A Classification and Recognition Method for Motor Imagery EEG Signals Using Fusion of DRSN and WPD

Noise reduction is a key factor in determining the effectiveness of motor imagery (MI) EEG signal classification and recognition. Deep Residual Shrinkage Network (DRSN) is suitable for denoising MI EEG signals due to its soft thresholding and attention mechanism characteristics. However, there is a lack of intermediate scale features in its processing, which are crucial for examining functional connectivity between brain regions. Therefore, this article proposes a multi-scale feature learning method that combines DRSN and Wavelet Packet Decomposition (WPD) to enhance DRSN's ability to capture intermediate scale features. Through WPD, EEG signals are decomposed into different frequency bands, covering from () wave to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$y$</tex> wave. The different frequency ranges of waves enable the network to effectively capture intermediate scale signal features of dynamic changes such as event related potentials, EEG synchronization and desynchronization. The experimental results show that the average accuracy of DRSN on the BCI competition IV 2a and 2b datasets is increased to 84.45% and 91.26% after the introduction of WPD, which can ensure the effective recognition of MI EEG signals.