Design and implementation of a subject identification system based on Electroencephalogram

Abstract

Biometrics are essential methods of identifying people nowadays. There are many types of biometrics, such as the classic methods for iris, face and fingerprint; but most of these are not robust or secure. Recently, biometrics based on electroencephalogram signals using machine learning algorithms have proven to be one of the highest quality and robust methods. Electroencephalograms have advantages over traditional modalities as they are extremely difficult to reproduce and cannot be captured stealthily from a distance. This work describes a system capable of acquiring real-time electroencephalogram signals, processing them using the PREP pipeline, to clean them and improve performance, and making subject identity predictions from electroencephalogram signals using different artificial intelligence algorithms. The system is portable, robust, low-cost and connected to the network to send the results to a server. It is composed of an acquisition system using an analog-to-digital converter and protection systems for electroencephalogram signals. The system is based on a Raspberry Pi Zero 2W as the computer in charge of performing all the computational work of the artificial intelligence algorithms and managing the different tasks. Several deep learning algorithms have been used and compared in terms of results and performance. The EEGNet model has provided the best results with an accuracy of 86.74% in its predictions. The data input to the model has been preprocessed with the PREP pipeline, which has proven to be effective in the results, as it improves the performance of all models that use it. The system provides a functional device with outstanding results that leads the way for future work and applications.