Frequency Domain CSP for Foot Motor Imagery Classification Using SVM for BCI Application

Abstract

The primary requirement for developing a BCI-based foot prosthetic for the lower limb amputation is to classify foot motor imagery through brain signals of both right and left limbs. It is necessary to effectively and accurately recognize the lower limb motor imagery through Electroencephalogram (EEG) signals as quickly as possible with the best possible combination of feature extraction and classification algorithms with maximum accuracy in the least time. Hence, an optimum option needs to be reached and so in this work, a combination of using Frequency Domain Common Spatial Pattern (FDCSP) for feature extraction and classification by SVM of two-class motor imagery which is left and right foot is proposed. Classical CSPs may often not retain the discriminatory functions among classes in the time domain. Therefore, the work to reduce the limitations of traditional CSP we have introduced a frequency domain CSP (FDCSP) technique and a set of features are selected for further process. Significant feature vectors are then applied to train SVM by which the right and left foot motor imagery movement have been differentiated. The average classification accuracy by using SVM has given the best result of 85 percent whereas for kNN the accuracy is 77%. Comparing with the other CSP methods it has been demonstrated that FDCSP outperforms other state of art methods. The attained results demonstrate that the frame described can be used as a tool to promote discrimination against lower limb motor imagery movement.