Wavelet transformation and classification with machine learning methods of electromyography signals for bionic hand control

Multi functional prothesis bionic arms/hand are limited in term of motion. To improve those prothesis functionality reserarchers are studying in many ways. Those are enhancig the ability of motion, providing the desired rotation and the motion by improving electronical parts and analyzing EMG signals that are source signals. It is crucial for efficiency of classification and system's performance that analyzing of signals should represent the aimed motion. Developments in signal analysis procedures of EMG and machine learning methods are the most important factors that effects system's performance. Selecting high frequency bands or low frequency bands in order to clear disturbance may deflect characteristics. So that if is important to analyze signals correctly smart algorithms should be used. EEG signals can be used as alternative to source signals. But EEG signals have lower bandwith and lower amplitude. So, they can be defected harder when compared with EMG. It is known that Wavelet Transformation Analysis is using widely to estimate characteristics of EMG signals. In this study, it has been focused on decomposition degree and prediction of wavelet that is the most representative property of EMG signals. In addition, the question of which method should be used for creating more efficient classification. 900 data set that belongs to EMG signals of 6 basic hand motions which are usually used for holding/grabbing has been used in order to improve the functionality of prothesis hands. Split Wavelet Transform and Markov Transformation Martix have been used to onlain vector of attribute that is the most representative speciality of EMG signals. Performance analyses have been performed to vector of attribute by using various algorithms of machine learning methods. When comparing with other worldwide studies, it is seen that the results of this study, are accurate, more over, it is more preferable in terms of real-time usage, beacuse of the stability between speed and performance. In addition to help developing prothesis-bionic hand/arms, it is expected that prediction of the characteristics of an efficient EMG signal will be able to help designing the prothesis more accurate, stiching body parts and identicating clinical diagnosis. By methods that have been used in this study, improvement of prediction of characteristics of EMG signals and performance is expected. All obtained results and suggested methods may contribute to new studies.