CRNN-Based Classification of EMG Signals for the Rehabilitation of the Human Arm

EMG signals have found applications in various fields like prosthetics and human–machine interfaces. In this study, we propose a Convolutional Recurrent Neural Network (CRNN) that leverages the strengths of both Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks. The CRNN model enables spatial feature extraction through CNNs and captures temporal patterns in EMG data using LSTM networks. Our aim is to utilize this model for human hand rehabilitation with the assistance of an exoskeleton robot. To accurately assess the performance of the CRNN approach, it is essential to compare it with existing studies in the field of EMG recognition. Our results, achieved on the Ninapro Dataset DB2, demonstrate great promise and are comparable to methods such as CNNs. Furthermore, the CRNN outperforms machine learning algorithms like Support Vector Machines (SVMs) and Random Forests. An important advantage of CRNNs is their ability to effectively capture the temporal dynamics of EMG signals, which cannot be achieved with standard CNN models. This study highlights the potential and effectiveness of CRNNs in classifying EMG signals. By combining the strengths of CNNs and LSTMs, the adopted CRNN-based approach proves to be valuable in EMG classification.