Deep-Learning Based Wearable 8-Channel sEMG Biometrics Modality for User's Verification

Recently, deep learning algorithms have gained prominence for their capacity to learn in diverse applications. However, in electromyography (EMG)-based biometrics, deep learning has not been widely involved, due to the extensive effort required to generate large databases. In this paper, data augmentation is utilized to expand a database, in order to reduce the data collection effort. A Convolutional Neural Network (CNN) is employed to accurately learn user characteristics within an EMG-based biometric system, with the SqueezeNet architecture chosen for its faster training time and reduced parameter requirements. Wavelet Transforms are applied to generate scalograms of the EMG signals. The system is evaluated in terms of testing accuracy, False Acceptance Rate, and False Rejection Rate. A classifier accuracy of 98.31% has been achieved.