Deep learning classification features visualization for arm movement brain-computer interface

Recently, several studies are focusing on the bimanual based arm movement brain-computer interface (BCI) study. In our previous bimanual BCI studies showed that classifying the arm movement type is significantly improved the BCI performance. In this study, to reduce the classification error, we were using a deep learning model as a classifier and focused to visualize the classification features using the Class Activation Map (CAM), which shows how the deep learning model uses brain information. In addition, we proposed a method for analyzing 3-dimensional brain signal information. Through this, we could understand the major features of the arm movement classifying, and we expected to compare these characteristics related to the neuroscience research field.