ECG Authentication Using Neural Network Wave Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing wearable devices face challenges in authenticating users effectively using electrocardiogram (ECG) signals due to variations in heart state and noise, which affect the accuracy of user identification.
Innovation Solution
An authentication apparatus utilizing a neural network that separates and processes ECG signals into P wave, QRS wave, and T wave components, applying machine learning to extract feature values and compare them with registered user signals, employing a Siamese neural network and k-nearest neighborhood algorithm to determine user authentication based on similarity ratios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If ECG signals are used for user authentication in wearable devices, then user identification capability is provided, but authentication accuracy deteriorates due to heart state variations and noise
Solution Approach 1:
The patent segments the ECG signal into distinct wave components (P wave, QRS complex, T wave) and processes each segment separately through dedicated neural networks. This segmentation allows the system to focus on specific characteristic features of each wave type, improving authentication accuracy despite variations in heart state and noise interference.
Solution Approach 2:
The patent introduces a Siamese neural network as an intermediary that compares the input ECG signal with reference ECG signals stored in the database. This intermediary mechanism enables the system to evaluate similarity ratios and determine authentication results by comparing characteristic features, thereby maintaining high accuracy even when heart state variations occur.
2Measurement precision
If neural network processing is applied to separate ECG wave components, then authentication accuracy is improved, but device complexity increases
Solution Approach 1:
The patent divides the complex ECG signal processing task into multiple simpler sub-tasks, with separate neural networks handling each wave component (P wave, QRS complex, T wave). This segmentation reduces the complexity of each individual neural network while maintaining overall authentication accuracy through the combined processing of all wave segments.
Solution Approach 2:
The patent employs a Siamese neural network architecture that serves multiple functions: it processes the input ECG signal, compares it with multiple reference signals, and determines authentication results. This multi-functional approach consolidates several processing operations into a single unified system, reducing overall device complexity.
3Reliability
If multiple reference ECG signals are stored and compared, then authentication reliability is improved, but processing time increases
Solution Approach 1:
The patent extracts and stores only the essential characteristic features of reference ECG signals in the database, rather than storing complete raw signals. During authentication, the system compares the input signal's extracted features with these stored characteristics, significantly reducing processing time while maintaining authentication reliability through the use of key identifying features.
Solution Approach 2:
The patent processes only a subset of the stored reference ECG signals by selecting those with the highest similarity ratios from the database. Rather than comparing the input signal with every single reference signal, the system performs partial comparisons with the most promising candidates, reducing processing time while maintaining high reliability through the use of similarity threshold criteria.
Data Source
AI summary
An authentication apparatus includes one or more processors configured to temporally implement a neural network, used to extract a feature value from hidden nodes, that is connected to input nodes to which an electrocardiogram (ECG) signal is input so as to share a weight set with the input nodes, and to match the ECG signal and the extracted feature value to a user for registration.