Research on Motor Imagery Signal Recognition Method Based on Wavelet Packet Transform and ConvLSTM

Brain Computer Interface (BCI), as a novel method of information communication and control, is an interdisciplinary research topic which involves neuroscience, signal processing, and pattern recognition. The BCI system based on motion imagination is considered the most promising type of brain computer interface system. The key to the BCI system lies in the analysis and processing of EEG signals. This article combines the characteristics of CNN and LSTM models, and combines wavelet packet transform to propose a ConvLSTM based imaginative signal recognition model. The model first uses wavelet packet transform to extract features of EEG signals, then uses convolution operation for spatial features extraction of EEG signals, and uses LSTM network for temporal features extraction of EEG signals. ConvLSTM model is able to extract local high-level features and take the advantage of LSTM model to predict the sequence and the Wavelet Packet Transform has the advantage including multi-resolution analysis, spatial locality, simple and fast computation, and sparsity. The simulation experiment results demonstrate that the method is feasible.