ECG Authentication Using Neural Network Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current biometric authentication methods using electrocardiogram (ECG) signals face challenges in accurately identifying individuals due to variations in heart rate and noise in ECG signals, which affect the robustness and reliability of authentication processes.
Innovation Solution
The proposed solution involves preprocessing ECG signals using band pass filters to remove noise, detecting fiducial points, and extracting semantic features using a neural network model trained on augmented ECG data. This model then compares the extracted features with registered features to determine authentication success based on similarity thresholds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If ECG signals are used for biometric authentication, then individual identification capability is provided, but noise in ECG signals and heart rate variations reduce authentication accuracy and reliability
Solution Approach 1:
The ECG signal is segmented into multiple data segments based on detected fiducial points (peak points and minimum points). This segmentation allows the system to extract multiple features from different portions of the ECG signal, thereby improving authentication accuracy despite noise and heart rate variations.
Solution Approach 2:
The system extracts multiple parameters including peak-to-peak intervals, amplitudes, and areas under curves from different segments of the ECG signal. By changing and comparing multiple parameters rather than relying on a single measurement, the system achieves more reliable authentication despite variations in heart rate and signal noise.
2Measurement precision
If multiple features are extracted from ECG signals to improve authentication accuracy, then authentication reliability is enhanced, but the complexity of signal processing increases
Solution Approach 1:
The system performs preliminary detection of fiducial points (peak points and minimum points) in the ECG signal before extracting features. This preliminary action identifies key locations in the signal, allowing subsequent feature extraction to focus only on relevant segments and simplifying the overall processing complexity while maintaining high accuracy.
3Reliability
If ECG signals are preprocessed with filtering and fiducial point detection to reduce noise, then authentication reliability is improved, but processing time and computational load increase
Solution Approach 1:
The system extracts and focuses processing on specific fiducial points (peak points and minimum points) rather than processing the entire ECG signal. By taking out only the critical portions of the signal containing authentication information, the system reduces processing time and computational load while maintaining high authentication reliability.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Disclosed are an electrocardiogram (ECG) authentication method and apparatus, and a training method and apparatus for training a neural network model used for ECG authentication, the ECG authentication apparatus being configured to acquire an ECG signal of a subject, extract a semantic feature of the ECG signal, and authenticate the subject based on the extracted semantic feature.