Motion Intention Classification of Multi-Class Upper Limbs Actions for Brain Machine Interface Applications

In this paper, we aim to achieve the construction of 4-class fNIRS-BCI system. The system performed in assistive-device could help patients with dyskinesia carry out multiple action paradigms of rehabilitation training, and it could provide power assistance to groups of people who carry heavy weights. This paper studied multi-class classification of upper limb movements based on using cerebral hemoglobin information. Seventeen healthy subjects participated in the experiment and accomplished a set of upper limbs action paradigm, including lifting-up, putting down, pulling back and pushing forward. Cerebral hemoglobin information was measured simultaneously by using fNIRS technology. To identify motion intention timely, the data obtained before actual motion was analyzed. Signals were decomposed into three frequency bands by wavelet packet. ReliefF and genetic algorithms were used to select optimal features. A library for support vector machines (LIBSVM) method was applied for pattern recognition and the average recognition rate was 70.6%. The results demonstrated the feasibility of classifying 4-classes of motion intentions based on cerebral hemoglobin information. It has a potential to provide multi-class commands for motion-assistant device.