A Study on the Feasibility of User Authentication Using EMG Signals from Finger Flexor Muscles

손가락 굴곡 근육의 근전도 신호를 활용한 사용자 인증 가능성 연구

Recently, biometric signals such as facial recognition and fingerprint recognition have become widely used for user authentication. However, as spoofing attacks targeting these biometric signals have increased, serious issues such as identity theft and financial fraud have arisen. To address this, this paper proposes a more secure user authentication method using sEMG signals, which, even for the same muscle, vary across individuals. Through this approach, the potential of sEMG signals for user authentication is validated. A Siamese network is employed to allow authentication even with a small amount of data. The participants performed a fist-clenching motion 120 times, and the data pairs were created such that pairs from the same person were labeled as 1, and pairs from different people were labeled as 0. Using this data, five-fold cross-validation was conducted on the Siamese network, achieving an average accuracy of 97.37%. During testing, an average accuracy of 95.83% was observed. Additionally, the system demonstrated excellent performance, with a Precision of 99.89%, Recall of 92.99%, and F1 Score of 96.28%.