Transradiyal kol hareketleri için iki kanal EMG sınıflandırması

In this thesis, the classification algorithm for electromyography (EMG) based prosthesis which can be developed for individuals who have undergone transradial arm amputation has been studied using artificial neural networks (ANN). Surface electromyography (sEMG) has been preferred for ease of application in order to receive EMG signals. The data were processed and various features were extracted. Mean absolute value (MAV), root mean square (RMS), simple square integral (SSI), variance of EMG (VAR), log detector (LOG), maximum fractal length (MFL), wavelength (WL), average amplitude change (AAC), difference absolute standard deviation value (DASDV), Willison amplitude (WAMP) and slope sign change (SSC) features are given as an input matrix to the ANN to be trained for classification purposes. In this study, four different movements as relaxed hand, hand close, wrist flexion, and forearm supination are classified. The accuracy, sensitivity, specificity, and precision performance metrics were calculated as a result of ANN training and examined and interpreted. As a result of the study carried out within the scope of this thesis, the effects of the number of hidden layer neurons, feature extraction, feature selection and individualization of ANN to classification performance were evaluated. A classifier that provides high classification performance metrics were established by training with feature sets formed by grouping and combining features with individual low accuracy rates. The two feature sets created achieved 88 percent accuracy with generalization and 95 percent accuracy with individualization.