Self-Supervised EEG Feature Learning for Cross-Subject Alertness Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning models for fatigue detection in humans rely heavily on the quality of feature data extracted from EEG signals, leading to performance issues when applied across different subjects.
Innovation Solution
A method utilizing self-supervised learning to train an alertness feature extraction model, which minimizes the distance in alertness feature space between samples with the same alertness level, improving cross-subject performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning models are used for fatigue detection based on EEG signals, then the detection accuracy for individual subjects can be improved with high-quality feature extraction, but the model performance deteriorates when applied to subjects outside the training data set
Solution Approach 1:
The patent applies self-supervised learning where the model learns to extract alertness features from EEG signals without requiring manual labeling of fatigue states. The system generates its own training signals by identifying temporal patterns and transitions in the EEG data, allowing it to learn robust cross-subject features independently for each subject's data, thereby improving both detection accuracy and cross-subject adaptability
Solution Approach 2:
The patent transforms the approach by changing from supervised learning with fixed labels to self-supervised learning with dynamically generated labels based on temporal patterns. This parameter change in the learning paradigm allows the model to capture subject-specific characteristics while learning generalizable alertness features, resolving the contradiction between individual accuracy and cross-subject performance
2Measurement precision
If feature data extraction from EEG signals is optimized for specific subjects, then detection precision improves for those subjects, but the model complexity increases when adapting to multiple subjects
Solution Approach 1:
The system performs self-supervised learning independently for each subject, where the model automatically adapts to individual characteristics without requiring complex retraining or manual intervention. This self-service approach simplifies the overall system complexity while maintaining high detection precision for each subject by learning from their own temporal patterns
Solution Approach 2:
The patent segments the learning process into subject-specific independent training stages, where each subject's data is processed separately to learn their unique alertness patterns. This segmentation allows the system to maintain simple, specialized models for each subject while avoiding the complexity of creating a single complex model that must accommodate all subjects simultaneously
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
The present invention provides a method for detecting the alertness of the human body. The method comprises receiving a bioelectric signal for detection of the alertness of the human body; using an alertness feature extraction model trained by a self-supervised learning method to obtain the feature data of the bioelectric signal for detection of the alertness of the human body; utilizing the alertness detection model for detection of the alertness of the human body based on the feature data of the bioelectric signal.