A New Bionic Model and Its Application to Epileptic Electroencephalograph Recognition

Abstract

Imitating the nervous system is not only an effective method to construct artificial neural networks with better performance, but also a research hotspots. In this paper, a new bionic model stimulating olfactory neural systems, KIII model, is introduced and its performance is researched based on epileptic electroencephalograph (EEG) recognition. In section 2, KIII model is introduced briefly. The structure of the KIII model is similar to that of olfactory neural systems. The model of neurons is based on the real action of neurons with the stimulus and optimized according to mathematical optimization. In section 3, KIII model was used as a classifier and the performance of the KIII model was evaluated to identify epileptic EEG. In the first group of experiments, the features are extracted based on EMD and the recognition rate of the KIII model is over 91 % with few training times. In the second group of experiments, extracting the feature of EEG is unnecessary and the raw EEG signals are used as the input directly. The KIII model gives the better recognition rate over 96 %. Experimental results show that KIII model has remarkable characteristics: (1) KIII model can learn the pattern with few training times, which is different from general deep learning models; (2) KIII model can recognize raw signals without special feature-extracting process, which is similar with deep learning model; (3) the structure of KIII model is very similar to that of real olfactory neural systems, which is partly similar with deep learning model.