Visual Brain Decoding for Short Duration EEG Signals

In this work, we propose a CNN-based approach for classification of short duration EEG signals for visual brain decoding. These signals are captured for a visual perception task by showing digit images on a computer screen, and the task involves classification of the EEG signals into 10 classes, corresponding to the digits shown. The captured EEG signals are of very short duration (approx. 2sec), which are typically very noisy. We use a correlation based technique for the removal of highly noisy samples. Further, a sample refinement approach for the selection of relevant channels is also proposed. Both these steps constitute the data refinement process, which we demonstrate has a significant effect on the CNN classification performance. We validate the proposed approach on a publicly available MindBigData (The “MNIST” of Brain Digits) dataset.