Grad Seminar: Open-access Dataset and Convolutional Feature Engineering for Long-term Biometric Authentication using Wrist Electromyography
Abstract
Surface electromyography (EMG) has been proposed as a novel biometric method. To address the performance degradation of EMG-based biometric methods in multi-day scenarios, robust EMG feature extraction methods are warranted. The advancements in artificial intelligence have led to the adoption of deep feature extraction and classification methods in the biosignal domain. Data was collected from 43 participants on three different days across one month while performing hand/wrist gestures. The data is made open access as the GRABMyo Dataset. This study proposes a convolutional feature extraction method and a two-stage network for an authentication application. The cross-day analysis involving training data and testing data from different days was performed to test the robustness of the EMG-based biometrics in practical scenarios. In cross-day authentication, the proposed method resulted in a median equal error rate (EER) of 0.003 and 0.008 when the gesture (code) is safe and compromised, respectively. These values were significantly lower than the traditional frequency domain feature sets. The results demonstrated the accurate and robust performance of convolutional engineering for more practical EMG-based biometric applications.
Presenter
Ashirbad Pradhan, PhD candidate in Systems Design Engineering
Join in-person or online via Microsoft Teams