Enhancing Hand Gesture Recognition Accuracy from Brain Signals Using Sparse Transforms

Abstract

Brain-Computer Interfaces (BCIs) offer a framework for classifying motor tasks such as hand movements and extracting features from electroencephalogram (EEG) signals. However, excessive noise levels and the processing requirements of existing approaches sometimes limit the accuracy of these classifications. For individuals with spinal cord injuries, whose limb movements are restricted, this challenge is particularly significant. Highly accurate classification algorithms are necessary for precise EEG-based control of neural prostheses. In this study, we propose a novel approach to EEG signal-based hand movement detection. Empirical Mode Decomposition (EMD) is used to preprocess EEG data before wavelet scattering and the Discrete Wavelet Transform (DWT) are applied to recover sparse features. Then, the Whale Optimization Algorithm (WOA) is employed to efficiently select features and reduce dimensionality. Three widely utilized algorithms—Random Forest, K-Nearest Neighbor (KNN) and Support Vector Machine (SVM)—are used to classify the refined features. According to experimental data, the proposed method classifies hand gestures with an accuracy of about 98% and a precision of about 93% using Random Forest. Comparative studies demonstrate that this approach outperforms most of the compared methods for identifying and categorizing hand movements in terms of accuracy and efficiency. These results show great potential for developing brain prosthetic devices and enhancing BCI systems.