摘要
针对最近正则化子空间(NRS)算法精度受限的问题,该文设计了基于随机子空间的最近正则化子空间集成算法(RSNRS)。该算法通过采用随机子空间的集成方式有效地提升了基分类器NRS的分类效果。考虑到充分利用高光谱遥感影像丰富的空间信息,进一步提出了一种基于随机子空间和马尔科夫随机场的高光谱遥感最近正则化子空间集成学习算法(RSNRS-MRF)。该算法利用马尔科夫随机场对RSNRS的初始分类结果进行结构规则化,得到最终分类结果。实验结果表明,RSNRS和RSNRS-MRF在分类效果上比NRS有明显的提升,RSNRS-MRF相较于RSNRS具有更好的分类效果,更能有效削弱噪声。
In view of the precision limit of nearest regularized subspace(NRS),a random subspace-based nearest regularized subspace(RSNRS)algorithm was proposed in this paper.Considering the spatial information of the hyperspectral image data,a random subspace-based nearest regularized subspace with Markov random field(RSNRS-MRF)model was further designed.The MRF was used to regularize the structure of initial classification results to obtain the final classification results.Experimental results showed that the classification effect of RSNRS and RSNRS-MRF were significantly improved than that of NRS.Compared with RSNRS,RSNRS-MRF had a better classification effect and could reduce noise effectively.
作者
虞瑶
杨锦
邱洁
YU Yao;YANG Jin;QIU Jie(Basic Geographic Information Center of Jiangsu Province,Nanjing 210013,China)
出处
《测绘科学》
CSCD
北大核心
2022年第10期161-167,175,共8页
Science of Surveying and Mapping
基金
江苏省自然资源科技项目(2021052)
关键词
随机子空间
协同表示
马尔科夫随机场
集成学习
random subspace
collaborative representation
Markov random field
ensemble learning
作者简介
虞瑶(1995—),女,硕士,主要研究方向为GIS、摄影测量与遥感等。E-mail:yuyaoyao_yy@163.om
