Forearm Surface EMG Palm Movement Classification

In this project we were handed a system consisted of a mechanical hand to which were connected 3 servo motors capable of moving the hand’s fingers. The motors get instructions on how to operate from an Edison board manufactured by Intel, which we can run the algorithms when needed. The purpose of the system is being a prosthetic for below the elbow amputees. Also, there’s an arm bracelet consisted of 8 electrodes that receive EMG (electromyography) signal from the muscles, transmit them to the Edison board which tells the motors how to work.
The system was constructed by a pair of students that in addition also embedded an algorithm capable of classifying 6 different hand gestures with relatively high accuracy. The problems that arose were long calibration time, strong dependency on how the gestures are performed during calibration and how the hand operates during test time, sensitivity of the signal to the armband’s orientation on the hand and the wearer (the user) as well as the need to characterize the gestures efficiently for classification. These are the problems addressed in this project. These difficulties can be solved using a part of the machine learning world called domain adaptation (i.e. DA). At this point we reviewed the literature on the characteristics of the EMG signal and ways to works with it and analyze it as well as reviewing possible DA algorithms. We found out that wavelet coefficients are widely used as EMG characterizing tools, and also found several DA algorithms that can help us such as FEDA,HAD,TCA,MMDT where MMDT was chosen as a winner.

Forearm Surface EMG Palm Movement Classification
Forearm Surface EMG Palm Movement Classification
Forearm Surface EMG Palm Movement Classification