This paper presents a cost-effective signal acquisition circuitry (SAC) for capturing surface electromyography (sEMG) data to classify different hand movements using advanced machine learning algorithms. The SAC, comprising an instrumentation amplifier, a Sallen–Key band-pass filter and a noninverting amplifier, is designed and tested on a portable printed circuit board. The purpose of this paper is to perform feature extraction and data segmentation for effective analysis and processing of the recorded sEMG signals.
The method involves acquiring sEMG signals through electrode placement on muscles and an SAC PCB board, followed by processing in MATLAB to amplify, filter and segment the signals. The features are extracted, and machine learning algorithms are used to classify muscle activity patterns for further analysis.
The sEMG signals are segmented using overlapping windows ranging from 220 to 550 ms with 50% and 75% overlaps. Time-domain features are extracted and fed into three classifiers: bilayered neural network (BNN), cubic support vector machine (CSVM) and weighted K-nearest neighbors. Among these, BNN achieved the highest accuracy of 92.7% with a window size of 550 ms and 75% overlap. By fixing the window size and overlap and using various feature sets, the highest accuracy is attained by BNN 95.1% and CSVM 92.4%.
The results underscore the feasibility of using low-cost SAC-acquired sEMG for movement detection, paving the way for advancements in prosthetic control systems and rehabilitation technologies. This study highlights the importance of effective signal preprocessing, feature extraction and classifier choice in enhancing hand gesture classification accuracy.
