SEMG Multi-Class Classification Based on S4VM Algorithm
A method using small amount of labeled instants and large unlabeled ones simultaneously involved in the training during the sEMG classification obtained a better effect is strongly needed. This paper introduces the S4VM proposed by Li et al. into surface
- PDF / 1,813,522 Bytes
- 9 Pages / 439.37 x 666.142 pts Page_size
- 11 Downloads / 192 Views
SEMG Multi-Class Classification Based on S4VM Algorithm Zhuojun Xu, Yantao Tian, Zhang Li and Yang Li
Abstract A method using small amount of labeled instants and large unlabeled ones simultaneously involved in the training during the sEMG classification obtained a better effect is strongly needed. This paper introduces the S4VM proposed by Li et al. into surface EMG pattern recognition with small labeled instants and extends to multi-class classification problems, which will represent the autoregressive model characteristic value of the human hand movements of the seven types of EMG signal as the object of classification. The experimental results show that the safety semisupervised support vector machine is suitable for the multi-pattern classification of surface EMG signal with high accuracy and good robustness. Keywords S4VM
EMG Multi-class Classification
62.1 Introduction SEMG signal as one kind of biological signal reflects human muscle movement characteristics which have the strength obtains directly and without invasive pain have been widely used in the fields of diagnostics, exercise physiology, design and development of myoelectric prostheses [1], and others. Through lots of research, people try to find the corresponding relation between the surface EMG signal and movement mode, and get the good effect by the method of neural network [2], support vector machine [3], discrete wavelet transforms [4], and decision trees [5], Z. Xu (&) Y. Tian School of Communication Engineering, Jilin University, Changchun, China e-mail: [email protected] Z. Li Y. Li Key Laboratory of Bionic Engineering, Ministry of Education Jilin University, Jilin, China
Z. Zhong (ed.), Proceedings of the International Conference on Information Engineering and Applications (IEA) 2012, Lecture Notes in Electrical Engineering 219, DOI: 10.1007/978-1-4471-4853-1_62, Ó Springer-Verlag London 2013
495
496
Z. Xu et al.
etc. However, it is a large workload to mark the EMG signals one by one, but it is relatively easier to obtain the unlabeled EMG signal. Therefore, it becomes a new issue of the surface EMG pattern recognition studies that combines a small amount of labeled instants with large amounts of unlabeled instants at the same time involved in the classification of training and get a better classification effect. In recent years, the semi-supervised learning method (semi-supervised learning, SSL) attracted wide attention as a kind of machine learning method among supervised. This paper is supported by the Key Project of Science and Technology Development Plan for Jilin Province (Grant No.20090350), Chinese College Doctor special scientific research fund (Grant No.20100061110029), Doctoral interdisciplinary scientific research projects fund of Jilin University (Grant No.2011J009), and the Jilin University ‘‘985 project’’ Engineering Bionic Sci. & Tech. Innovation Platform. Learning and unsupervised learning, learning instants include labeled data and unlabeled data [6]. This method uses a small amount of labeled data fo
Data Loading...
