Character Identification in Farwell Imaginary Matrix Based on SVM Feature Optimization
Baikun Wan · Journal of Tianjin University Science and Technology · 2011
Traditional classification in Farwell imaginary character matrix BCI only uses EEG features on several electrodes located on the middle line.For the limited recognition information,the classification efficiency is low.In this paper,a support vector machine feature optimization method is introduced,which evaluates the feature’s contribution to recognition by disturbing cost function of support vector machine and then optimizes the feature combination.EEG signals from 6 subjects evoked by Farwell imaginary matrix task,which includes 80 characters,are collected.Support vector machine feature optimization method is used on these EEG data and classifiers were constructed to recognize the character.The results show that optimal feature combination can improve task recognition accuracy significantly(the recongnition error rate is 0.9%).This study suggests that the support vector machine feature optimization method can provide effective feature selection,so it can be applied to the reduction of feature dimension in high-dimensional evoked EEG signal processing of BCI and is worthy of further development.