30 results about "Eeg classification" patented technology

The invention discloses a motor imagery EEG classification method based on HELM combined with PTSNE manifold and LDA feature fusion, and improves the classification accuracy. In terms of feature extraction, on the one hand, using PCA combined with LDA to extract linear features can not only eliminate noise, but also consider the label information of training data; Complex nonlinear intrinsic manifold features. In terms of feature classification, the HELM algorithm with high classification accuracy is used to classify and identify motor imagery EEG signals.

An emotion recognition system and application based on deep learning and brain-computer interface, comprising a sequentially connected portable EEG acquisition device, an emotional EEG classification module and an emotion classification display module, the portable EEG acquisition device collects data from the brain of a subject Emotional EEG signal, the emotional EEG classification module analyzes the emotional EEG signal, establishes a relationship model between the emotional EEG signal and emotion, and through the relationship model, the input emotional EEG signal is analyzed. Emotional classification, the emotional classification display module displays the recognized emotional types and prompts the emotional state of the subject. The emotion recognition system and application based on deep learning and brain-computer interface of the present invention can realize accurate acquisition, effective identification and correct classification of emotional EEG signals, intuitively prompt emotional states, and realize the function of emotional state monitoring. The invention directly establishes the corresponding relationship between the EEG signal and the emotion type, and realizes the feedback to the emotion.

An electroencephalogram (EEG) classification detection device based on lacuna characteristics belongs to the technical field of electroencephalogram automatic detection. The EEG classification detection device comprises a multi-way EEG amplifier, a data collection card and a computer which are sequentially connected through a circuit. A signal preprocessing module, a signal segmentation module, a lacuna characteristic extraction module, a Bayes linear discriminant analysis classification module and a threshold judgment module are built in the computer. The multi-way EEG amplifier first amplifies EEG signals, then the data collection card collects the EEG signals and transmits the signals to the computer, finally the modules in the computer are utilized to conduct preprocessing and segmentation on the EEG signals and calculate the lacuna characteristics of the signals, a Bayes linear discriminant analysis classification device is utilized to classify the EEG lacuna characteristics, and the threshold judgment module is used for marking the classification and obtaining a result. The EEG classification detection device has the advantages of being simple in characteristic operation, high in practice and classification speed, high in classification accuracy and capable of achieving good classification detection effect.

The present invention provides an EEG intent decoding method based on a non-negative CP decomposition model. The method extracts the time component features of the EEG of different subjects in the boundary avoidance task, optimizes the feature dimension by using 2-DPCA, and uses The support vector machine classifies the features, which can reflect the differences in the EEG of the subject in different states, and the accuracy of the EEG classification of a single subject is high; and this method uses the interaction between the various modes of the EEG, and uses the channel The components and frequency components obtain the time component characteristics of the EEG, and the characteristics of the obtained EEG time components are well separable. By optimizing its dimension, it can effectively detect the left and right hand movements in the boundary avoidance task. EEG intent to decode.

The present application provides a method for classifying EEG signals, the method comprising: obtaining an EEG signal sequence to be identified; inputting the EEG signal sequence to be identified into an EEG signal classification model to obtain the EEG signal sequence to be identified EEG classification information, the EEG classification information includes motor imagery EEG signals and steady-state visual evoked potential EEG signals.

The invention belongs to the field of mode recognition and a brain-machine interface and discloses a motor imagery electroencephalogram voting strategy sorting method based on extreme learning machines. The motor imagery electroencephalogram voting strategy sorting method comprises the following steps: dividing an original motor imagery electroencephalogram into S sections of sub-signals; carrying out dimensionality reduction on each section of sub-signal by a principal component analysis method; carrying out secondary dimensionality reduction on a feature vector subjected to the dimensionality reduction by a linear discrimination analysis method; carrying out the same processing on the S sections of sub-signals to finally obtain S (K-1)-dimensional feature vectors, and combining the S (K-1)-dimensional feature vectors to finally obtain an S*(K-1)-dimensional feature; and transmitting the S*(K-1)-dimensional feature into a plurality of ELM (Extreme Learning Machine) sorting devices so as to obtain a final sorting result by utilizing a voting sorting strategy. The invention provides a voting sorting strategy based on the ELMs; compared with a traditional multi-time ELM average accuracy scheme, the method provided by the invention has the advantages that the sorting accuracy is improved under the condition of not influencing the training sorting low consumption.

The present invention relates to the algorithm of brain-computer interface, artificial intelligence and pattern recognition, more specifically, relates to the SSVEP EEG classification method based on the convolutional neural model enhanced by EMD data, which can be applied to the fields of medical equipment, human-computer interaction, robot control, etc. . This method first preprocesses the original EEG data. After preprocessing, the empirical mode decomposition method is used to decompose the original EEG data, and a large number of artificial EEG data that conform to the time-frequency domain characteristics of the original EEG signal are generated by mixing. The original EEG data is merged and used for parameter training of the neural network, so as to achieve the effect of effectively training network parameters with a small amount of EEG data. Finally, the complex Morlet wavelet transform is used to generate the EEG tensor, and the original time domain data is converted into a tensor dictionary integrating time domain, frequency domain and spatial information as the input of the neural network, and the convolutional neural network model is used to enhance the data. EEG training set for classification.

The invention belongs to the field of pattern recognition and EEG signal processing, and relates to an EEG signal classification method based on an improved ResNeXt network; it includes four parts: EEG signal acquisition, preprocessing, feature extraction, and training ResNeXt classification network; training ResNeXt classification network is Refers to: dividing the training set and test set; building an improved ResNeXt EEG signal classification network; training the improved ResNeXt EEG signal classification network; building an improved ResNeXt EEG signal classification network refers to: improving on the basis of ResNeXt, and convolving the groups The middle layer of the convolutional layer of each block module increases the direct connection operation, speeds up the speed of model convergence, reduces the test error of the model, and improves the generalization ability; the invention speeds up the convergence speed of the classification model, compared with the convolutional neural network brain The electrical classification model, the improved ResNeXt classification model is easier to optimize, effectively improves the gradient explosion problem in the deep training model, and can greatly deepen the number of layers of the network while avoiding the degradation of the classification model.

The present application provides a method for generating an EEG classification model, a device for generating an EEG classification model, electronic equipment, and a computer-readable storage medium. The method for generating an EEG classification model includes: obtaining sample data of K subjects, wherein, The sample data includes: classified EEG information and classification results of the corresponding EEG information, the K is greater than or equal to 2; based on the sample data of K subjects and the preset first objective function, the calculation makes the first An orthogonal transformation matrix with the minimum value of the objective function, wherein the first objective function is a function related to the orthogonal transformation matrix and the EEG information of K test subjects, and the orthogonal transformation matrix is ​​used to transform the K test subjects The respective EEG information is transformed into correlation information among K subjects; an EEG classification model is generated based on the orthogonal transformation matrix. The technical solution of the present application is used to generate an EEG classification model that can be applied to multiple subjects, saving the maintenance cost of the EEG classification model.

The present invention relates to an ELM motor imagery EEG classification method based on the AR coefficient space. First, the motor imagery EEG signal of a single channel is used as a random signal, and the p-order AR model is used for fitting, expressed as: ; and then the Burg algorithm is used Solve the undetermined coefficients a1, a2, ..., ap of the AR model; then combine the p-order AR coefficients of m channel motor imagery EEG signals into an AR coefficient vector, and embed it into the original optimization problem of ELM network parameter training Among them, the optimal external weight β is solved, and the ELM classification algorithm based on the AR coefficient space is constructed. The method of the invention improves the classification accuracy and classification speed of motor imagery EEG signals.

The invention discloses an EEG classification method based on the entropy value of dynamic functional connection. First, the obtained original EEG signals are preprocessed, and then filtered; then the phase synchronization analysis method is used to calculate the EEG signals of each frequency band at each time. Point the phase relationship between each two channels to obtain the dynamic functional connection matrix; then calculate the time domain entropy of the phase relationship value between the two channels one by one to obtain the information entropy of each edge to measure the time of each edge of the EEG functional network The complexity of the domain; the dynamic functional connection entropy of each frequency band is used as the classification feature of the EEG functional network, and the adaptive improvement classifier is trained to obtain multiple adaptive improvement classifiers and the corresponding classification accuracy; The samples are combined for classification. The problem of low classification accuracy in the existing EEG signal classification methods is solved.

The invention proposes a multi-class EEG classification method based on a dual-rule active over-limit learning machine. The method of the present invention adopts the core idea of ​​active learning, first evaluates the uncertainty of unlabeled samples according to the extreme learning machine classifier, and secondly eliminates unlabeled samples with high similarity according to the cosine similarity rule, and obtains the most valuable small number of unlabeled samples Labeling, and then use the filtered data to train the extreme learning machine, maximize the use of the internal information of the labeled EEG signal, thereby reducing the dependence on the labeled EEG data, and obtain a higher multi-classification of motor imagery tasks accuracy. This method has broad application prospects in the field of brain-computer interface.

An EEG classification method based on frequency-band attention residual network belongs to the field of pattern recognition and bioinformatics. First, preprocess the original EEG signal to remove noise and artifacts in the signal; then, use a fixed-length sliding window to segment the preprocessed N-lead EEG signal to obtain a total of S-segment N-lead sub-signals ; Then use the wavelet packet decomposition to decompose and reconstruct the S segment N derivation sub-signals, each sub-band signal of each data segment is decomposed into F sub-band signals; after that, the multi-conductor decomposition results of each frequency band are converted into electrode correlation matrix; and then use the electrode correlation matrix of F frequency bands as the input of the frequency band attention residual network to complete the classification of EEG signals. Compared with the existing technology, the present invention has the advantages that the frequency band attention module is used to obtain frequency band importance weights, and a personalized frequency band attention distribution is given to each sample.

The invention discloses a transfer learning method and system for EEG classification based on Euclidean alignment and Procrustes analysis, and belongs to the field of brain-computer interface based on motor imagery. Including: covariance alignment of the feature matrix of previous users and new users; mean alignment of feature matrices of previous users and new users; according to the feature matrix of previous users after the label and mean alignment of previous users, calculate the category center of previous users, Calculate the category center of the new user according to the pseudo-label of the new user and the feature matrix of the new user after the mean value alignment; construct the matrix of the category center of the previous user and the new user, and calculate the relationship between the category center of the new user and the new user through the orthogonal Platts analysis method. The rotation matrix corresponding to the center alignment of the categories of the previous users; the feature matrix of the new user after the mean alignment is multiplied by the rotation matrix to obtain the final aligned new user data.

The invention discloses a method for calculating symbol transfer entropy and brain network characteristics based on time-frequency energy. First, the collected motor imagery EEG signals (MI-EEG) are preprocessed based on a common average reference; MI-EEG performs continuous wavelet transform to obtain its time-frequency-energy matrix, and splices the time-energy sequences corresponding to each frequency in the frequency band closely related to motor imagery in turn to obtain the one-dimensional time-frequency energy of the lead sequence; then, calculate the sign transfer entropy between any two lead time-frequency energy sequences, construct a brain connectivity matrix, and use the Pearson feature selection algorithm to optimize the matrix elements; finally, calculate the degree and betweenness centrality of the brain functional network , constituting the feature vector for MI‑EEG classification. The results show that the present invention can effectively extract the frequency domain features and nonlinear features of MI-EEG, and has obvious advantages compared with the traditional feature extraction method based on brain function network.