A dual-branch deep learning model based on fNIRS for assessing 3D visual fatigue.
Yan Wu1,2,3, TianQi Mu1, SongNan Qu1
1School of Computer Science and Technology, Changchun University of Science and Technology, Changchun, China.
Frontiers in Neuroscience
|June 20, 2025
Summary
A new deep learning model using functional near-infrared spectroscopy (fNIRS) can automatically detect 3D visual fatigue. This advanced method improves user experience in stereoscopic 3D technology by accurately assessing fatigue levels.
Area of Science:
- Neuroscience
- Computer Science
- Human-Computer Interaction
Background:
- Extended 3D content viewing causes visual fatigue, impacting user experience and stereoscopic 3D technology performance.
- Functional near-infrared spectroscopy (fNIRS) shows promise for assessing 3D visual fatigue by measuring cerebral hemodynamic responses.
- Traditional fNIRS methods require manual feature extraction, limiting their efficiency and effectiveness.
Purpose of the Study:
- To develop a novel deep learning framework for automated 3D visual fatigue evaluation using fNIRS data.
- To overcome the limitations of manual feature extraction in traditional fNIRS-based fatigue assessment.
- To enhance user experience and optimize stereoscopic 3D technology through accurate fatigue detection.
Main Methods:
- An fNIRS-based experimental paradigm was designed to collect data under both comfort and fatigue conditions from 20 subjects.
- A dual-branch convolutional network was employed to extract temporal and spatial features from time-series fNIRS data.
- A transformer module and channel attention mechanism were integrated to enhance long-range dependency extraction and adaptive feature weighting.
Main Results:
- The proposed deep learning model achieved high classification accuracy: 93.12% within subjects and 84.65% across subjects.
- The model demonstrated superior performance compared to traditional machine learning and existing deep learning approaches.
- Automated end-to-end feature extraction and classification were successfully enabled, removing the need for manual intervention.
Conclusions:
- A novel deep learning framework for automatic 3D visual fatigue evaluation using fNIRS was successfully constructed.
- The integrated transformer and channel attention mechanisms significantly improved the model's feature extraction capabilities.
- The framework offers a valuable tool for enhancing user experience in stereoscopic 3D applications and advancing fatigue assessment research.
