For the deficit of sensitivity in attention level extraction from a single parameter of EEG, a method to extract attention level from multi-parameters of EEG was designed and implemented. In this method, 7 characteristic parameters, which were used for a full assessment of spectral characteristics of EEG, were first derived, and then the back propagation (BP) neural network algorithm was applied to the nonlinear mapping from 7 parameters to the attention level, so that it could achieve a more accurate attention level extraction. In addition, various improvements were employed on the BP algorithm in order that it could overcome the defect of slow training speed and escape from the local minimum quickly. In the network training using 34 samples, the neural network converged quickly, and the test of attention level extraction after convergence verified that the network had high identification ratio.
|