비전 트랜스포머를 이용한 단일 리드 심전도 기반 사용자 인증 시스템

Abstract

This study aims to user authentication based on dynamic characteristics using Electrocardiogram (ECG) signals, which are unique to each individual and offer high security potential. Unlike traditional methods focusing on static characteristic like fingerprints, iris, and facial recognition, which are vulnerable to replication and manipulation, our approach utilizes the Vision Transformer (ViT) to effectively learn complex patterns within ECG data, transforming time-series signals into images for input. By extracting high-dimensional feature vectors, we accurately authenticate users. Experiments conducted with real ECG datasets demonstrate the efficiency and practicality of our system. When trained and tested on the Physionet ECG-ID database, the model achieved the following performance metrics for user authentication using a patch sequence length of 8: accuracy of 98.55%, precision of 86.55%, recall of 89.49%, and F1-score of 87.96%.