Development of soft computing technique for classification of EEG signal

Brain Machine Interface (BMI) using EEG signal provides the direct communication pathway between the human brain and an external device. Research on Brain Computer Interface (BCI) begins in 1970s at university of California, Los Angeles. BMI allows us to manipulate computer and other machinery with our thoughts. Motor Imagery (MI) basically involves imaginations of any body part that leads to the activation of sensorimotor cortex which modulates the oscillation in EEG. BCI can be done by both Invasive and Non-invasive methods. In invasive, BCI are implanted into the gray matter of individual's brain by neurosurgery. Invasive method produce highest quality signals as they are directly lie in the gray matter. But gradually its signal strength weakens as scare-tissue build ups and as body always react to the foreign object in the brain [1]. Non-invasive BCI is preferred as it is easy to wear and do not require surgery. EEG is most studied non-invasive BCI because of its high temporal resolution and ease to use and low coast set up. Individual difference is the major issue in MI based BCI. Algorithms that can successfully overcome individual differences is to create diverse ensemble classifiers which uses sub-band common spatial pattern method (SBCSP). To aggregate the output of this ensemble, we use fuzzy integral with particle swarm optimization (PSO) techniques. The combination of fuzzy with PSO produces robust performance for offline classification of EEG signal.