Cognitive state deep learning classification method, device and equipment based on electroencephalogram signals, medium and product

Abstract

The invention discloses a cognitive state deep learning classification method, device and equipment based on an electroencephalogram signal, a medium and a product, and relates to the field of biological feature recognition, and the method comprises the steps: obtaining an original electroencephalogram signal; the original electroencephalogram signal comprises a plurality of channel signals; dividing the original electroencephalogram signal in a frequency domain to obtain a plurality of frequency bands, and in each frequency band, according to phase synchronism between every two channel signals of the frequency band, obtaining in-band characteristics corresponding to each frequency band; fusing the intra-band features of all the bands to obtain fused features; according to the fusion features, a cognitive state classification model is adopted, and the cognitive state corresponding to the original electroencephalogram signals is determined. According to the method, the characteristics in the frequency bands are obtained by analyzing the internal relation among different channel signals, and the final fusion characteristics are obtained by further carrying out characteristic fusion among the frequency bands, so that the electroencephalogram signal characteristic extraction method is optimized, and the accuracy of cognitive state classification is improved.

Description

Technical Field

[0001] The present application relates to the field of biometric recognition, and in particular, to a deep learning classification method, device, equipment, medium and product for cognitive state based on electroencephalogram signals. Background Art

[0002] Dementia caused by cognitive dysfunction, which is manifested by severe impairment of cognitive functions such as memory, language, reasoning and spatial orientation, will seriously affect personal autonomy and bring a huge economic burden to society, and has become a global concern. It is estimated that by 2050, the number of people suffering from cognitive dysfunction will reach 152 million. Mild cognitive impairment is considered to be the early stage of dementia, between normal cognition and dementia. Research shows that about 6% to 25% of patients with mild cognitive impairment develop into dementia every year. Dementia symptoms are irreversible, but effective intervention measures taken in the mild cognitive impairment stage can delay cognitive decline and improve the quality of life of patients. Therefore, it is very important to classify and evaluate the cognitive state at an early stage.

[0003] In recent years, advanced neuroimaging techniques for early classification and evaluation of cognitive states have become increasingly feasible. These techniques include functional magnetic resonance imaging, positron emission tomography, magnetoencephalography, and functional near-infrared spectroscopy. Although the above techniques can capture subtle changes related to brain activity, their clinical applications are limited by high costs and time consumption. In contrast, electroencephalogram, as a non-invasive, economical and portable technology, can effectively reflect the dynamic changes of brain activity by placing electrodes on the scalp and detecting the electrical signals generated by thousands of neurons. However, in practical applications, the existing electroencephalogram-based detection systems still have the following deficiencies: the classification performance of the feature classification method depends to a large extent on feature engineering and feature selection under the condition of sufficient prior knowledge; the current deep learning models used for cognitive state classification and evaluation are good at capturing the physical space information of the brain, but often ignore the internal connections between different brain regions (channels). And the cognitive process often involves the coordination of multiple brain regions, and different brain regions are responsible for different functions. Therefore, obtaining the internal connections between multiple channels is also important for cognitive state classification and evaluation. Summary of the Invention

[0004] The purpose of the present application is to provide a deep learning classification method, device, equipment, medium and product for cognitive state based on electroencephalogram signals, which can analyze the internal connections between signals of different channels, obtain the features within the frequency band, further perform feature fusion between frequency bands, obtain the final fusion features, optimize the electroencephalogram signal feature extraction method, and improve the accuracy of cognitive state classification.

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

[0006] In a first aspect, the present application provides a deep learning classification method for cognitive states based on electroencephalogram (EEG) signals, including:

[0007] Obtain the original EEG signal; the original EEG signal includes multiple channel signals;

[0008] Divide the original EEG signal in the frequency domain to obtain multiple frequency bands, and within each frequency band, obtain the intra-band features corresponding to each frequency band according to the phase synchrony between the channel signals of the frequency band in pairs;

[0009] Fuse the intra-band features of all frequency bands to obtain fused features;

[0010] According to the fused features, use a cognitive state classification model to determine the cognitive state corresponding to the original EEG signal; the cognitive state classification model is a neural network model pre-built according to a training sample set; the training sample set includes the feature information of multiple sample EEG signals and the cognitive state corresponding to each sample EEG signal.

[0011] Optionally, before dividing the original EEG signal in the frequency domain to obtain multiple frequency bands, and within each frequency band, obtaining the intra-band features corresponding to each frequency band according to the phase synchrony between the channel signals of the frequency band in pairs, the deep learning classification method for cognitive states based on EEG signals further includes:

[0012] Perform downsampling, band-pass filtering, and independent component analysis processing on the original EEG signal in sequence.

[0013] Optionally, dividing the original EEG signal in the frequency domain to obtain multiple frequency bands, and within each frequency band, obtaining the intra-band features corresponding to each frequency band according to the phase synchrony between the channel signals of the frequency band in pairs specifically includes:

[0014] Divide the original EEG signal in the time domain to obtain multiple time periods;

[0015] For any time period, divide the time period in the frequency domain to obtain multiple frequency bands of the time period, and within each frequency band of the time period, calculate the phase synchrony between the channel signals of the frequency band in pairs to obtain the channel pair features of each frequency band of the time period;

[0016] According to the channel pair features of all frequency bands of all time periods, use a deep learning algorithm to obtain the intra-band features corresponding to each frequency band.

[0017] Optionally, the deep learning algorithm is a multi-head attention mechanism.

[0018] Optionally, the method for building the cognitive state classification model includes:

[0019] Obtain multiple sample electroencephalogram (EEG) signals; each sample EEG signal contains multiple channel signals;

[0020] For any sample EEG signal, divide the sample EEG signal in the frequency domain to obtain multiple sample frequency bands, and within each sample frequency band, obtain the in-band features corresponding to each sample frequency band according to the phase synchrony between the channel signals of the sample frequency band in pairs;

[0021] For any sample EEG signal, fuse the in-band features of all the sample frequency bands of the sample EEG signal to obtain the feature information of the sample EEG signal;

[0022] Train a neural network model according to each feature information and the cognitive state corresponding to each feature information to obtain the cognitive state classification model.

[0023] Optionally, for any sample EEG signal, divide the sample EEG signal in the frequency domain to obtain multiple sample frequency bands, and within each sample frequency band, obtain the in-band features corresponding to each sample frequency band according to the phase synchrony between the channel signals of the sample frequency band in pairs, specifically including:

[0024] Divide the sample EEG signal in the time domain to obtain multiple sample time periods;

[0025] For any sample time period, divide the sample time period in the frequency domain to obtain multiple sample frequency bands of the sample time period, and within each sample frequency band of the sample time period, calculate the phase synchrony between the channel signals of the sample frequency band in pairs to obtain the channel pair features of each sample frequency band of the sample time period;

[0026] According to the channel pair features of all the sample frequency bands of all the sample time periods, use a deep learning algorithm to obtain the in-band features corresponding to each sample frequency band.

[0027] In a second aspect, the present application provides a deep learning classification device for cognitive state based on EEG signals, which is applied to the deep learning classification method for cognitive state based on EEG signals described in any one of the above, and includes:

[0028] A signal acquisition module, configured to acquire an original EEG signal; the original EEG signal contains multiple channel signals;

[0029] An in-band feature extraction module, configured to divide the original EEG signal in the frequency domain to obtain multiple frequency bands, and within each frequency band, obtain the in-band features corresponding to each frequency band according to the phase synchrony between the channel signals of the frequency band in pairs;

[0030] A fusion feature extraction module, configured to fuse the in-band features of all frequency bands to obtain fused features;

[0031] A cognitive state classification module, configured to determine the cognitive state corresponding to the original EEG signal according to the fused features by using a cognitive state classification model; the cognitive state classification model is a neural network model pre-established according to a training sample set; the training sample set includes the feature information of multiple sample EEG signals and the cognitive state corresponding to each sample EEG signal.

[0032] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the above-mentioned deep learning classification method for cognitive state based on EEG signals.

[0033] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the above-mentioned deep learning classification method for cognitive state based on EEG signals.

[0034] In a fifth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the above-mentioned deep learning classification method for cognitive state based on EEG signals.

[0035] According to the specific embodiments provided by the present application, the present application has the following technical effects:

[0036] The present application provides a deep learning classification method, device, device, medium and product for cognitive state based on EEG signals. By dividing the original EEG signal in the frequency domain and analyzing the phase synchrony between different channel signals in different frequency bands as in-band features, the internal connection between different brain regions is obtained; and further by performing inter-band feature fusion on the in-band features of different frequency bands, the final fused features are obtained. The fused features consider both in-band features and inter-band features, optimize the EEG signal feature extraction method, and improve the accuracy of cognitive state classification. Description of the Drawings

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0038] Figure 1 It is an application environment diagram of a deep learning classification method for cognitive state based on EEG signals in an embodiment of the present application;

[0039] Figure 2 Schematic flow chart of a deep learning classification method for cognitive state based on electroencephalogram (EEG) signals provided by an embodiment of the present application;

[0040] Figure 3 Schematic flow chart of the processing of raw EEG signals provided by an embodiment of the present application;

[0041] Figure 4 Schematic overall framework diagram of a deep learning classification method for cognitive state based on EEG signals provided by an embodiment of the present application;

[0042] Figure 5 Schematic flow chart of the extraction of in-band features within the frequency band of EEG signals for any time period provided by an embodiment of the present application;

[0043] Figure 6 Schematic flow chart of the method for building a cognitive state classification model provided by an embodiment of the present application;

[0044] Figure 7 Schematic diagram of the functional modules of a deep learning classification device for cognitive state based on EEG signals provided by an embodiment of the present application;

[0045] Figure 8 Schematic diagram of the structure of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0046] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0047] The present application proposes a deep learning classification method, device, equipment, medium and product for cognitive state based on EEG signals. By dividing the raw EEG signals in the frequency domain and analyzing the phase synchronization between signals of different channels within different frequency bands as in-band features, the internal connections between different brain regions are obtained; and further, by performing inter-band feature fusion on the in-band features of different frequency bands, the final fusion features are obtained. The fusion features consider both in-band features and inter-band features, optimize the method for extracting EEG signal features, and improve the accuracy of cognitive state classification.

[0048] To make the above objects, features and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0049] The deep learning classification method for cognitive state based on electroencephalogram (EEG) signals provided by the embodiments of the present application can be applied to an application environment as shown in Figure 1 the following figure. Among them, the terminal 102 communicates with the server 104 through a network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the original EEG signals to the server 104. After receiving the original EEG signals, the server 104 divides the original EEG signals in the frequency domain to obtain multiple frequency bands, and within each frequency band, according to the phase synchrony between the channel signals of the frequency band, obtains the in-band features corresponding to each frequency band. The in-band features of all frequency bands are fused to obtain the fused features. According to the fused features, a cognitive state classification model is used to determine the cognitive state corresponding to the original EEG signals. The server 104 can feedback the cognitive state corresponding to the original EEG signals to the terminal 102. In addition, in some embodiments, the deep learning classification method for cognitive state based on EEG signals can also be implemented independently by the server 104 or the terminal 102. For example, the terminal 102 can directly classify and evaluate the cognitive state corresponding to the original EEG signals, or the server 104 can obtain the original EEG signals from the data storage system and classify and evaluate the cognitive state corresponding to the original EEG signals.

[0050] Among them, the terminal 102 can be, but is not limited to, various desktop computers, laptop computers, smartphones, tablet computers, Internet of Things (IoT) devices, and portable wearable devices. The IoT devices can be smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. The portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server 104 can be implemented by an independent server or a server cluster composed of multiple servers, and can also be a cloud server.

[0051] In an exemplary embodiment, as shown in Figure 2 the following figure, a deep learning classification method for cognitive state based on EEG signals is provided. This method is executed by a computer device, and specifically can be executed independently by a computer device such as a terminal or a server, or jointly executed by a terminal and a server. In the embodiments of the present application, taking this method applied to the server 104 in Figure 1 as an example for illustration, it includes the following steps 201 to 204. Among them:

[0052] Step 201, obtain the original EEG signals; the original EEG signals include multiple channel signals.

[0053] Step 202: Divide the original EEG signal in the frequency domain to obtain multiple frequency bands, and within each frequency band, obtain the intra-band features corresponding to each frequency band according to the phase synchrony between the channel signals of the frequency band in pairs.

[0054] Step 203: Fuse the intra-band features of all frequency bands to obtain the fused features.

[0055] Step 204: According to the fused features, use the cognitive state classification model to determine the cognitive state corresponding to the original EEG signal; the cognitive state classification model is a neural network model built in advance according to the training sample set; the training sample set includes the feature information of multiple sample EEG signals and the cognitive state corresponding to each sample EEG signal.

[0056] In an exemplary embodiment, as Figure 3 shown, before the above step 202, the deep learning classification method for the cognitive state based on EEG signals further includes: successively performing downsampling, band-pass filtering, and independent component analysis on the original EEG signal.

[0057] The purpose of downsampling is to reduce the computational amount while retaining sufficient signal details for subsequent analysis and processing. After downsampling, the time resolution of the original EEG signal decreases, but the main information in the signal can still be effectively retained, especially in the target frequency range. In this embodiment, the original EEG signal is downsampled to 200 Hz.

[0058] Band-pass filtering can remove the low-frequency drift (such as respiration or heartbeat artifacts) and high-frequency noise (such as power supply noise, etc.) in the original EEG signal by setting appropriate frequencies, while retaining the frequency band information crucial for the EEG activities related to the cognitive state. The common EEG signal activities are mainly distributed in the frequency band of 0.1 Hz - 45 Hz, including the θ, α, β, and γ bands related to brain activities. The filtered signal will have a higher signal-to-noise ratio, which is beneficial for subsequent feature extraction and analysis.

[0059] Independent component analysis is a blind signal source separation method that can decompose complex EEG signals into multiple independent components. Through independent component analysis processing, it is possible to effectively separate the electrooculogram artifacts (such as artifacts caused by blinking and eye movement) and other possible physiological artifacts (such as electromyogram artifacts), thereby ensuring that the EEG activity components in the signal are more pure. The application of independent component analysis significantly improves the quality of the EEG signal, providing a more reliable basis for subsequent functional connectivity analysis and feature extraction.

[0060] Through the above three-step processing of the original EEG signal, clear and clean signal data are provided for subsequent analysis steps, thereby improving the accuracy and reliability of the research results.

[0061] In an exemplary embodiment, Figure 2 Step 202 in

[0062] can be replaced by the following steps:

[0063] Divide the original EEG signal in the time domain to obtain multiple time segments.

[0064] For any time segment, divide the time segment in the frequency domain to obtain multiple frequency bands of the time segment, and calculate the phase synchrony between every two channel signals within each frequency band of the time segment to obtain the channel pair features of each frequency band of the time segment.

[0065] In an exemplary embodiment, as Figure 4 shown, divide the original EEG signal into N non-overlapping time segments, each with an appropriate length. For any time segment, the in-band feature extraction process is as Figure 4 and Figure 5 shown. Using frequency band decomposition techniques (such as wavelet transform, fast Fourier transform, etc.), divide the EEG signal corresponding to the time segment into four frequency bands in the frequency domain. The four frequency bands are: θ (4H Z -7H Z ), α (8H Z -13H Z ), β (14H Z -30H Z ), and γ (31H Z -45H Z ). These frequency bands cover the activities of different frequency ranges of the brain, and each frequency band plays a different role in cognitive functions.

[0066] Cross-spectral density is a frequency domain measurement method that can quantify the mutual relationship between two signals of different frequencies, thereby revealing the synchronization strength between them. In this embodiment, calculate the Phase Locking Value (PLV) according to the cross-spectral density, and determine the phase synchrony between every two channel signals within the frequency band according to the value of PLV. In this way, a deeper understanding of the functional connections between different channels in different frequency bands can be obtained, not limited to time domain synchronization measurement. If the PLV value of each pair of channels is close to 1, it indicates that the phase synchrony between the signals remains consistent over time, reflecting strong functional connectivity. On the contrary, if the PLV value is close to 0, it indicates that the phase relationship between the signals changes inconsistently over time, indicating weak functional connectivity between the two channels.

[0067] Within each frequency band, taking any two different channel signals X and Y as an example. First, X and Y are each divided into multiple frames to facilitate averaging and improve the accuracy of calculations. The cross-spectral density of channel signals X and Y at the k-th frame can be expressed as where f is the current frequency band; e is the base of the natural logarithm; i is the imaginary unit; A(f) is the amplitude component of the cross-spectral density of channel signals X and Y at the k-th frame; θ k (f) is the angular component of the cross-spectral density of channel signals X and Y at the k-th frame.

[0068] The cross-spectral density contains the frequency-dependent relationship between X and Y, including amplitude and phase information. To obtain the PLV, only the phase information in the cross-spectral density is needed, that is, the angular component of the cross-spectral density. Within the current frequency band f, the PLV value of channel signals X and Y can be obtained by the following formula:

[0069]

[0070] where f is the current frequency band; e is the base of the natural logarithm; i is the imaginary unit; PLV is the phase-locking value of channel signals X and Y in the current frequency band; θ k (f) is the angular component of the cross-spectral density of channel signals X and Y at the k-th frame; K is the total number of frames included in channel signals X and Y.

[0071] By calculating the phase synchrony between each pair of channel signals within each frequency band in turn, that is, the PLV value, a connectivity matrix with a shape of (4, ch, ch) is obtained, where "4" are the four frequency bands in this embodiment, and ch is the number of channels included in the original EEG signals. Due to the symmetric property of the connectivity matrix, for each frequency band, the lower triangular part of its corresponding connectivity matrix is extracted to obtain a matrix with a shape of (1, d), where d is the total number of all elements included in the lower triangular part of the connectivity matrix. For all four frequency bands in this embodiment, a matrix with a shape of (4, d) is obtained. Further, as Figure 4 and Figure 5 shown, for all N time periods, within each frequency band, a feature sequence with a shape of (N, d) is obtained, which is the channel pair feature corresponding to the frequency band. The channel pair features of all frequency bands form a feature sequence with a shape of (N, 4, d).

[0072] In an exemplary embodiment, according to the channel pair features of all frequency bands of all time periods, a multi-head attention mechanism is adopted to obtain the intra-band features corresponding to each frequency band.

[0073] The channel pair features across all frequency bands form a feature sequence shaped like (N, 4, d). By converting the frequency band dimension and the sequence dimension, it is transformed into (4, N, d). In this way, the feature sequences for each frequency band are obtained, each shaped like (N, d). Then, based on the feature sequences for each frequency band, a multi-head attention mechanism is adopted to obtain the intra-band features for the corresponding frequency band. The design purpose of the multi-head attention mechanism is to assign different degrees of importance to different channel pair features and integrate this information to construct the final representation of the current frequency band, thereby promoting the learning of intra-band information.

[0074] In this embodiment, first, a learnable classification token is added to each input sequence. Then, a multi-head attention mechanism is used to capture the global temporal dependence of the feature sequence, enabling the final model to simultaneously focus on different segments of the input and learn more complex relationships between the encoded feature sequences. This process includes projecting the input features into h different subspaces and independently processing each subspace through the attention mechanism. The intra-band features across all frequency bands are fused to obtain the inter-band mapping result, that is, the fused features. This step transforms the extracted features into the final attention representation through feature fusion, promoting the integration of complementary information between frequency bands. In this embodiment, the intra-band features of the four frequency bands are concatenated, and then the concatenated feature vectors are linearly processed to obtain the fused features.

[0075] In an exemplary embodiment, as Figure 6 shown, the method for building a cognitive state classification model includes:

[0076] Step 601, obtain multiple sample EEG signals; each sample EEG signal contains multiple channel signals.

[0077] Step 602, for any sample EEG signal, divide the sample EEG signal in the frequency domain to obtain multiple sample frequency bands, and within each sample frequency band, obtain the intra-sample-frequency-band features corresponding to each sample frequency band according to the phase synchrony between the channel signals of the sample frequency band.

[0078] Step 603, for any sample EEG signal, fuse the intra-sample-frequency-band features of all sample frequency bands of the sample EEG signal to obtain the feature information of the sample EEG signal.

[0079] Step 604, train a neural network model according to each feature information and the cognitive state corresponding to each feature information to obtain a cognitive state classification model.

[0080] In an exemplary embodiment, the above step 602 can be replaced by the following steps:

[0081] Divide the sample EEG signal in the time domain to obtain multiple sample time periods.

[0082] For any sample period, divide the sample period in the frequency domain to obtain multiple sample frequency bands of the sample period, and within each sample frequency band of the sample period, calculate the phase synchronization between every two channel signals of the sample frequency band to obtain the channel pair features of each sample frequency band of the sample period.

[0083] According to the channel pair features of all sample frequency bands of all sample periods, use a deep learning algorithm to obtain the in-band features corresponding to each sample frequency band.

[0084] In an exemplary embodiment, as Figure 4 shown, the cognitive state classification model of the present application consists of two fully connected layers, aiming to learn the complex decision boundaries between different cognitive states. To reduce the risk of overfitting, dropout regularization is applied between these layers, thereby ensuring that the model has strong generalization ability. In addition, the exponential linear unit is used as the activation function in the fully connected layer, and this function can improve the training dynamics by alleviating the vanishing gradient problem and accelerating the convergence speed. The above design enables the cognitive state classification model of the present application to effectively map the learned features to the final classification decision, thereby improving the overall performance of the model in the cognitive state classification task.

[0085] In an exemplary embodiment, the present application verifies the performance of the cognitive state classification model in two scenarios: cross-subject and cross-session. For the above two scenarios, the model performance verification has different data definitions and dataset partitions.

[0086] Assume that there are S subjects, and each subject is divided into D different session (period) experiments. The entire sample set can be expressed as where l is the subject number, j is the session (period) number, B l is the sample of the l-th subject, and C j is the sample label corresponding to the j-th session. The sample label set corresponding to the subject sample set includes different cognitive state categories, specifically including the healthy control group, the mild cognitive impairment group, and the dementia group.

[0087] For both the cross-subject and cross-session scenarios, the leave-one-out method is used to perform cross-validation on the dataset.

[0088] For the cross-session cognitive state classification task, in each subject, the data of the last cognitive trial of all subjects in the session is used as the test set; the remaining D - 1 sessions are used as source domains in the training set, with each session as a source domain in the training set, and finally D - 1 source domains are obtained as the training set. A total of D experiments are conducted and the average accuracy is calculated.

[0089] For the cross-subject cognitive state classification task, in each experiment, all the data of D cognitive trials of one subject are iteratively taken out and assumed that the cognitive state label is unknown, as the test set; then, from the data of the remaining subjects, R subjects are randomly and non-repeatedly selected to form a source domain, and finally (rounded down) source domains are used as the training set. Conduct S experiments and calculate the average accuracy of these experiments.

[0090] The cognitive state predicted by the model in which the target domain sample set converges during training and the true state E T are compared to obtain the accuracy result to evaluate the model performance. The accuracy is the number of samples correctly classified by the model during testing divided by the total number of test samples. The formula for calculating the model accuracy is as follows:

[0091]

[0092] where acc is the model accuracy; T is the number of samples correctly classified by the model; F is the number of samples misclassified by the model.

[0093] Based on the same inventive concept, the embodiments of the present application also provide an electroencephalogram (EEG)-based deep learning classification device for cognitive state for implementing the above-mentioned EEG-based deep learning classification method for cognitive state. The implementation solutions provided by this device to solve problems are similar to the implementation solutions described in the above method. Therefore, the specific limitations in one or more embodiments of the EEG-based deep learning classification device for cognitive state provided below can refer to the limitations on the EEG-based deep learning classification method for cognitive state in the above text, and will not be repeated here.

[0094] In an exemplary embodiment, as Figure 7 shown, an EEG-based deep learning classification device for cognitive state is provided, including:

[0095] A signal acquisition module 701, configured to acquire original EEG signals; the original EEG signals include multiple channel signals. A within-band feature extraction module 702, configured to divide the original EEG signals in the frequency domain to obtain multiple frequency bands, and in each frequency band, obtain the within-band features corresponding to each frequency band according to the phase synchrony between the channel signals of the frequency band. A fused feature extraction module 703, configured to fuse the within-band features of all frequency bands to obtain fused features. A cognitive state classification module 704, configured to determine the cognitive state corresponding to the original EEG signals according to the fused features, using a cognitive state classification model; the cognitive state classification model is a neural network model pre-built according to a training sample set; the training sample set includes the feature information of multiple sample EEG signals and the cognitive state corresponding to each sample EEG signal.

[0096] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as shown in Figure 8 . The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store raw electroencephalogram signals. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a deep learning classification method for cognitive states based on electroencephalogram signals.

[0097] Those skilled in the art can understand that Figure 8 the structure shown in

[0098] is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0099] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, it implements the steps in the above method embodiments.

[0100] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, it implements the steps in the above method embodiments.

[0101] The beneficial effects of this application are as follows:

[0102] This application uses PLV for feature extraction, analyzes the internal connections between different brain regions, and to a certain extent solves the problem of insufficient alignment of the distribution differences of EEG signals. The multi-head attention mechanism is used to selectively focus on the specific frequency band features related to the cognitive state. Execute the mapping between frequency bands to transform the extracted features into the final attention representation, thereby promoting the integration of complementary information between different frequency bands. The deep learning classification method for cognitive state based on EEG signals provided by this application trains a high-precision cognitive state classification model across time periods and subjects, which can effectively combine time-frequency information and complementary features between frequency bands, and provides strong support for the early detection of cognitive impairment. The deep learning classification method for cognitive state based on EEG signals proposed in this application uses a variety of frequency band analysis methods to evaluate the cognitive state from multiple perspectives, and has the advantages of small time complexity, high computational efficiency, strong generalization ability, etc., and has a wide application prospect in actual brain-computer interaction.

[0103] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with relevant regulations.

[0104] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0105] The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logics, data processing logics based on quantum computing, etc., without limitation.

[0106] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0107] Specific examples are used in this article to elaborate on the principles and implementation manners of the present application. The descriptions of the above embodiments are only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A deep learning classification method for cognitive states based on EEG signals, characterized in that: The deep learning classification method of cognitive state based on EEG signals includes: Acquire an original EEG signal; the original EEG signal includes multiple channel signals; The original EEG signal is divided in the frequency domain to obtain a plurality of frequency bands, and in each frequency band, according to the phase synchronization between the channel signals of the frequency band, the frequency band features corresponding to each frequency band are obtained; The intra-band features of all frequency bands are fused to obtain fused features; According to the fusion features, a cognitive state classification model is used to determine the cognitive state corresponding to the original EEG signal; the cognitive state classification model is a neural network model pre-built based on a training sample set; the training sample set includes feature information of multiple sample EEG signals and the cognitive state corresponding to each sample EEG signal.

2. The method for deep learning and classification of cognitive states based on EEG signals according to claim 1, characterized in that: Before dividing the original EEG signal in the frequency domain to obtain a plurality of frequency bands, and obtaining the intra-band features corresponding to each frequency band according to the phase synchronization between the channel signals of the frequency bands, the EEG signal-based cognitive state deep learning classification method further includes: The original EEG signal is sequentially subjected to downsampling, bandpass filtering and independent component analysis.

3. The method for deep learning and classification of cognitive states based on EEG signals according to claim 1, characterized in that: The original EEG signal is divided in the frequency domain to obtain multiple frequency bands, and in each frequency band, according to the phase synchronization between the channel signals of the frequency band, the frequency band features corresponding to each frequency band are obtained, which specifically include: Dividing the original EEG signal in the time domain to obtain multiple time periods; For any time period, the time period is divided in the frequency domain to obtain a plurality of frequency bands of the time period, and in each frequency band of the time period, the phase synchronization between the channel signals of the frequency band is calculated to obtain the channel pair characteristics of each frequency band of the time period; According to the channel pair features of all frequency bands in all time periods, a deep learning algorithm is used to obtain the intra-band features corresponding to each frequency band.

4. The method for deep learning and classification of cognitive states based on EEG signals according to claim 3, characterized in that: The deep learning algorithm is a multi-head attention mechanism.

5. The method for deep learning and classification of cognitive states based on EEG signals according to claim 1, characterized in that: The method for building the cognitive state classification model includes: Acquire multiple sample EEG signals; each sample EEG signal contains multiple channel signals; For any sample EEG signal, the sample EEG signal is divided in the frequency domain to obtain a plurality of sample frequency bands, and in each sample frequency band, according to the phase synchronization between the channel signals of the sample frequency band, the sample frequency band features corresponding to each sample frequency band are obtained; For any sample EEG signal, the sample frequency band features of all sample frequency bands of the sample EEG signal are integrated to obtain feature information of the sample EEG signal; According to each feature information and the cognitive state corresponding to each feature information, the neural network model is trained to obtain the cognitive state classification model.

6. The method for deep learning and classification of cognitive states based on EEG signals according to claim 5, characterized in that: For any sample EEG signal, the sample EEG signal is divided in the frequency domain to obtain multiple sample frequency bands, and in each sample frequency band, according to the phase synchronization between the channel signals of the sample frequency band, the sample frequency band features corresponding to each sample frequency band are obtained, specifically including: Dividing the sample EEG signal in the time domain to obtain a plurality of sample time periods; For any sample period, the sample period is divided in the frequency domain to obtain a plurality of sample frequency bands of the sample period, and in each sample frequency band of the sample period, the phase synchronization between the channel signals of the sample frequency band is calculated to obtain the channel pair feature of each sample frequency band of the sample period; According to the channel pair features of all sample frequency bands in all sample time periods, a deep learning algorithm is used to obtain the intra-sample frequency band features corresponding to each sample frequency band.

7. A deep learning classification device for cognitive states based on EEG signals, applied to the deep learning classification method for cognitive states based on EEG signals as claimed in any one of claims 1 to 6, characterized in that: The cognitive state deep learning classification device based on EEG signals includes: A signal acquisition module, used to acquire an original EEG signal; the original EEG signal includes multiple channel signals; An intra-band feature extraction module is used to divide the original EEG signal in the frequency domain to obtain multiple frequency bands, and in each frequency band, according to the phase synchronization between the channel signals of the frequency band, obtain the intra-band feature corresponding to each frequency band; A fusion feature extraction module is used to fuse the intra-band features of all frequency bands to obtain fusion features; A cognitive state classification module is used to determine the cognitive state corresponding to the original EEG signal based on the fusion features and using a cognitive state classification model; the cognitive state classification model is a neural network model pre-built based on a training sample set; the training sample set includes feature information of multiple sample EEG signals and the cognitive state corresponding to each sample EEG signal.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the deep learning classification method for cognitive states based on EEG signals as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for deep learning and classification of cognitive states based on EEG signals described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for deep learning and classification of cognitive states based on EEG signals described in any one of claims 1 to 6 is implemented.

Images