as an output vector O G ＝(O G1 ，O G2 ，...，O GL ) (ii) a The method comprises the following steps:
(11) constructing a multilayer graph convolution network;
the construction method comprises the following steps:
wherein,
representing the addition of a self-connected adjacency matrix,
A adj an adjacency matrix for representing connection relation of adjacent parts in an undirected graph G, I N Is a matrix of characteristics of the image data,
and W (l) To train a matrix, superscript (l) indicates that the matrix belongs to layer l. σ () is the activation function of the network. H (l) ∈R N×D Is the activation matrix of layer I, the activation matrix of layer 0 is
In the first embodiment, a two-layer graph convolution network is used to model the attention feature sequence, and an adjacency matrix a is used adj Expressed as:
wherein, when i ═ j +1 or i ═ j-1, a ij 1, otherwise a ij ＝0。
(12) Computing adjacency matrices with added self-joins and normalization
Calculated according to the following formula:
in the formula,
representing a diagonal matrix.
(13) Calculating an output vector O according to a formula Gl The sequence of (a);
the calculation formula is as follows:
in the formula,
representing an adjacency matrix, W, added self-join and normalized (0) ∈R M×T The weight matrix is a weight matrix which can be learnt from an input layer to a hidden layer and is provided with T characteristic graphs; w (1) ∈R T×F It is a weight matrix that can be learned from the hidden layer to the output layer.
In the first embodiment, all residual layers share network parameters, and in order to obtain proper residual combination, a full-link layer network is used to combine X l Conversion to and O RL Vectors of the same structure.
In the bidirectional recurrent neural network part, attention feature vector sequence (X) 1 ,…,X L ) Is encoded as an output vector (O) R1 ,O R2 ,…,O RL ) Wherein L is more than or equal to 1 and less than or equal to L. Computing an output sequence O by computing its predecessor hidden state vector and back hidden state vector R 。
The method specifically comprises the following steps:
(21) calculating a forward hidden state vector:
(22) calculating a backward hidden state vector:
(23) calculating an output sequence:
in the formula,
is a learnable weight matrix;
represents a backward-hidden state vector and a forward-hidden state vector,
S4, classifying the output vector sequences based on the output vector sequences to obtain the stage result of the output vector sequences;
adding the output vectors of the two branches of the bidirectional cyclic neural network and the multilayer graph convolution network to obtain a sequence of the output vectors: o is l ＝O Rl +O Gl L is more than or equal to 1 and less than or equal to L; wherein L represents the number of electroencephalogram signal segments.
Output vector sequence O ═ O (O) 1 ，O 2 ，...，O L ) Inputting a shared softmax layer, wherein the shared softmax layer can effectively reduce parameters in the network, and calculating the probability value of the sleep stage corresponding to each sequence
The method for acquiring the sleep stage probability value corresponding to each sequence comprises the following steps: and sequentially calculating the probability value of each sequence for each sleep stage, and taking the maximum probability value as the sleep stage corresponding to the sequence.
For example: the probability values for a sequence of each sleep stage are: 0.6, 0.1 and 0.1, the maximum probability value 0.6 is taken as the probability value of the sleep stage corresponding to the sequence, and the sleep stage corresponding to the sequence is the waking period.
Example two
In order to judge the reliability of the constructed model, the invention introduces a loss function to calculate the loss of the model prediction result and the real result. The method comprises the following steps:
in one embodiment, the output vector sequence O ═ O (O) 1 ，O 2 ，...，O L ) After inputting a shared soft max layer, calculating the probability value of sleep stage corresponding to each sequence
When network training is performed, a sequence (S) is input 1 ，S 2 ，...，S L ) And a coded vector (y) corresponding thereto 1 ，y 2 ，...，y L ) As ground-truth, to evaluate the reliability of the model. When the loss is large, the parameter adaptability is adjusted. The loss function formula for the network is as follows:
where θ represents a parameter of the network and λ represents a trade-off error and l 2 Hyper-parameters of the non-regularization term.
EXAMPLE III
As shown in fig. 3, the present invention further provides a sleep staging system for extracting sleep spectrogram features based on GCN, including: the device comprises a signal acquisition unit, a coding processing unit, a modeling unit and a classification unit;
the signal acquisition unit is connected with the coding processing unit and is used for acquiring electroencephalogram signals and processing the electroencephalogram signals to obtain electroencephalogram time-frequency images;
the method for obtaining the electroencephalogram time-frequency image comprises the following steps:
carrying out short-time Fourier transform on the electroencephalogram signal to obtain a power spectrum;
transforming the power spectrum by adopting a logarithmic scaling method to obtain a logarithmic power spectrum; and obtaining the electroencephalogram time-frequency image based on the logarithmic power spectrum.
The encoding processing unit is also connected with the modeling unit and is used for encoding the electroencephalogram time-frequency image to obtain a characteristic vector sequence;
the method for obtaining the characteristic vector sequence comprises the following steps:
processing the time-frequency image S into a sequence of T spectrum columns, and coding through sine and cosine functions; take row i, column 2j, and column (2j +1) as examples:
wherein, P represents a coding matrix; f denotes the total number of frequency intervals between all samples in the frequency domain.
Then, the position information coding is carried out on the time frequency image S:
in the formula,
representing the time-frequency image after encoding the position information, P ep ∈R T×F Representing a position-coding matrix.
N Using Epochtransformer E The converter is used for coding, and the coding method comprises the following steps:
X (i) ＝EpochTransformer(X (i-1) ),X (i) ∈R T×F ，1≤i≤N E ，
wherein,
X (i) show that
The sequence of spectral columns of (a) is encoded.
To reduce the output of all transformers to a compact eigenvector representing epoch, the sequence is reduced
Obtaining the sequence (X) by weighted combination through an attention mechanism 1 ，…，X L ) (ii) a The calculation method is as follows:
wherein X ∈ R F A derived feature vector representing the input epoch;
is the output sequence of an Epoch Transformer
One of (a), α t Is the attention weight learned by the soft max layer.
Wherein, the attention weight calculation formula is as follows:
a e ＝tanh(W a X t +b a ),W a ∈R A×F ,b a ∈R A ；
wherein, W a ∈R A×F And b a ∈R A Respectively learnable weight matrix and bias，a e ∈R A Is the trained epoch context vector. A represents the attention size.
The modeling unit is also connected with the classification unit and is used for modeling the characteristic vector sequence to obtain an output vector sequence;
the method for obtaining the output vector sequence comprises the following steps:
the method adopts a multilayer graph convolution network as a main branch to deeply mine the potential characteristics of the electroencephalogram signal, and adopts a bidirectional circulation neural network as an auxiliary branch to extract the characteristics. The specific method comprises the following steps:
in the multi-layer graph convolution network part, attention feature vector sequence (X) 1 ，…，X L ) Is encoded as an output vector O G ＝(O G1 ，O G2 ，...，O GL ) (ii) a The method comprises the following steps:
constructing a multilayer graph convolution network;
the construction method comprises the following steps:
wherein,
representing the addition of a self-connected adjacency matrix,
A adj for the adjacency graph G, I N Is a matrix of characteristics that is,
and W (l) To train a matrix, superscript (l) indicates that the matrix belongs to layer l. σ () is the activation function of the network. H (l) ∈R N×D Is the activation matrix of layer I, the activation matrix of layer 0 is
Modeling attention feature sequence by using two-layer graph convolution network and adopting adjacency matrix A adj Expressed as:
wherein, when i ═ j +1 or i ═ j-1, a ij 1, otherwise a ij ＝0。
Computing a join matrix with added self-joins and normalization
Calculated according to the following formula:
in the formula,
representing a diagonal matrix.
Calculating an output vector O according to a formula Gl The sequence of (a);
the calculation formula is as follows:
in the formula,
representing the joining matrix W with the addition of self-joins and normalization (0) ∈R M×T The weight matrix is a weight matrix which can be learnt from an input layer to a hidden layer and is provided with T characteristic graphs; w is a group of (1) ∈R T×F It is a weight matrix that can be learned from the hidden layer to the output layer.
All residual layers share network parameters, and in order to obtain the proper residual combination, a full-link layer network is usedMixing X l Conversion to and O RL Vectors of the same structure.
In the bidirectional recurrent neural network part, attention feature vector sequence (X) 1 ,…,X L ) Is encoded as an output vector (O) R1 ,O R2 ,…,O RL ) Wherein L is more than or equal to 1 and less than or equal to L. Computing an output sequence O by computing its predecessor hidden state vector and back hidden state vector R 。
The method specifically comprises the following steps:
calculating a forward hidden state vector:
calculating a backward hidden state vector:
calculating an output sequence:
in the formula,
is a learnable weight matrix;
represents a backward-hidden state vector and a forward-hidden state vector,
represents a forward hidden state vector;
is a learnable offset.
The classification unit is used for classifying the output vector sequence to obtain the stage result of the output vector sequence.
The method for obtaining the stage result of the output vector sequence comprises the following steps:
adding the output vectors of the two branches of the bidirectional cyclic neural network and the multilayer graph convolution network to obtain a sequence of the output vectors: o is l ＝O Rl +O Gl L is more than or equal to 1 and less than or equal to L; wherein L represents the number of electroencephalogram signal segments.
Output vector sequence O ═ O (O) 1 ，O 2 ，...，O L ) Inputting a shared softmax layer, wherein the shared softmax layer can effectively reduce parameters in the network, and calculating the probability value of the sleep stage corresponding to each sequence
Wherein, the signal acquisition unit can adopt a head-wearing electrode electroencephalogram signal acquisition device.
The above-described embodiments are merely illustrative of the preferred embodiments of the present invention, and do not limit the scope of the present invention, and various modifications and improvements of the technical solutions of the present invention can be made by those skilled in the art without departing from the spirit of the present invention, and the technical solutions of the present invention are within the scope of the present invention defined by the claims.
Claims (8)
1. The sleep staging method for extracting the sleep spectrogram characteristics based on the GCN is characterized by comprising the following steps of:
s1, acquiring an electroencephalogram signal, and processing the electroencephalogram signal to obtain an electroencephalogram time-frequency image;
s2, carrying out position coding processing based on the electroencephalogram time-frequency image to obtain a characteristic vector sequence;
s3, modeling based on the characteristic vector sequence to obtain an output vector sequence;
and S4, classifying the output vector sequence based on the output vector sequence to obtain the stage result of the output vector sequence.
2. The sleep staging method for extracting sleep spectrogram features based on GCN as claimed in claim 1, wherein said S1 processing said EEG signal comprises:
s11, carrying out short-time Fourier transform on the electroencephalogram signal to obtain a power spectrum;
s12, transforming the power spectrum by adopting a logarithmic scaling method to obtain a logarithmic power spectrum; and obtaining the electroencephalogram time-frequency image based on the logarithmic power spectrum.
3. The sleep staging method for extracting sleep spectrogram features based on GCN as claimed in claim 1, wherein said position encoding method in S2 comprises: and (3) coding the electroencephalogram signals by adopting a Transformer as a main structure and combining an attention mechanism.
4. The sleep staging method for extracting sleep spectrogram features based on GCN as claimed in claim 1, wherein said modeling method in S3 comprises: and modeling the feature vector sequence by adopting a graph convolution network and a bidirectional recurrent neural network.
5. The sleep staging method based on GCN extraction sleep spectrogram feature of claim 4, wherein the output vector sequence classification in S4 employs a softmax function.
6. The sleep staging method for extracting sleep spectrogram features based on the GCN as claimed in claim 1, wherein said staging results in S4 correspond to five stages of sleep staging: awake phase, N1 phase, N2 phase, N3 phase and rapid eye movement phase.
7. Sleep staging system based on GCN draws sleep spectrogram characteristic, its characterized in that includes: the device comprises a signal acquisition unit, a coding processing unit, a modeling unit and a classification unit;
the signal acquisition unit is connected with the coding processing unit and is used for acquiring electroencephalogram signals and processing the electroencephalogram signals to obtain electroencephalogram time-frequency images;
the encoding processing unit is also connected with the modeling unit and is used for encoding the electroencephalogram time-frequency image to obtain a characteristic vector sequence;
the modeling unit is also connected with the classifying unit and is used for modeling the characteristic vector sequence to obtain an output vector sequence;
the classification unit is used for classifying the output vector sequence to obtain a stage result of the output vector sequence.
8. The GCN-based sleep stage classification system for extracting sleep spectrogram features according to claim 7, wherein said signal acquisition unit is a head-mounted electrode EEG signal acquisition device.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210687782.5A CN115054270A (en) | 2022-06-17 | 2022-06-17 | Sleep staging method and system for extracting sleep spectrogram features based on GCN |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210687782.5A CN115054270A (en) | 2022-06-17 | 2022-06-17 | Sleep staging method and system for extracting sleep spectrogram features based on GCN |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115054270A true CN115054270A (en) | 2022-09-16 |
Family
ID=83201860
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210687782.5A Pending CN115054270A (en) | 2022-06-17 | 2022-06-17 | Sleep staging method and system for extracting sleep spectrogram features based on GCN |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115054270A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115969329A (en) * | 2023-02-08 | 2023-04-18 | 长春理工大学 | Sleep staging method, system, device and medium |
| CN116898455A (en) * | 2023-07-06 | 2023-10-20 | 湖北大学 | Sleep electroencephalogram signal detection method and system based on deep learning model |
-
2022
- 2022-06-17 CN CN202210687782.5A patent/CN115054270A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115969329A (en) * | 2023-02-08 | 2023-04-18 | 长春理工大学 | Sleep staging method, system, device and medium |
| CN116898455A (en) * | 2023-07-06 | 2023-10-20 | 湖北大学 | Sleep electroencephalogram signal detection method and system based on deep learning model |
| CN116898455B (en) * | 2023-07-06 | 2024-04-16 | 湖北大学 | Sleep electroencephalogram signal detection method and system based on deep learning model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Parmar et al. | Learning to score olympic events | |
| CN115054270A (en) | Sleep staging method and system for extracting sleep spectrogram features based on GCN | |
| CN112766172B (en) | Facial continuous expression recognition method based on time sequence attention mechanism | |
| CN114564991B (en) | Electroencephalogram signal classification method based on transducer guided convolutional neural network | |
| CN111080032A (en) | Load prediction method based on Transformer structure | |
| CN110070895B (en) | Mixed sound event detection method based on factor decomposition of supervised variational encoder | |
| CN113423005B (en) | Intelligent music generation method and system based on improved neural network | |
| CN116486308A (en) | Teaching management system and method based on intelligent education | |
| CN113707331B (en) | Traditional Chinese medicine syndrome differentiation data generation method and system | |
| CN111626296A (en) | Medical image segmentation system, method and terminal based on deep neural network | |
| Luo et al. | Croup and pertussis cough sound classification algorithm based on channel attention and multiscale Mel-spectrogram | |
| CN110853764B (en) | Diabetes syndrome prediction system | |
| CN117333497A (en) | Mask supervision strategy-based three-dimensional medical image segmentation method for efficient modeling | |
| CN114626424B (en) | Data enhancement-based silent speech recognition method and device | |
| CN116570284A (en) | Depression recognition method and system based on voice characterization | |
| CN113361505B (en) | Non-specific human sign language translation method and system based on contrast decoupling element learning | |
| CN115456025A (en) | Electroencephalogram emotion recognition method based on layered attention time domain convolution network | |
| CN113553917A (en) | Office equipment identification method based on pulse transfer learning | |
| CN113269702A (en) | Low-exposure vein image enhancement method based on cross-scale feature fusion | |
| CN113096070A (en) | Image segmentation method based on MA-Unet | |
| CN114847968A (en) | Electroencephalogram sleep staging method based on long-term and short-term memory network | |
| Ting et al. | Performance analysis of single and combined bit-planes feature extraction for recognition in face expression database | |
| CN116738359B (en) | Mongolian multi-mode emotion analysis method based on pre-training model and high-resolution network | |
| CN118094139A (en) | Current load decomposition method based on time sequence deep analysis | |
| CN116756571A (en) | Behavior recognition model training method based on time-frequency fusion enhancement |
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 |