Low-computational EMG gesture recognition for prosthetic control via handcrafted features and lightweight MLP

• A low-computation method for EMG signal classification is proposed for enhanced prosthetic control and rehabilitation. • Manual feature extraction significantly reduces computational overhead while maintaining high classification accuracy. • The streamlined three-layer MLP achieves 97.7% accuracy using selected time and frequency domain features. • The approach enables real-time, cost-effective deployment on low-power hardware for broader accessibility in assistive technologies. This study introduces an efficient and accurate approach to classifying electromyography (EMG) signals for advanced prosthetic control and rehabilitation, addressing the need for practical, real-time human-computer interaction (HCI) systems. Unlike conventional deep learning methods that require substantial computational power, our method significantly reduces computational overhead through handcrafted feature extraction. Specifically, we extract four discriminative features from both time and frequency domains—waveform length (WL), absolute value of the summation of the exponential root (ASM), absolute value of the summation of the square root (ASS), and amplitude of the first dominant frequency (AFDF). These features are used to train a compact three-layer Multilayer Perceptron (MLP) neural network, which achieves a classification accuracy of 97.7% across six common hand gestures. The system demonstrates real-time viability with an average inference time of approximately 3.2 milliseconds per gesture on standard consumer hardware (Intel Core i5, 8 GB RAM). The low computational complexity enables deployment on affordable embedded platforms, making this approach a strong candidate for use in wearable prosthetics and rehabilitation tools. Our results validate the potential of this lightweight, accessible method for enhancing the functionality of assistive devices and improving quality of life for individuals with motor impairments.