Inception Based Deep Learning: Biometric Identification Using Electroencephalography (EEG)

Biometric systems to measure inherent human characteristics in combination with methods of Artificial Intelligence (AI), can result in innovative applications, especially in the fields of remote access systems and human-machine interfaces. One of these involves the use of Electroencephalography (EEG) equipment to monitor and visualize brain activity. Besides traditional EEG applications, such as medical aspects and neuroscience purposes, it can be used to distinguish individuals based on unique and characteristic signals. The ability to recognize a person without their physical attendance is the first step for future remote access to Brain-Computer Interface (BCI) applications. Therefore, in this work, a novel Deep Neural Network (DNN), based on inception modules with a kernel-adapted modification, is developed. Inception-based DNNs recently gained much interest for Time Series Classification (TSC), since they provide comparable performance to very deep Convolutional Neural Networks (CNN) with much less computational complexity. To the best of knowledge, this is the first inception-based DNN developed for authentication purposes using EEG. To validate the proposed network's performance it is compared to three other inception-based DNNs, namely: InceptionTime, EEG-Inception, and EEG-ITNet. Using an anonymized dataset of 22 participants with a length of 1 s per epoch, the proposed network achieves an average recall and precision of 99.1 % and 99.4 %, respectively, outperforming the other DNNs, with EEG-Inception achieving the best results with an average recall and precision of 96.8 % and 97.1 %, respectively.