Enhancing Adversarial Attack Detection in EEG Signals With Covariance Entropy: A Novel Framework for BCI Security

Brain-computer interfaces (BCIs) facilitate direct brain-to-external device connection by using machine learning to interpret EEG data. However, these systems are vulnerable to adversarial attacks that can lead to faulty outputs and potentially severe consequences. This work presents a new method, Covariance Entropy (CovEn), for measuring an EEG signal's uncertainty. It provides a more elaborative analysis of signal complexity compared to other entropy measures since it accounts for inter-channel variance and explicit correlation in EEG data. In our research, we used a variety of noisy and adversarial attacks to show that CovEn is a valuable tool for detecting signal contamination. This work introduces the advancement of reliable and safe neurotechnologies by highlighting the potential of CovEn as a protective mechanism against adversarial threats in multivariate signals.