Residual Self-Calibrated Network With Multi-Scale Channel Attention for Accurate EOG-Based Eye Movement...

Area of Science:

• Biomedical Engineering
• Computer Science
• Human-Computer Interaction

Background:

• Electrooculography-based Human-Computer Interaction (EOG-HCI) is gaining traction in fields like assistive robotics and augmented reality.
• Accurate eye movement classification (EMC) is crucial for EOG-HCI but is hindered by limitations in extracting discriminative features.

Purpose of the Study:

• To propose an efficient deep learning model for enhanced feature extraction and performance improvement in EOG-based EMC.
• To establish a new state-of-the-art benchmark for EOG-based EMC.

Main Methods:

• Development of a Residual Self-Calibrated Network with Multi-Scale Channel Attention (RSCA).
• Utilizing self-calibrated convolution blocks within a hierarchical residual framework for multi-scale feature extraction.
• Incorporating a multi-scale channel attention module to adaptively weight features and aggregate multi-scale context.

Main Results:

• The RSCA network demonstrated significant performance improvements over seven prevailing methods across five public datasets.
• Ablation studies confirmed the effectiveness of the individual modules within the RSCA architecture.
• The RSCA network established a new state-of-the-art benchmark for EOG-based EMC.

Conclusions:

• The proposed RSCA network effectively addresses the challenge of discriminative feature extraction in EOG-based EMC.
• The findings offer valuable insights for designing advanced deep learning models for EOG-HCI applications.
• RSCA significantly advances the accuracy and potential of EOG-HCI technology.