DWT and CNN based multi-class motor imagery electroencephalographic signal recognition

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Brain-computer interface (BCI) systems enable control of external devices via motor imagery (MI) signals.
• Existing feature extraction algorithms struggle to account for individual differences, limiting BCI universality.
• There is a need for robust algorithms that minimize inter-subject variability in BCI performance.

Purpose of the Study:

• To develop and validate a novel processing algorithm for BCI systems.
• To reduce the impact of individual differences on MI signal classification.
• To improve the accuracy and universality of BCI technology.

Main Methods:

• Discrete wavelet transform was used to determine the optimal frequency band by calculating sub-band energy.
• Power spectral density was employed for feature extraction.
• A convolutional neural network based on a visual geometric group network was utilized for classification.

Main Results:

• The proposed algorithm demonstrated superior performance on the BCI Competition IV dataset IIa.
• It achieved an average classification accuracy rate of 96.21%, surpassing current state-of-the-art methods.
• The algorithm reduced classification calculation time and improved accuracy compared to conventional techniques.

Conclusions:

• The algorithm's effectiveness stems from a reduced parameter count, lower computational resource demands, and mitigation of individual differences.
• This approach enhances classification performance in BCI systems.
• The developed algorithm is suitable for real-time, multi-class BCI applications.