Research On Upper Limb Action Recognition Based On EEG

The rehabilitation of the athletic ability is very important for people who are disabled. Although the rehabilitation treatment for movement disorders has made significant progresses, but the overall effect is still not satisfactory. In recent years, the idea of bringing the brain-computer interface technology into rehabilitation methods has attracted the interest of a growing number of researchers. The rehabilitation method based on brain-computer interface technology collects patients'neural activity in real-time, decodes the neural signal, and generates movement for rehabilitation training. This method can combine patients'actual movement intention and the actual body actions to achieve the most effective training.First, according to the specific needs of the experiment system which consisted of the EEG acquisition equipment, movement acquisition equipment and indicating device was set up. In consideration of convenience and real-time, an embedded system with LM3S9B96 driving a TFT LCD was designed to serve as the indicating device.For the further feature extraction the signal was first preprocessed. The preprocessing included data filtering and segmentation. EEG data of each task was separated. Power estimation, AR model parameters and the wavelet coefficients were extracted as the EEG features.Support vector machine was used as classification algorithm to distinguish the EEG signals during different direction hand movements with classification accuracy maximized at 55.56%. Since the parameters of support vector machine have a great impact on the results and difficult to determine. This paper then used the particle swarm optimization algorithm to optimize the parameters of support vector machine and then discussed the optimization process and results,which revealed that the optimization slightly decreased the classification accuracy and greatly reduced the amount of calculation.