MST-net: A multi-scale swin transformer network for EEG-based cognitive load assessment

Area of Science:

• Neuroscience
• Computer Science
• Human-Computer Interaction

Background:

• Cognitive load assessment is vital for safe production, resource management, and human-computer interaction.
• Electroencephalography (EEG) is a preferred method for brain monitoring due to its high temporal resolution and ease of use.

Purpose of the Study:

• To propose and validate a novel Multi-scale Swin Transformer network (MST-Net) for accurate cognitive load assessment using EEG signals.

Main Methods:

• Developed MST-Net integrating multi-scale parallel convolution for local feature extraction and Swin Transformer attention for multi-scale feature correlation.
• Validated MST-Net performance on EEG data from cognitive and N-back tasks with varying load levels.

Main Results:

• MST-Net demonstrated superior classification accuracy compared to standard Swin Transformer and Convolutional Neural Network (CNN) models on both local and public datasets.
• Ablation studies and feature visualization confirmed MST-Net's capability in effectively distinguishing different cognitive loads.

Conclusions:

• The proposed MST-Net offers a powerful new tool for cognitive load assessment.
• MST-Net shows significant potential for widespread application in brain-computer interface (BCI) systems.