An EigenECG Network Approach Based on PCANet for Personal Identification from ECG Signal

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Personal identification using biosignals is crucial for security and healthcare.
• Electrocardiogram (ECG) signals offer unique physiological characteristics for identification.
• Existing methods often require complex feature extraction or back-propagation.

Purpose of the Study:

• To propose a novel EigenECG Network (EECGNet) for robust ECG-based personal identification.
• To develop a method that efficiently extracts features without relying on back-propagation.
• To evaluate the performance of EECGNet against conventional identification algorithms.

Main Methods:

• ECG signals are preprocessed (normalization, spike removal) and R-peak detected.
• Signals are transformed into 2D images for input into the EECGNet.
• The network employs cascaded Principal Component Analysis (PCA) stages, followed by quantization and histogram computation.

Main Results:

• The proposed EECGNet demonstrated strong performance in personal identification.
• Experimental results showed EECGNet outperforms PCA, Auto-Encoder (AE), Extreme Learning Machine (ELM), and Ensemble Extreme Learning Machine (EELM).
• The method effectively extracts features from visual representations of ECG data.

Conclusions:

• EECGNet provides an effective and efficient approach for ECG-based personal identification.
• The network's ability to perform without back-propagation simplifies implementation and reduces computational cost.
• This method holds significant potential for secure and reliable biometric authentication systems.