文献类型：期刊文献

中文题名：基于多类特征融合的极限学习在四足机器人野外地形识别中的应用

英文题名：Application of extreme learning based on multi-class feature fusion in field terrain recognition of quadruped robots

作者：刘彩霞[1];方建军[1];刘艳霞[1];马慧姝[1]

第一作者：刘彩霞

机构：[1]北京联合大学自动化学院

第一机构：北京联合大学城市轨道交通与物流学院

年份：2018

卷号：0

期号：2

起止页码：97-105

中文期刊名：电子测量与仪器学报

外文期刊名：Journal of Electronic Measurement and Instrumentation

收录：CSTPCD;;Scopus;北大核心:【北大核心2017】；CSCD:【CSCD_E2017_2018】；

基金：北京市属高等学校高层次人才引进与培养计划(CIT&TCD20150314);北京市自然科学基金(4142018)资助项目

语种：中文

中文关键词：四足机器人;纹理特征;小波特征;极限学习机;地形识别

外文关键词：quadruped robot;texture feature;wavelet feature;extreme learning machine;terrain recognition

摘要：针对四足机器人在野外环境下对多地形的识别能力较弱的问题,提出了一种基于多类特征融合的极限学习识别算法。该算法首先针对野外不同地形表面性质和组织结构的特点,利用纹理特性和小波变换获其低维和高维特征,作为分类器的训练特征。然后引入极限学习机分类算法对多种地形进行识别。结果显示,算法识别率为97.5%,比传统的BP神经网络算法、支持向量机算法分别高出30.89%和20.45%。并且重复的实验证明了该算法具有很好的泛化能力和鲁棒性,这为四足机器人关于提高其自主移动能力的研究提供了一种新的思路。
Aiming at the problem that quadruped robot has a weak ability to identify terrain in off-road environments,the extreme learning recognition algorithm based on multi-class feature fusion is proposed in the paper. Firstly,for the characteristics of the different terrain surface nature and structure in the field,the low dimensional and high dimensional feature is respectively achieved by the texture characteristic and the wavelet transform,and they are used as the training feature of classifier. Then the extreme learning machine is introduced to identify multiple terrains. The results show that the recognition rate of the algorithm proposed in this paper is 97. 5%,which is 30. 89% and 20. 45% higher than the traditional BP neural network algorithm and the support vector machine algorithm,respectively. And the repeated experiments show that the method has a good generalization ability and robustness,which provides a new idea for quadruped robot to improve its autonomous mobility.