ELEKTRA: ELEKTRokardiomatrix Application to Biometric Identification with Convolutional Neural Networks
Published in Neurocomputing, 2022
Recommended citation: Fuster-Barceló, C., Peris-Lopez, P., & Camara, C. (2022). ELEKTRA: ELEKTRokardiomatrix application to biometric identification with convolutional neural networks. *Neurocomputing, 506*, 37–49. https://doi.org/10.1016/j.neucom.2022.07.059 https://doi.org/10.1016/j.neucom.2022.07.059
ELEKTRA: ELEKTRokardiomatrix Application to Biometric Identification with Convolutional Neural Networks
Biometric systems are becoming an increasingly common identification method in everyday life. We introduce a novel approach using Elektrokardiogramms (ECGs), transformed into heatmaps of aligned R-peaks (heartbeats), creating a matrix we call an Elektrokardiomatrix (EKM).
Our method uses a simple Convolutional Neural Network (CNN) — just one convolutional layer and pooling — to identify users with outstanding accuracy. Evaluated on three datasets (NSRDB, MIT-BIH, and PTB), we achieved:
- 99.53% accuracy (NSRDB)
- 0.02% FAR, 0.05% FRR
This lightweight solution delivers high performance with low computational cost. It shows strong potential as a secure and efficient biometric identification system.