Analysis of the Effect of the Optimization Method and the Regulating Parameters in the Deep Network to Improve the Classification Accuracy of Finger Movements based on Electromyogram Signal

Hand and finger movements are crucial for daily activities, and their impairment due to illness or accidents can have a significant impact. In the rehabilitation and treatment of finger injuries, it is vital to classify finger movements and assess their condition accurately. One approach to establishing a connection between the surface electromyogram (EMG) signal and finger movement classes is through the application of deep learning techniques. Deep learning has made remarkable advancements across various domains in recent years, and leveraging its knowledge can be beneficial in this context. In this study, the focus is on classifying and identifying eight different finger movements using deep convolutional neural networks. The researchers utilized EMG signals obtained from the Ninapro database for their analysis. The results indicate that the classification accuracy for certain movements reaches as high as 98.9%. Certain regulators and optimizers have a significant impact on the classification accuracy. By carefully selecting regulators, such as the random removal layer and L2, it is possible to improve the accuracy of the classification.