GestProNet: A Hybrid Deep Learning Framework for High-Precision EMG-Based Hand Gesture Classification

The hand gesture classification system utilizes electromyography (EMG) signals as a robust tool for recognizing and evaluating muscle activity patterns. Challenges such as signal variation, noise, and high computation requirements often pose problems to the recognition performance of traditional methods. To tackle these problems, this paper presents GestProNet, an advanced hybrid deep learning framework for hand gesture classification based on EMG signals. It utilizes a hybrid architecture combining convolutional neural network (CNN) and Progressive Feedback Residual Attention Network with Snow Geese Algorithm (CNN-PFRAN+SGA). The proposed framework employs Adaptive and Propagated Mesh Filtering (APMF) for preprocessing to enhance signal quality, followed by feature extraction using the Inception Transformer (IT) to capture critical EMG patterns. The CNN-PFRAN model performs gesture classification, while the Snow Geese Algorithm (SGA) optimizes hyperparameters to improve accuracy and efficiency. Evaluated on the UC2018 DualMyo benchmark dataset, the proposed system demonstrates state-of-the-art performance, achieving exceptional metrics of 99.9% accuracy, 99.5% specificity, and 99.7% precision. The model also demonstrates superior computational efficiency, with a processing time of 0.20 seconds and an error rate of 0.2%. The system sets new benchmarks in EMG gesture recognition, outperforming existing methods. Its robustness shows strong potential for prosthetics, rehabilitation, and human computer interface (HCI), with future work focusing on optimization, adaptability, and real-time implementation.