Feature Extraction Process with an Adaptive Filter on Brain Signals Motion Intention Classification

摘要

Identifying motor imagery from an electroencephalogram (EEG) has been researched from different perspectives and methods of classification. Translating a brain signal into a language understandable for machines relies on feature extraction techniques, which vary from working on the frequency domain to dealing with raw data. Using statistical information to classify motor imagery has shown encouraging results. In this paper we benefit from statistical approaches and propose a different perspective to boost results obtained through brain signals provided by a low cost EEG. Our motivation is based on the natural separability of classes exhibited by statistical indicators such as the mean and standard deviation. A special emphasis in our method is made on filtering data to subject readings in an adaptive manner, leading to a successful classification rate of 97%, outperforming Hjorth's mobility and complexity measure, a state-of the art technique used in EEG signal classification.