An application of common spatial pattern algorithm for accuracy improvement in classification of cortical activation pattern according to finger movement

In this paper, we tried to suggest feature extraction method using CSP Algorithm to improve the accuracy of classifier according to finger movement to develop the Upper Limb Rehabilitation Robot System based on brain signal. Four subjects participated in the experiment and they conducted four kind of tasks in three times. The task is divided by two kind of movement(Digit Flexion/Extension, Thumb Flexion/Extension), and two kind of mode(Active, Passive). We measured brain signal according to finger movement using fNIRS(functional Near Infrared Spectroscopy, FOIRE-3000, Shimadzu, Japan). Also, sampling rate of measured sign is 7.6923Hz and there are 24 channels. We conducted preprocessing process using HRF(Hemodynamic Response Function) and Wavelet-MDL(minimum description length) to remove the noise and global bias in selected signal. After preprocessing process, we extracted feature using CSP(Common Spatial Pattern) Algorithm and calculated the accuracy of classification in each task using Support Vector Machine(SVM). There is accuracy of classification result by applying CSP Algorithm. Accuracy of classification using SVM is about 69.53% and after applying CSP Algorithm is about 71.82%. we can find that it improved 2.29%.