Study of EMG-based Authentication Algorithm Using Artificial Neural Network

Abstract

사용자 인증을 위한 패스워드 입력 방식은 해킹에 취약하다. 그래서 심전도, 근전도와 같은 생체 신호를 이용한 보안 기술이 개발되고 있다. 본 연구에서는 개인 인증 기술의 취약점을 보완하기 위해 근전도를 이용한 개인 인증 알고리즘을 제안한다. 인증률을 향상시키기 위해 인공신경망 알고리즘을 사용하였다. 이 방식은 전처리, 특징점 추출, 분류 등의 과정을 포함한다. 개인 인증은 근전도 신호로부터 다섯 가지의 특징점을 추출하여 진행하였다. 제안된 알고리즘은 실험을 통해 95.0 %의 정확도를 나타내었다. 진행하였다. 제안된 알고리즘은 실험을 통해 95.0 %의 정확도를 나타내었다.|Biometric authentication, which uses body parts as a password, does not require memorization and is widely used owing to its excellent security. Primarily, fingerprints, face, iris, voice, etc. are used for personal authentication.

However, when a vulnerability is found in each method, an authentication method for securing it is required. Therefore, this study proposes a personal authentication method using EMG.

EMG is a µV-sized electrical signal that occurs in the muscles as the body moves. This signal can detect muscle health, nerve abnormalities, and body movements. EMG data were acquired from the right arm after constructing multiple measurement channels. Further, the operation of holding the fist was repeated for several times while measuring the data. The EMG data were acquired using the EMG measurement module that was directly produced.

To improve the personal authentication rate, we designed a digital filter using Matlab and applied it to the EMG signals. To classify the individuals, Variance value, Mean value, Zero crossing value, Length value, and Median Frequency value parameters were extracted from the EMG signals.

The extracted parameters were classified into artificial neural networks that implemented the program for brain information processing. The model used for the artificial neural network is a feedforward neural network. The result of the EMG-based personal authentication showed 95.0 % accuracy.