One-Dimensional Shallow Neural Network Using Non-Fiducial Based Segmented Electrocardiogram for User Identification System

In order to remotely identify users in the public, medical, and access management sectors, recent user recognition technologies have concentrated on biometric signals. Since the electrocardiogram (ECG) signal is a distinct electrophysiological signal produced by each person's body and is hard to fake or alter, it can be used to identify a user specifically. Current ECG-based user identification systems use a fiducial-based segmentation approach for data normalization and identify R peaks based on morphological characteristics. However, because of motion artifacts brought on by the subject's movement, this method fails to identify a clear R peak, which reduces the accuracy of identification. This paper suggests a person identification system based on one-dimensional neural networks that uses periodic non-fiducial-based segmentation data that do not overlap in the time domain in order to overcome the issue of declining peak accuracy of the fiducial-based segmentation data. A preprocessing stage for data denoising, a non-fiducial-based and non-overlap segmentation step, and a step for user classification utilizing a one-dimensional shallow neural network make up the suggested system. Consequently, the most effective and precise method for achieving the best classification results is to use wave modelling for feature extraction. As a new research we are focusing on automatic heartbeat categorization for healthy/unhealthy status based on the authenticated result from the previous stage. This can be achieved by using the Support Vector Classifier (SVM) classifier at the end which will be trained on the average heart rates extracted from the previous stage samples.

Keywords: Electrocardiogram (ECG), Non-Fiducial-Point, 1D-Convolution Network, Preprocessing (1D-CNN), Security, User Identification, Support Vector Classifier (SVM) classifier.