k-NN法を用いた表面筋電位による指動作識別

概要

論文の詳細を見る
Human's motion by the muscle's contraction conforms to the signal(The signal is given from the brain). Electric signal generated by muscle operation is called Electromyogram(EMG). The one observed on the body surface is called Surface EMG(S-EMG). In this paper, we propose a finger motion recognition system from surface-EMG using k-NN method. First, we tested on computer simulations of the method of finger motion recognition from the S-EMG. So far, the authors used only Nearest Neighbor method. However, the Authors did not a experiment on other recognition methods. We compare k-NN method performance with Nearest Neighbor method and Multilayer Perceptron and tested on several computer simulations. The results obtained from the computer simulation show the effectiveness of k-NN method (Recognition rate 90.0%). Next, we propose k-NN method with on-line adjustment technique and test own system by two subjects. The proposed system is online adjustment method. This adjustment method adjusts proposed system side at the same time as training by experimental subject The usefulness of the system is confirmed by comparing temporal change in recognition rate that k-NN method with on-line adjustment technique and original k-NN method. From the experiment results, k-NN method with on-line adjustment technique is more appropriate than original k-NN method.
宮崎大学の論文
2007-08-30