Enhancing Mind Controlled Smart Living Through Recurrent Neural Networks.

Abstract

While smart living based on the controls of voices, gestures, mobile phones or the Web has gained momentum from both academia and industries, most of existing methods are not effective in helping the elderly or people with muscle disordered or motor disabilities. Recently, the Electroencephalography (EEG) signal based mind control has attracted much attentions, due to the fact that it enables users to control devices and to communicate to outer world with little participation of their muscle systems. However, the use of EEG signals face challenges such as low accuracy, arduous and time-consuming feature extraction. This paper proposes a 7-layer deep learning model that includes two layers of Long-Short Term Memory (LSTM) cells to directly classify raw EEG signals, avoiding the time-consuming pre-processing and feature extraction. The hyper-parameters are selected by an Orthogonal Array experiment method to improve the efficiency. Our model is applied to an open EEG dataset released by PhysioNet and achieves 95.05% accuracy over 5 categorical EEG raw data. The applicability of our proposed model is further demonstrated by two use cases of smart living in terms of assisted living with robotics and home automation.