在线半监督Kohonen网络的预抓取手势识别

Prefetching gesture recognition based on online semi supervised Kohonen network

为实现智能仿生手的抓取,提高模式识别的实时性和灵敏性,提出一种在线半监督Kohonen网络。该网络针对表面肌电信号(s EMG)的特性,在有监督Kohonen网络基础上,将有监督和无监督网络的优势进行结合,应用数据剪辑方法处理训练集更新识别网络,在线识别侧边抓取、球形抓取、三指精确抓取和圆柱形抓取4种预抓取手势。实验表明,与不同Kohonen网络相比,此识别方法具有很好的在线识别能力和正确率。

In order to realize the grasp of intelligent bionic hand, and improve the timeliness and sensitivity of pattern recognition, a semi supervised Kohonen network is presented in this paper. According to the characteristics of surface eleetromyography （sEMG）, the network combines the advantages of supervised and unsupervised network, and applies the data editing method to handle the training set, then to update the identification network based on the supervised Kohonen network. The presented network is used to identify the four prefetehing patterns online： lateral, spherical, fingertip and cylindrical. Experiments show that, compared with the different Kohonen network, the recognition method has a good ability to identify online and correct rate.