Identificação Biométrica A Partir De ECG De Derivação Única Utilizando Perceptron Multicamadas e Transformada de Stockwell

Abstract

Nowadays digital security systems are ordinary tools on a day by day, going by biometrics such as digital and facial recognition to the traditional users and passwords methods. Noticing the popularization of wearables, this paper proposes a biometric recognition method using ECG signals that can be obtained by the current technology of smartwatches, that differ from traditional 12-lead ECGs. First it was used a blind segmentation method to the electrocardiogram signal, which was treated with a Butterworth filter to attenuate signal noise. It was applied the Stockwell transform to the signal, shifting it from time domain to time- frequency domain, therefore obtaining the unidimensional complex trajectories of each sample as the MLP’s input. The proposed MLP model were able to obtain an 95,96% of accuracy using a data window of 500 discrete samples, showing its ability to set apart each subject signal.