Classification of electromyography signals from hand grasping of different object shapes and positions

In order to use the prosthesis hand, the dexterous and robust algorithm is proposed. However, the effects of object shapes and object positions on the accuracy from hand grasping classification system have to be considered. We present the hand grasping classification system based on electromyography signals (EMG) from three different object shapes with five different object placement positions using a linear discriminant analysis (LDA) classifier. The three different object shapes include parallelepiped, cylinder, and cone. There are two protocols for training the classification system, namely, the intraposition protocol and the interposition protocol. In the intraposition protocol, the trained dataset and the tested dataset are from same object placement positions. On the other hand, in the interposition protocol, the trained dataset and the tested dataset are from different object placement positions. Results show that the error rates from the intraposition protocol (0-4.94%) are much lower than those from the interposition protocol $(\gt20$%). Moreover, when the trained dataset is from all object placement positions, the obtained error rate is 1.98-15.31%. These results suggest the future work on using the trained dataset and the tested dataset that are from same object placement positions to achieve better error rates.