Integrated EMD-Enhanced Analysis and Neural Network Classification for Robust Processing of Motor Imagery EEG Signals

The rapid evolution of deep learning in the current era has prompted extensive exploration in the realm of Brain-Computer Interface (BCI). Despite the substantial progress made in enhancing the accuracy of BCI systems employing Motor Imagery (MI) through deep learning techniques, especially when compared to certain traditional algorithms, there remains a substantial challenge in clearly interpreting these sophisticated models. This holds particular importance due to the labor-intensive and costly nature of the data collection process. This challenge is addressed in our manuscript through the introduction of a unique methodology to improve the effectiveness and precision of BCI systems using EEG data. The outlined methodology comprises five fundamental steps. The signal pre-processing begins with band-pass filtering, succeeded by Empirical Mode Decomposition (EMD) to capture underlying oscillatory components. Subsequently, correlation calculation is conducted for all the Intrinsic Mode Functions (IMFs) with the original signal. The IMF exhibiting the highest correlation factor is selected for feature extraction, and Classification is executed using the adaptive Graph Convolutional Network (GCN). It plays a pivotal role in feature extraction, leveraging its capabilities to capture intricate patterns within EEG signals. Implemented in Python, this model, achieved an accuracy level of 91.3%, underscoring its significance in augmenting the overall functionality of BCI systems based on EEG data.