A Hybrid Feature Extraction Technique for Optimized Motor Imagery Classification in BCI
Muhammad Ahmed Abbasi,Hafza Faiza Abbasi,Muhammad Zulkifal Aziz,Junzhe Wang,Xiaohua Wu,Xiaojun Yu
DOI: https://doi.org/10.1109/ICICN59530.2023.10392790
2023-01-01
Abstract:Motor Imagery (MI) Classification is critical in Brain-Computer Interface (BCI) systems, allowing mental intentions to control external equipment. In this research, we propose a hybrid feature extraction strategy for optimal MI Classification in BCI. The technique uses Multi-Scale Principal Component Analysis (MSPCA) to denoise and improve the signal’s quality. Following that, the preprocessed signals are subjected to independent applications of the Power Spectral Density (PSD), Continuous Wavelet Transform (CWT), and Hilbert Transform (HT), with each transformation extracting distinct features. These features are then merged to create a complete feature set. AlexNet, a re-known and efficient deep learning architecture, is then used for MI task categorization, which has shown promising results. Experiment findings on a publicly available dataset show that our proposed technique works impressively, with an amazing classification accuracy of around 99.2%.This hybrid strategy has various advantages over traditional methods. First, including MSPCA improves signal quality, reducing the impact of noise and other artifacts on classification performance. Second, combining PSD, CWT, and Hilbert Transform features yields a very comprehensive representation of MI patterns that extracts both spectral and temporal information. Third, by exploiting the capabilities of AlexNet, a cutting-edge deep learning model, excellent classification accuracy is achieved by efficiently learning complicated patterns from the combined feature space. All of these benefits add up to make our hybrid feature extraction technique a highly viable solution for improving MI Classification in BCI systems.