摘要
近年来随着小卫星数量与传感器类型的快速增加,急需研究和发展快速可靠的小卫星遥感影像分类方法。针对分类方法各具局限性、具体应用中最优分类器选取困难等问题,本文基于多分类器集成学习的思路,引入随机森林(Random Forests)方法用于小卫星遥感影像分类。采用灾害监测预报小卫星(HJ-1)、北京1号小卫星(BJ-1)两种国产小卫星多光谱遥感影像进行试验,并与传统分类方法进行比较,结果表明,随机森林比最大似然分类器(MLC)、支持向量机分类器(SVM)等具有更好的稳定性、更高的分类精度和更快的运算速度,具有很好的适用性。
Small satellite remote sensing,characterized by wide coverage,all weather observation ability,and flexible operation mode,plays a significant role in resource and environment monitoring,emergency response,and after-disaster relief and rebuilding.As the foundation of remote sensing image processing and application,small satellite image classification attracts more and more attention of researchers,with the increasing number of small satellite sensors and the widening applications in recent years.Because of the limitation of traditional classifiers and the difficulty of selecting a strongest classifier in practical use,image classification using single classifier is always not satisfactory.Aiming to overcome this problem,Random Forests,an advanced classifiers ensemble method,was employed to small satellite remote sensing image classification in the paper.Classification results of HJ-1 and BJ-1 images by random forests demonstrated that random forests could outperform conventional MLC and SVM in terms of stability,computation speed and classification accuracy.
出处
《测绘科学》
CSCD
北大核心
2012年第4期194-196,共3页
Science of Surveying and Mapping
基金
国家自然科学基金资助项目(40871195)
关键词
随机森林
多分类器
集成学习
决策树
国产小卫星
random forests
multiple classifiers
ensemble learning
decision tree
China small satellite