Deep-BBiIdNet: Behavioral Biometric Identification Method Using Forearm Electromyography Signal

Abstract

The purpose of this study is to use behavioral biometric features of bioelectric signals for classifying and identifying people. This paper describes the development of a high-predictive-accuracy Convolutional Neural Network (CNN)-based human recognition system. When four participants make seven separate finger/wrist movements, their bioelectric signals are captured and recorded by the bio-armband sensor. The developed CNN model is used to generate features from EMG signals and to classify humans. The model's success is assessed in two cases in the study. In the first case, separate human classification is made for seven different movements. In the second case, the human classification performance is tested using the entire dataset, regardless of the movement type. It is observed that the average accuracy of 98.833%, 99.166%, 98.333%, 100%, 99.708%, and 99.791% is reached for seven different subcases in Case 1, respectively. It is observed that the network model developed for Case 2 has 100% accuracy in human classification/identification, 100% recall, 100% sensitivity, and 100% F1-score performance.