L1-norm based discriminative spatial pattern for single-trial EEG classification

Spatial filtering provides an efficient method for single-trial EEG classification and has been widely used in EEG-based brain computer interfaces. However, scalp-recorded EEG signals are usually very noisy since they could be contaminated by various outliers, such as EOG or EMG artifacts. The outliers may seriously distort the performance of spatial filters. To solve this problem, we propose a new robust spatial filtering algorithm, namely DSP-L1, which is L1-norm based discriminative spatial pattern (DSP). Compared with the conventional DSP, DSP-L1 takes advantage of the robust L1-norm modeling that expects to perform better in suppressing the effect of outliers. Computationally, an iterative approach is introduced to find the spatial filters of DSP-L1. Experimental results on two EEG data sets of motor movements demonstrate the efficiency of the proposed method.

► A new discriminant approach, termed DSP-L1, is proposed. ► It is for movement-related potentials (MRPs)-based single-trial electroencephalogram (EEG) classification. ► An iterative algorithm is presented to solve DSP-L1.