Recognition of EEG Based on Stacked Sparse Denoising Auto-Encoder

An improved algorithm, stacked sparse denoising auto-encoder (SSDAE), is proposed in this paper. In the new algorithm, the noise of original input data is processed, the hidden layers is limited to sparse restrictions, finally, EEG features are classified with the softmax. Experiments results on two different data sets (Laboratory data sets and 2005 BCI competition data set IVa) demonstrate that SSDAE had the highest recognition accuracy than traditional algorithms, which proves that SSDAE has the stronger learning ability and robustness in motor imagery EEG feature extraction.