著者

小林 正幸 小西 康夫 石垣 博行

出版者

一般社団法人日本機械学会

雑誌

日本機械学會論文集. C編 (ISSN:03875024)

巻号頁・発行日

vol.71, no.710, pp.2989-2995, 2005-10-25

---

Support Vector Machines (SVM) are learning machines with high performance about pattern recognition. But, SVM has some parameters decided preliminarily. As these parameters vary, SVM's recognition performance severely varies. In this paper, we propose an optimizing method of SVM's parameters using Reinforcement Learning. By the proposed method, the Actor-Critic method that is a kind of reinforcement learning was used. Actor-Critic method can use continuous action space. The effectiveness of the proposed method is confirmed by a one-dimension pattern recognition simulation. And, we verify practical effectiveness of the proposed method by experiments about handwritten character recognition. As results, the recognition rate of characters increased to about 90% with parameters of SVM optimized by the proposed method.

著者

荒木 望 帆足 勇希 小西 康夫 満渕 邦彦 石垣 博行

出版者

The Institute of Electrical Engineers of Japan

雑誌

電気学会論文誌. C, 電子・情報・システム部門誌 = The transactions of the Institute of Electrical Engineers of Japan. C, A publication of Electronics, Information and System Society (ISSN:03854221)

巻号頁・発行日

vol.131, no.4, pp.736-741, 2011-04-01

被引用文献数

1

---

This paper proposed an active finger recognition method using Bayesian filter in order to control a myoelectric hand. We have previously proposed a finger joint angle estimation method based on measured surface electromyography (EMG) signals and a linear model. However, when we estimate 2 or more finger angles by this estimation method, the estimation angle of the inactive finger is not accurate. This is caused by interference of surface EMG signal. To solve this interference problem, we proposed active finger recognition method from the amplitude spectrum of surface EMG signal using Bayesian filter. To confirm the effectiveness of this recognition method, we developed a myoelectric hand simulator that implements proposed recognition algorithm and carried out real-time recognition experiment.