Evaluation of popular features and entropy as a new feature for hand gesture classification by electromyography signals

The topic of EMG signals classification by pattern recognition methods has more than 40 years of history. Today, efforts are being made to make this classification more applicable, using the developed methods to make more accurate predictions in a shorter time. While scanning the literature, it is found that a group of features became popular among researchers and repeatedly used without giving any other reason than popularity. It is also found that entropy has many usages within bio-signal classification research. Still, it is unused for classifying EMG signals of hand gestures. This research investigates the classification ability of nine "popular" features of EMG research and proposes entropy as a new feature. For this purpose, experiment protocols are formed using SVM classifiers with the NinaPro DB5 dataset. The first nine popular features are tested with different combinations. Later, the entropy feature is analyzed with window-length sensitivity, its problem with LPH is solved with the proposed GPH approach. Then activation detection is developed and implemented for a real-time test. Results of both parts are given in their respective chapters in a way it is easy to compare with other works. The evaluation of popular features showed that some features could harm the system; however, this highly changes from application to application. On the other hand, the entropy feature showed that it could be used to classify EMG signals, and it can classify feature vectors extracted from windows with lengths different than the training set.