Classification of finger movements using deep learning

This study focuses on the classification of finger movements using EMG signals and deep learning model. Electromyography signals are biomedical signals that record the electrical activity of muscles, and these signals are analyzed and used in various fields such as the production of prosthetic limbs, diagnosing conditions, and even controlling prosthetic limbs. The data used in this study is the data collected from ten healthy volunteers. Each volunteer performed seven tasks related to finger movements, and each task was completed five times; this means that the total number of experiments from all volunteers is 350. The data was recorded using electromyography signals, and then these data were processed and made usable. In this study, the data was made ready for classification after being processed. First, the data belonging to each finger was separated, and since more than one channel belonged to some fingers, the FastICA method was used to reduce the data of these channels to one. Then, specific features were extracted using time domain descriptors (TDD), convolutional neural network (CNN), and long short-term memory (LSTM). After extracting these features, the most important features for the classification process were selected using the mutual information method (MI). After this process, the classification was performed by selecting the fully connected neural network model (FCNN). The classification process was successful due to the effective learning of the model and the high accuracy rates. After these results were obtained, it can be concluded that deep learning models can be an effective tool in the classification of vital signs and EMG signals. It can be concluded from this study that deep learning can provide effective contributions to the fields of classification and analysis.