Spatio-Temporal Progressive Attention Model for EEG Classification in Rapid Serial Visual Presentation Task.

Area of Science:

• Neuroscience
• Computer Science
• Signal Processing

Background:

• Electroencephalogram (EEG) signals are multi-dimensional sequential data.
• Investigating spatial and temporal dependencies in EEG is crucial for accurate classification.
• Rapid serial visual presentation (RSVP) tasks present challenges due to complex signal patterns.

Purpose of the Study:

• To propose a novel spatial-temporal progressive attention model (STPAM) for enhanced EEG classification in RSVP tasks.
• To develop a new Infrared RSVP Dataset (IRED) for evaluating EEG classification models.
• To improve the understanding and modeling of spatial and temporal dependencies in EEG signals.

Main Methods:

• Developed a spatial-temporal progressive attention model (STPAM) with sequential spatial and temporal experts.
• Employed a progressive approach for refining EEG electrode selection and focusing on significant spatial information.
• Utilized attention mechanisms to capture crucial temporal dependencies in EEG time slices.
• Introduced a novel Infrared RSVP Dataset (IRED) using dim infrared images with small targets.

Main Results:

• The proposed STPAM model demonstrated superior performance compared to all baseline methods.
• STPAM achieved a 2.02% improvement on a public dataset.
• STPAM achieved a 1.17% improvement on the newly created IRED dataset.

Conclusions:

• The STPAM model effectively captures spatial and temporal dependencies in EEG signals for improved classification.
• The novel IRED dataset provides a valuable resource for future research in EEG-based RSVP tasks.
• The progressive attention mechanism offers a promising direction for advanced EEG signal analysis.