Hybrid Cnn-Bi-Lstm Model-Based Intention Decoding from Fingers of the Same Hand in High Demanding Boundary Voidance Task

Abstract

Capability for dexterous decoding of control intention involving electroencephalography (EEG) is of paramount significance to the research of cognitive perception and movement in computational neuroscience. A reasonable experimental task is a crucial prerequisite for evoking high-quality EEG signals. However, most EEG evoking tasks in computational neuroscience are low demanding sensory-motor tasks, which easily make the subjects fatigue and slack. In this study, a period of high concentration of users can be maintained by a high demanding boundary avoidance task. Meanwhile, instead of decoding control intention from large limbs’ movements, a finer and more challenging control intention decoding from fingers of the same hand was studied under this high demanding task. Specifically, EEG signals during control period of fingers within the same hand are collected under the boundary avoidance task. A novel hybrid model of CNN-Bi-LSTM is proposed for decoding dexterous EEG-based control intention. The experimental results from ten subjects assessed by 5-fold cross-validation method show that analyzing the EEG signals by the proposed model can enable accurate decoding of fingers control intentions within the same hand in this boundary avoidance task.

Keywords: Electroencephalography (EEG), boundary avoidance task (BAT), dexterous control, hybrid model of CNN-Bi-LSTM