Continuous neural networks for electroencephalography waveform classification

Nowadays classification of electroencephalography (EEG) signals have brought new perspectives in the understanding of the brain. Establishing associated characteristics to certain stimulus in EEG is a monumental work due to complexity of the brain responses. For EEG classification several methods have been proposed. Among various statistical methods, Neural Networks (NN) have demonstrated capability in EEG classification using static and recurrent structures. In this paper, we propose a classification method based on Continuous Neural Networks (CNN). Such class of algorithm can handle the raw EEG signal. The method is divided in three stages, first the CNN is trained by using a part of a known database, secondly a parallel structure of the CNN is build with the weights obtained after training, third the parallel structure is tested with the rest of the database that was not used for the training process. All the previously mentioned process is developed by using the raw EEG signals presented on the database and introducing them directly to the CNN without any previously process. The classification algorithm produces a 97% of efficiency.