A Capsule Decision Neural Network Based on Transfer Learning for EEG Signal Classification.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Brain-computer interfaces (BCI) require robust methods to handle individual differences in neural signals.
• Existing transfer learning techniques face challenges with feature distortion in electroencephalography (EEG) signal processing.
• Developing adaptive algorithms is crucial for personalized BCI applications.

Purpose of the Study:

• To propose a novel capsule decision neural network (CDNN) that utilizes transfer learning for enhanced BCI performance.
• To address the issue of feature distortion in EEG signal extraction using a deep capsule network architecture.
• To improve the independent decoding capabilities of BCI systems for individual users' EEG signals.

Main Methods:

• A deep capsule decision network (CDNN) architecture was constructed with primary capsules and a neural decision routing algorithm.
• The neural decision network computes capsule relationships probabilistically, differing from traditional dynamic routing.
• EEG covariance matrix distribution alignment in Riemann space and a regional adaptive method were employed.

Main Results:

• The proposed CDNN effectively handles individual differences and feature distortion in EEG signals.
• The neural decision routing algorithm demonstrated superior performance compared to dynamic routing.
• Experiments on two motor imagery EEG datasets confirmed CDNN's superior performance over advanced transfer learning methods.

Conclusions:

• The developed CDNN offers a promising approach for improving transfer learning in BCI applications.
• The probabilistic routing and Riemann space alignment enhance EEG signal decoding accuracy.
• CDNN shows significant potential for personalized and effective brain-computer interface systems.