CNN-SVM-based event-related potential signal classification method

Abstract

The invention discloses a CNN-SVM-based event-related potential signal classification method. The method comprises the following steps: firstly, carrying out band-pass filtering on an acquired electroencephalogram signal through a band-pass filter; making the electroencephalogram signals into a data set with labels, and dividing the data set into a training set, a verification set and a test set; inputting the training set into a designed convolutional neural network model for training, and selecting optimal network parameters by using the verification set; inputting the training set, the verification set and the test set into the trained convolutional neural network, outputting downsampling layer features of the network, wherein features corresponding to the training set and the verification set are training features, and features corresponding to the test set are test features; and using the training features to complete support vector machine model training, and using the trained support vector machine model to classify the test features so as to obtain the identification result of the event-related potential signal. According to the invention, accurate identification of event-related potential signals is realized, and the practical value of a brain-computer interface system is improved.

Description

technical field [0001] The invention relates to the technical field of event-related potential brain-computer interface, in particular to a method for classifying event-related potential signals based on CNN (convolutional neural network)-SVM (support vector machine). Background technique [0002] Brain-computer interface (BCI) is a technology that allows the brain to interact directly with the outside world without the help of peripheral neural pathways. Because of its low cost, high time resolution, and good safety, this technology is widely used in language communication, environmental control, It has been widely used in sports rehabilitation and other fields. Brain-computer interface applications based on EEG signals can use a variety of EEG forms, including steady-state visual evoked potential (SSVEP) and event-related potential (ERP). [0003] ERP is a transient brain response evoked by a series of specific stimuli, reflecting the brain's processing mechanism for phys...

AI Technical Summary

Problems solved by technology

[0004] Traditional ERP signal classification is generally achieved by manually extracting the frequency domain or time-frequency domain feature information in the EEG signal, and then performing supervised classification on the extracted features, but the traditional method generally requires multiple superposition of signal waveforms. Effective identification of ERP signals can only be achieved by strengthening the signal-to-noise ratio. Compared with traditional feature extraction methods, deep learning can automatically mine deeper features of signals to avoid information loss. However, due to the weak signal of ERP signals, individual differences are large. And because of the small sample size of EEG signals, overfitting often occurs in the process of deep learning, which limits the application of brain-computer interface engineering based on ERP signals.

Images