A sleep staging method and device based on brain connectivity features and domain adaptation

Abstract

The application discloses a sleep staging method and device based on brain connection characteristics and domain adaptation, wherein the device comprises a signal acquisition module, a sleep staging module and a local storage module. The signal acquisition module acquires multi-channel electroencephalogram signals. The sleep staging module comprises the following steps: 1. filtering and denoising the acquired signals, removing power frequency interference and performing independent component analysis to remove artifacts; 2. extracting correlation characteristics based on discrete wavelet transform and synchronization likelihood, and calculating energy proportion as a supplementary feature; 3. using a domain adaptation method to transform the features, thereby reducing the difference between the database data and the distribution of the collected data; 4. constructing a classification model, and predicting the sleep stage according to the transformed features. The local storage module records the electroencephalogram signals and the sleep staging results. The application can perform sleep staging based on multi-channel electroencephalogram, has good robustness, and has a positive significance for the diagnosis and treatment of sleep disorders.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of signal detection and medical electronics, and particularly relates to a sleep staging method and device based on brain connectivity features and domain adaptation. BACKGROUND

[0002] About one-third of a person's life is spent in sleep, so it is very important for us to have good sleep quality. Diseases such as insomnia and sleep disorders (such as sleep apnea syndrome) are very common, which can seriously affect the physical and mental state of patients. Poor sleep quality can affect different types of sleep-related diseases, such as sleep apnea syndrome, sleepiness, insomnia, depression, and cardiovascular disease. Sleep staging and sleep condition assessment are of great significance to the mental health and quality of life of human beings.

[0003] Sleep quality assessment is usually evaluated by the duration of sleep stages and the rate of change thereof. Polysomnography (PSG) technology is generally considered the gold standard for objective sleep stage division. PSG records various physiological signals, including electroencephalogram, electrocardiogram, electromyogram, and electrooculogram. Sleep staging based on the rhythm of electroencephalogram is a basic method for evaluating sleep quality, and the staging results are an important basis for diagnosis and subsequent treatment of sleep disorders. Generally, human experts perform sleep stage division by reading PSG data, which is very time-consuming because PSG contains many channels and records for a long time. It usually takes several hours to stage a night of PSG, and is prone to errors. In addition, PSG itself has problems such as inconvenience, high cost, and inconvenience to implement. For electroencephalogram signals, there are differences in age and disease conditions among different subjects, resulting in significant deviations between electroencephalogram signals. Considering the above problems, it is meaningful to design an automatic sleep staging system based on domain adaptation technology. SUMMARY

[0004] To solve the above problems, the present application provides a sleep staging method and device based on brain connectivity features and domain adaptation, which can achieve accurate emotion recognition through the collected multi-channel signals, thereby evaluating sleep quality and other indicators.

[0005] To achieve the above purpose, the present application provides the following technical solutions:

[0006] A sleep staging method based on brain connectivity features and domain adaptation, comprising the following steps:

[0007] S1: In the sleep staging process, multi-channel electroencephalogram signals are acquired, and the electroencephalogram signals are preprocessed, including: filtering and denoising, removing artifacts by independent component analysis, and removing baseline.

[0008] S2: Extracting sub-band from the processed electroencephalogram signal, including theta wave (4-8 Hz), alpha wave (8-13 Hz), beta wave (13-30 Hz), and gamma wave (30-100 Hz). Estimate the functional connection between each sub-band and different EEG channels corresponding to the brain region by synchronous likelihood algorithm, and extract the brain function connection feature.

[0009] S3: Discrete wavelet transform is performed on the electroencephalogram signal, db5 wavelet is selected, decomposition level is 5, different levels of wavelet coefficients are obtained, corresponding to different electroencephalogram rhythms. The Kendall correlation coefficient is used to calculate the correlation between the wavelet coefficients corresponding to different electroencephalogram channels, and the second group of brain function connection features are extracted; the energy ratio feature is calculated according to the wavelet coefficient as a supplement.

[0010] S4: In order to improve the generalization performance of the algorithm, the maximum independent domain adaptive algorithm based on clustering is used to transform the above extracted features, so as to reduce the difference between the database and the newly collected data distribution, and improve the performance of the classification algorithm.

[0011] S5: Use SVM to build a classification model to predict sleep stages, and save the results in the storage module.

[0012] Further, the features are extracted from the multi-channel electroencephalogram signal, the number of electroencephalogram channels is not fixed, and can depend on the specific collection device, generally not less than 3.

[0013] Further, in step S2, the commonly used low-frequency δ wave is excluded, and the γ wave is used as a supplementary frequency band to exclude the influence of volume conduction effect, wherein the δ wave is 0.5-4 Hz.

[0014] The brain function connection features include the following two kinds:

[0015] (1) WC: correlation based on discrete wavelet transform;

[0016] (2) SL: brain function connection based on synchronous likelihood;

[0017] The synchronous likelihood feature is obtained by the following method:

[0018] For a given time series x k (i), where k represents the kth channel, and i represents the serial number of the sample point.

[0019] The definition of the embedding vector X k (i) is:

[0020] X k (i) = [x k (i), x k (i+d), x k (i+2d), …, xk (i + (m - 1)d T

[0021] where d represents the time delay, m represents the embedding dimension, and T is the transpose operator. The synchronization likelihood S k,l (i) describes the signal X k (i) and the signal X l (i) of the kth channel, defined as:

[0022]

[0023] where ω1 and ω2 are two window parameters used to Taylor correct the autocorrelation effect and improve the temporal resolution, respectively. S k,l (i,j) represents the synchronization likelihood of the kth channel and the lth channel at each discrete time pair (i,j), calculated as:

[0024]

[0025] where H k,l (i,j)∈{0,1,2} indicates the embedding vector X k (i), X k (j), X l (i) and X l (j) that are less than the critical distance.

[0026] H k,l (i,j) is calculated as:

[0027]

[0028] where ε k (i) can be determined by setting where p ref is a predefined probability and the value is much smaller than 1.

[0029] Further, in step S3, the Kendall rank correlation coefficient is used to estimate the nonlinear correlation of wavelet coefficients. Based on the correlation and energy proportion characteristics of discrete wavelet transform, the following method is used to obtain:

[0030] The wavelet coefficients are calculated as:

[0031]

[0032] where the coefficients of level y contain the information of sub-band F y (2 -y-1 F s < F y <2 -y Fs ), where F s is the sampling frequency of the signal. The Kendall's tau correlation coefficient τ k (y,z) and C l (y,z) is calculated as follows: k,l,y

[0033]

[0034] where

[0035] The energy proportion is obtained as follows:

[0036]

[0037] Further, the step S4 specifically includes the following process:

[0038] The samples in the database are clustered by K-means, the marginal probability distribution and the conditional probability distribution of the data are estimated according to the clustering result, and the to-be-predicted sample is mapped into the source domain, the distance between the source domain and the target domain is calculated, so as to realize the selection of the source domain and the domain adaptation, which is realized by the following way:

[0039] Let the sample set of a domain be X = {x1, x2,..., x n}, and |·| represents the cardinality of the set. The samples in the set are divided into N k clusters by using the k-means algorithm, and the center position of each cluster is obtained. The marginal probability distribution is a distribution function about the features, and the MPD is estimated by calculating the ratio of the number of samples in each mode to the cardinality of the set, and the calculation method is as follows:

[0040]

[0041] where MPD k is the value of the MPD in the kth mode, and X k is the number of samples contained in the kth (k e {1, 2,..., N k}) mode. The calculation method of the CPD is as follows:

[0042]

[0043] The samples of the target domain are mapped into the source domain, for each sample x of the target domain, the nearest cluster center is selected, that is, After mapping, the cluster to which each sample in the target domain belongs is obtained, and the MPD of the target domain sample set is calculated, denoted as When there is no sample of the target domain in the kth cluster, ​When calculating the CPD, the classifier trained on the source domain is used to make predictions on the samples of the target domain, and the prediction results are used as the labels of the samples, and the CPD is calculated, denoted as denotes the proportion of samples of the target domain mapped to the kth cluster that belong to the cth class, and when no sample is mapped to the kth cluster, let to avoid numerical errors. satisfy the following conditions

[0044] Further, the step S5 specifically includes the following process:

[0045] After the extracted brain connection features are aligned by the domain adaptation algorithm, the selected source domain features are learned by using a nonlinear support vector machine; the target domain samples are subjected to the trained nonlinear support vector machine model to obtain the corresponding sleep staging. The nonlinear support vector machine maps the features extracted in steps S2-S4 to a high-dimensional space; wherein the kernel function adopts a Gaussian kernel; since the support vector machine does not support multi-classification, an error correction output code model is used to construct a multi-classifier, and for a 5-classification problem, 10 classifiers need to be constructed.

[0046] One or more embodiments provide a sleep staging device based on brain connection features and domain adaptation, comprising a signal acquisition module, a sleep staging module, and a local storage module. The sleep staging module comprises a signal preprocessing module, a feature extraction module, a domain adaptation module, and a classifier prediction module.

[0047] The signal acquisition module is configured to acquire multi-channel electroencephalogram signals.

[0048] The signal preprocessing module acquires the electroencephalogram signals from the signal acquisition module, and filters and denoises the acquired signals through a signal conditioning circuit and an MCU.

[0049] The feature extraction module extracts brain connection features from the multi-channel electroencephalogram signals after preprocessing for sleep staging.

[0050] The domain adaptation module aligns the database features and the target domain features by using a clustering-based maximum independent domain adaptation algorithm to improve the classification accuracy.

[0051] The classifier prediction module constructs a multi-classification model by using a nonlinear support vector machine and an error correction output code, and learns the training set samples to predict the sleep stage of new samples.

[0052] The local storage module records the acquired electroencephalogram signals and sleep staging results, and has the characteristics of long-time recording, large-capacity storage, and convenient data exchange with other devices.

[0053] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0054] The present application can realize accurate and robust automatic sleep staging based on multi-channel electroencephalogram signals. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 The sleep staging method based on brain connectivity features in the present application is shown in the flow chart.

[0056] Figure 2 The device example diagram of the sleep staging method based on brain connectivity features in the present application is shown in the flow chart.

[0057] Figure 3 The schematic diagram of the field adaptive module in the present application is shown in the flow chart.

[0058] Figure 4 The feature fusion diagram of the brain connectivity in the present application is shown in the flow chart. DETAILED DESCRIPTION

[0059] The present application is further described in detail below in conjunction with the accompanying drawings and implementation examples. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0060] It should be noted that the terms used herein are only for the purpose of describing the specific embodiments, and are not intended to limit the exemplary embodiments according to the present application. As used herein, the singular form is intended to include the plural form unless the context clearly indicates otherwise, and furthermore, it should be understood that when the terms "comprise" and / or "include" are used in the specification, there is a feature, step, operation, device, component and / or combination thereof.

[0061] Example 1

[0062] The present embodiment provides a sleep staging method based on brain connectivity features and field adaptation, as shown in Figure 1 The method comprises the following steps:

[0063] Step 1: Obtain multi-channel electroencephalogram signals from wearable measurement devices and other various forms of devices, including polysomnography; then pre-process the electroencephalogram signals, including: filtering and denoising, independent component analysis to remove artifacts, and removing baseline;

[0064] Step 2: Sub-band extraction is performed on the processed EEG signal, including theta waves (4-8 Hz), alpha waves (8-13 Hz), beta waves (13-30 Hz), and gamma waves (30-100 Hz). The functional connectivity between each sub-band and different EEG channels corresponding to brain regions is estimated using the synchronization likelihood algorithm, and brain functional connectivity features are extracted. For a given time series x k (i), where k (k ∈ {1, …, M}, M = 6) represents the kth channel, and i (i ∈ {1, …, N}, N = 30 × 200 = 6000) represents the serial number of the sample point. In this example, the signal sampling rate is 200 Hz, and the number of sample points is 6000.

[0065] The embedding vector X k (i) is defined as:

[0066] X k (i) = [x k (i), x k (i+d), x k (i+2d), …, x k (i+(m-1)d] T

[0067] where d represents the time delay, m represents the embedding dimension, and T is the transpose operator. The synchronization likelihood S k,l (i) at each time i describes the synchronization degree between the signal X k (i) of the kth channel and the signal X l (i) of the lth channel, which is defined as:

[0068]

[0069] where ω1 and ω2 are two window parameters used for Taylor correction of autocorrelation effects and improvement of time resolution, respectively. S k,l (i,j) represents the synchronization likelihood of channel k and channel l at each discrete time pair (i,j), which is calculated as:

[0070]

[0071] where H k,l (i,j) ∈ {0,1,2} represents the number of channels whose embedding vectors X k (i), X k (j), X l (i) and X l (j) are less than the critical distance. The calculation of H k,l (i,j) is as follows:

[0072]

[0073] where ε k (i) can be determined by letting where p ref is a predefined probability and has a value much smaller than 1, in this example p ref = 0.05 is chosen.

[0074] Step 3: Discrete wavelet transform is performed on the electroencephalogram signal, db5 wavelet is selected, and the decomposition level is 5, to obtain wavelet coefficients of different levels corresponding to different electroencephalogram rhythms. The Kendall correlation coefficient is used to calculate the correlation between the wavelet coefficients corresponding to different electroencephalogram channels, and a second group of brain function connection features is extracted; energy proportion features are calculated according to the wavelet coefficients as a supplement.

[0075] The calculation method of the wavelet coefficient is as follows:

[0076]

[0077] The coefficients of level y contain the information of the sub-band F y (2 -y-1 F s <F y <2 -y F s ), where F s is the sampling frequency of the signal.

[0078] For the y-level wavelet coefficients C k (y, z) and C l (y, z) of channel k and channel l, the Kendall tau correlation coefficient τ k,l,y is calculated as follows:

[0079]

[0080] where

[0081] The energy proportion is obtained as follows:

[0082]

[0083] Step 4: The maximum independent domain adaptive algorithm based on clustering is used to transform the features extracted above, to improve the performance of the classification algorithm, and the source domain selection process is shown in Figure 3 . Let the sample set of a domain be X = {x1, x2, …, x n}, and |·| represents the cardinality of the set. The k-means algorithm is used to divide the samples in the set into N k clusters, and 10 clustering centers are selected in this example.

[0084] The MPD is estimated by calculating the ratio of the number of samples and the cardinality of the set in each mode, and the calculation is as follows:

[0085]

[0086] where MPD k is the value of the MPD in the kth mode, X k is the number of samples contained in the kth (k∈{1,2,...,N k}) mode.

[0087] The calculation method of the CPD is as follows:

[0088]

[0089] Mapping the samples of the target domain to the source domain, for each sample x of the target domain, the nearest cluster center is selected, that is, After mapping, the cluster to which each sample in the target domain belongs is obtained, and the MPD of the target domain sample set is calculated, denoted as When there is no sample of the target domain in the kth cluster, When calculating the CPD, the classifier trained on the source domain is used to predict the samples of the target domain, the prediction result is used as the label of the sample, and the CPD is calculated, denoted as , which represents the proportion of samples of the target domain mapped to the kth cluster, and when there is no sample mapped to the kth cluster, let to avoid numerical errors. The following conditions are met

[0090] Step 5: A classification model is constructed using SVM, in this example, the SVM uses a Gaussian kernel, and the hyperparameter optimization uses a Bayesian optimization method, the trained classifier is used to predict the sleep stage, and the result is saved in the storage module.

[0091] Embodiment 2

[0092] The embodiment provides a sleep staging device based on brain connection features and domain adaptation, as shown in Figure 2 , which comprises:

[0093] A signal acquisition module 1 is used to acquire multi-channel electroencephalogram signals, the sampling frequency is 200Hz, the number of channels is 6, and the channels are F3-A2, C3-A2, O1-A2, F4-A1, C4-A1 and O2-A1 six channels.

[0094] The sleep staging module 2 acquires signals from the signal acquisition module 1, and performs sleep staging on the acquired electroencephalogram signals, and can be built into a mobile device or other terminal; comprising: a signal preprocessing module 21, a feature extraction module 22, a domain adaptation module 23, and a classifier prediction module 24.

[0095] The signal preprocessing module 21 acquires electroencephalogram signals from the signal acquisition module, filters and denoises the acquired electroencephalogram signals, removes artifacts through independent component analysis, and removes the baseline.

[0096] The feature extraction module 22 extracts brain connectivity features for sleep staging using the preprocessed multi-channel electroencephalogram signals. The functional connectivity between the corresponding brain regions of each sub-band and different EEG channels is estimated by a synchronization likelihood algorithm, and the brain functional connectivity features are extracted. The electroencephalogram signals are subjected to discrete wavelet transform, db5 wavelet is selected, and the decomposition level is 5, obtaining different levels of wavelet coefficients corresponding to different electroencephalogram rhythms. The Kendall correlation coefficient is used to calculate the correlation between the wavelet coefficients corresponding to different electroencephalogram channels, and the second group of brain functional connectivity features are extracted; and the energy proportion feature is calculated based on the wavelet coefficients as a supplement.

[0097] The domain adaptation module 23 aligns the database features and the target domain features by a clustering-based maximum independent domain adaptation algorithm, and the number of cluster centers is set to 10. The domain background information is encoded in one-hot encoding mode, the independence of sample features and domain information is maximized to solve the transformation matrix, and the trace of the matrix is maximized to retain the variance properties of the features, and the Lagrange multiplier method is used to solve the objective function.

[0098] The classifier prediction module 24 uses a nonlinear support vector machine and an error correction output code to construct a multi-classification model, and the classifier learns the training set samples to predict the sleep stage of new samples. The SVM uses a Gaussian kernel, and the kernel hyperparameter and the regularization parameter are optimized by the Bayesian optimization method.

[0099] The local storage module 3 records the six-channel electroencephalogram signals and the sleep staging prediction results in detail, has the characteristics of long-time recording, large-capacity storage, and convenient data exchange with other devices, and assists patients in preventing and treating sleep disorders, and provides doctors with the basis for sleep disorder diagnosis;

[0100] The technical features of the above embodiments can be combined in any way. To make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combinations of the technical features do not exist, they should be considered as the scope of the description.

[0101] It should be noted that the terms "first", "second", "third" involved in the embodiments of the present application are only to distinguish similar objects, and do not represent a specific order for the objects. Understandably, "first", "second", "third" can be interchanged in a specific order or sequence as appropriate. It should be understood that the objects distinguished by "first", "second", "third" can be interchanged under appropriate circumstances, so that the embodiments of the application described herein can be implemented in an order other than those illustrated or described herein.

[0102] The terms "comprising" and "having" and any variations thereof in the exemplary embodiments of the present application are intended to cover non-exclusive inclusion. For example, a process, method, device, product or equipment including a series of steps or modules is not limited to the listed steps or modules, but optionally further includes steps or modules not listed, or optionally further includes other steps or modules inherent to the process, method, product or equipment.

[0103] The above-described exemplary embodiments only express several embodiments of the present application, which are described in detail and in detail, but cannot be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application. Therefore, the scope of protection of the present application patent should be subject to the appended claims.

Claims

1. A sleep staging method based on brain connectivity features and domain adaptation, characterized in that, Comprise the following steps: S1: In the sleep staging process, acquire multi-channel electroencephalogram signals, and pretreat the electroencephalogram signals, including: filtering and denoising, removing artifacts by independent component analysis, and removing baseline; S2: Extract sub-band of the processed electroencephalogram signals, including theta wave, alpha wave, beta wave and gamma wave, wherein the theta wave is 4-8 Hz, the alpha wave is 8-13 Hz, the beta wave is 13-30 Hz, and the gamma wave is 30-100 Hz; estimate the functional connection between each sub-band and different EEG channels corresponding to the brain region by synchronous likelihood algorithm, and extract brain functional connection features; S3: Discrete wavelet transform is performed on the electroencephalogram signals, db5 wavelet is selected, decomposition level is 5, different level wavelet coefficients are obtained, corresponding to different electroencephalogram rhythms; the Kendall correlation coefficient is used to calculate the correlation between the wavelet coefficients corresponding to different electroencephalogram channels, and a second group of brain functional connection features are extracted; S4: In order to improve the generalization performance of the algorithm, the maximum independent domain adaptive algorithm based on clustering is used to transform the extracted features, so as to reduce the difference between the database and the newly collected data distribution, and improve the performance of the classification algorithm; S5: A classification model is constructed by using SVM to predict the sleep stage, and the result is saved in the storage module. The domain adaptation method in step S4 is to estimate the conditional probability distribution and marginal probability distribution of the samples by K-means clustering, and measure the distance between the source domain and the target domain; it is realized by the following process: Let a sample set of a domain be , represent the cardinality of the set; use the k-means algorithm to divide the samples in the set into clusters and obtain the center position of each cluster; the edge probability distribution is a distribution function about the features, and the MPD is estimated by calculating the ratio of the number of samples under each mode to the cardinality of the set, and the calculation method is: wherein is the value of the first mode under the second mode, is the number of samples contained in the first mode; and the CPD is calculated by: wherein is the proportion of samples in the first category in the second mode, is the set of samples in the first category in the second mode, is the set of samples in the second category in the second mode.​ 2. The sleep staging method based on brain connectivity features and domain adaptation according to claim 1, characterized in that, The number of electroencephalogram channels is not fixed and depends on the specific acquisition device, which is not less than 3.

3. The sleep staging method based on brain connectivity features and domain adaptation according to claim 1, characterized in that: In step S2, the commonly used low-frequency δ wave is excluded, and the γ wave is used as a supplementary frequency band to exclude the influence of volume conduction effect, wherein the δ wave is 0.5-4 Hz.

4. The sleep staging method based on brain connectivity features and domain adaptation according to claim 1, characterized in that: In step S3, the Kendall rank correlation coefficient is used to estimate the nonlinear correlation of the wavelet coefficients.

5. The sleep staging method based on brain connectivity features and domain adaptation according to claim 1, characterized in that: The brain functional connection features include the following two kinds: (1) WC: correlation based on discrete wavelet transform; (2) SL: brain functional connection based on synchronous likelihood; And obtained by the following way: For a given time series wherein denotes the k-th channel, i denotes the index of the sample point; Embedding vectors is defined as: Where d represents the time delay, m represents the embedding dimension, and T is the transpose operator; Synchronization likelihood at each time i , describes the synchronization of the signal of the kth channel and the signal of the lth channel is defined as: where and are two window parameters, respectively used to Taylor correct the autocorrelation effect and to improve the time resolution; represents the synchronization likelihood of channel k and channel l at each discrete time pair is calculated as wherein ∈ {0,1,2} denotes an embedding vector and a number of channels less than a critical distance; The calculation method is as follows: ， wherein is determined by letting wherein is a predefined probability and the value is much smaller than 1; The calculation method of the wavelet coefficient is: where the coefficients of level y contain information of the sub-band , where is the sampling frequency of the signal; the wavelet coefficients of level y for channel k and channel l and , the Kendall's tau correlation coefficient is calculated as:​ wherein ; The supplementary features include the energy proportion, and are obtained by the following way: 。 6. The sleep staging method based on brain connectivity features and domain adaptation according to claim 1, characterized in that: The step S5 specifically Comprise the following process: After the extracted brain connection features are aligned by the domain adaptation algorithm, the selected source domain features are learned by using a nonlinear support vector machine; The target domain samples are subjected to the trained nonlinear support vector machine model to obtain the corresponding sleep staging.

7. The sleep staging method based on brain connectivity features and domain adaptation according to claim 6, characterized in that: The nonlinear support vector machine maps the features extracted in steps S2-S4 to a high-dimensional space; wherein the kernel function adopts Gaussian kernel; since the support vector machine does not support multi-classification, an error correction output code model is used to construct a multi-classifier, and for a 5-classification problem, 10 classifiers need to be constructed.

8. A sleep staging device based on brain connectivity features and domain adaptation, configured to implement the method of any one of claims 1-7. Comprise: Signal acquisition module, sleep staging module, local storage module; The sleep staging module comprises: signal preprocessing module, feature extraction module, domain adaptation module, and classifier prediction module; The signal acquisition module is used for multi-channel electroencephalogram signals. The signal preprocessing module acquires the electroencephalogram signals from the signal acquisition module, filters and denoises the acquired signals through a signal conditioning circuit and an MCU. The feature extraction module extracts brain connection features from the multi-channel electroencephalogram signals after preprocessing for sleep staging. The domain adaptation module aligns the database features and the target domain features through a clustering-based maximum independent domain adaptation algorithm to improve the classification accuracy. The classifier prediction module constructs a multi-classification model by using a nonlinear support vector machine and an error correction output code, and learns the training set samples to predict the sleep stage of a new sample. The local storage module records the acquired electroencephalogram signals and sleep staging results, has the characteristics of long-time recording, large-capacity storage and convenient data exchange with other devices.

Images