Abstract
In Indonesia, many people experience upper-limb functional loss, which limits their ability to perform daily activities. Myoelectric prosthetic hands offer a promising solution for transradial amputees by using electromyography (EMG) signals to control hand movements. However, conventional wired systems may introduce noise that affects EMG signal quality and gesture recognition performance. Therefore, this study proposes an enhanced hand gesture classification approach for myoelectric prosthetic hands using wearable EMG sensors, Savitzky-Golay (SG) filtering, and a compact convolutional neural network (CNN). The dataset consists of EMG signals acquired from eight subjects using an eight-channel wearable EMG sensor, covering four hand gestures: rock, paper, scissors, and OK. The SG filter is applied to smooth the EMG signals and reduce noise while preserving important signal characteristics. Since filtering alone is insufficient to distinguish all gesture types, a compact CNN model is employed for classification. The proposed model achieved an accuracy of 95.15%, precision of 95.16%, recall of 95.15%, and F1-score of 95.15%, demonstrating its effectiveness for EMG-based hand gesture classification in myoelectric prosthetic hand applications.