Visual Explanations of Deep Convolutional Neural Network for EEG Brain Fingerprint

Brain fingerprint with electroencephalogram (EEG) is widely employed in identification. However, the security of brain fingerprint and the identification system is greatly reduced due to their unclear mechanism. Thus, this study conducted a visual explainable research to comprehend the key features the model focuses on for identification. We used CNN to extract brain fingerprints for identification, with an accuracy of 98.06%. To explain the brain fingerprints, we used the gradient-weighted class activation mapping (Grad-CAM) method and found very meaningful visualization results. Limited to the motor imagery (MI) dataset we collected, the most significant EEG segments correspond to the phases of the presented MI instruction, and the most effective channels correspond to motor areas of the human cerebral cortex. Our findings demonstrate that the visual explainable research provides an understanding of which features are better involved in the model learned behavior and provides insights into the neural processes.