A Scalable Open-Set ECG Identification System Based on Compressed CNNs.

Area of Science:

• Biometrics
• Machine Learning
• Signal Processing

Background:

• Deep learning (DL) excels at feature learning but faces challenges in ECG biometric recognition.
• Existing methods struggle with open-set identification and efficient enrollment of new subjects.

Purpose of the Study:

• To apply DL, specifically CNNs, for robust ECG biometric identification.
• To enable open-set identification and efficient, scalable enrollment of new users.
• To optimize CNN models using quantum evolutionary algorithms (QEA) for reduced computational cost.

Main Methods:

• Utilized convolutional neural networks (CNNs) for ECG feature extraction and identification.
• Implemented a scheme for open-set identification using prestored user-specific feature vectors.
• Employed transfer learning for scalable enrollment of new subjects.
• Applied quantum evolutionary algorithms (QEA) for pruning CNN filters.

Main Results:

• Achieved over 99% identification accuracy in closed-set scenarios using the PTB dataset (285 subjects).
• Demonstrated effective exclusion of unregistered subjects, enhancing resistance to dictionary attacks.
• QEA-based filter pruning reduced computational operations to 1.20% and 0.22% without significant accuracy loss.

Conclusions:

• The proposed DL-based ECG biometric system offers high accuracy and security.
• The approach supports efficient open-set identification and scalable user enrollment.
• QEA-based model optimization significantly enhances computational efficiency for ECG biometrics.