Detection Of Single-Trial Evoked Potential Based On The Normalized Rbf Neural Network

The problem of detecting the single-trial evoked potentials with noise is investigated. One of main task for evoked potentials estimation is to extract the amplitude of the response and the corresponding latency without losing the individual properties of each epoch that is meaningful to clinicians and practical recognition problems. For this purpose, an approach with normalized RBF neural network is proposed to obtain preferable responses against other nonlinear methods which include RBF adaptive noise canceling and RBF neural network. Based on the MSE and the ability of peak tracking, the presented method is also compared with other common techniques. Several results show that the proposed algorithm can provide the convergent evidence that our method can significantly attenuate noise and successfully detect the changes between trials. The proposed method also gives a robust estimation of the noisy signal.