Abstract: Brain-Computer Interface (BCI) is an emerging technology for communication between human brain and external devices. The traditional feature extraction method based on time-frequency cha- racteristics cannot reflect the nonlinear characteristics of EEG signals. In order to further improve the accuracy of classification, the pretreatment method of wavelet threshold denoising is firstly used to improve the signal-to-noise ratio of EEG signals. Then, the feature extraction of the three kinds of imaginary motion EEG signals is carried out by the parameter-sample entropy of nonlinear dynamics, and the nonlinear features of EEG signals are preserved. Among them, the research of Motor-Imagery (MI) EEG has always been the focus of BCI that is a high-speed development field. This paper also studies three classifiers including support vector machine, LVQ neural network and BP neural network. The experimental results show that BP neural network has higher recognition rate for classification and recognition of EEG signals.

Key words: sample entropy, feature extraction, BP neural network, pattern recognition