Bioelectric Signal Regression for Human Alertness Transitions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning methods for detecting human fatigue based on brain electrical signals (EEG) are less accurate and may not provide reliable judgments, especially when the human body is on the edge of fatigue or has just entered a fatigued state, leading to potential safety hazards.
Innovation Solution
A method and device using a regression neural network to estimate human alertness by obtaining bioelectric signals, such as EEG, EOG, and EMG, preprocessing them to remove noise, extracting feature data, and employing an encoder-decoder architecture with an attention mechanism to estimate the alertness value.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If classification models (CNN or RNN) are used to detect fatigue state, then the detection method is simple to implement, but the detection accuracy is insufficient especially at fatigue transition states
Solution Approach 1:
The patent changes the fundamental parameter of the detection task from classification (discrete categories) to regression (continuous alertness values 0-1). This allows the model to output precise alertness levels rather than simple fatigued/not-fatigued classifications, significantly improving detection accuracy at transition states while using a relatively simple neural network architecture.
Solution Approach 2:
The patent introduces a dynamic alertness estimation approach where the model outputs continuous values that can adapt to varying fatigue states. The regression framework allows the system to dynamically adjust alertness predictions based on input features, capturing subtle transitions that classification models miss.
2Reliability
If classification tasks are used to predict fatigue state, then the computational process is simple, but the model cannot make accurate judgments at the edge of fatigue states
Solution Approach 1:
The patent transforms the output parameter from discrete classification labels to continuous alertness values ranging from 0 to 1. This continuous parameter allows the model to express uncertainty and partial states, making judgments at fatigue edges more reliable while maintaining computational simplicity through standard regression neural network architecture.
3Measurement precision
If simple classification models are used, then the processing speed is fast, but the detection accuracy drops when human body is on the edge of fatigue
Solution Approach 1:
The patent uses regression to output continuous alertness values directly, avoiding the multiple processing steps required for classification (prediction, threshold comparison, category assignment). This direct regression approach maintains fast processing speed while achieving higher accuracy in detecting subtle fatigue transitions.
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
The present invention provides a method for estimating the alertness of the human body. The method comprises: Obtaining a bioelectric signal from the human body; obtaining feature data for estimating the alertness of the human body based on the bioelectric signal; and estimating the alertness value of the human body using a regression neural network based on the feature data of the bioelectric signal.