Research On Analysis & Recognition Method For Multi-motor Erd/ers Signal In Bci System

Brain-Computer Interface (BCI) is to establish a direct information and control channel between the human brain and a computer or other electronic equipment. For its great value in the research areas of the rehabilitation engineering, brain science, artificial intelligence and so on, BCI research is becoming one of the hot issues.The focus of this article is on Event Related Desynchronization (ERD)/Event Related Synchronization (ERS) which exists in the process of motor imagery. For the ERD/ERS signal, there have been a lot of discussions on the imagination of left and right hand movements with many reliable results for its characteristic frequency band and the corresponding cortical activity in the region of cerebral cortex in the literatures.But the research on the foot and tongue ERD/ERS signals is still limited and, their characteristic frequency band and the corresponding brain region are still unconvincing.The main task of this article is to analyze and classify the ERD/ERS signals evoked by the motor imagery of left hand, right hand, foot and tongue, to find the features of the four motor imagery ERD/ERS signals (with the emphasis on the foot and tongue movements) and to use these features to find some effective movement pattern classification algorithms.Our data is from the BCI competition 2005 dataⅢa, collected by BCI lab of Graz University of Technology, Austria good reliability.We use a variety of methods, including Time-Frequency Spectrogram (TFS), the Power Spectral Density (PSD) curve, ERD/ERS coefficient, brain topographic mapping,Power Spectral Entropy(PSE),Wavelet Entropy(WE)and Independent Component Analysis (ICA) based frequency-spatial filtering method, to analyze the four modes of motor imagery signals in detail which not only confirmed the conclusions of the literature on the left and right hand movements, but also gave us a clear understanding about the features of ERD/ERS evoked by foot and tongue movements, including feature frequency band, good channel-frequency differences from other movements, and the corresponding brain areas.On this basis, we have used several methods to finish the feature extraction and classification to the experimental signal.The results showed that the combination of ICA-based signal extraction algorithm and Support Vector Machine (SVM) based classification method is an effective tool for the identification of motor imagery potentials. This classification algorithm achieved a high accuracy rate (91.4% for the highest group,77.6% for the lowest group).The improvements in this algorithm reduced the calculation time (to 15%-20% of the original), and maintained a relatively high rate of correct classification. It may lay the foundation for the final realization'of real-time BCI system based on motor imagery potential.