特征优选的GF-1影像植被遥感精细分类研究  

Study on fine classification of vegetation using GF-1 image based on feature optimization

作　　者：葛俊勇 葛俊涛[2] 刘朝霞 GE Junyong;GE Juntao;LIU Zhaoxia(Urumqi Jingweitiandi Mining Engineering Technology Co.,Ltd.,Beitun Branch,Beitun,Xinjiang 836000,China;Surveying and Mapping Geographic Information Institute of Henan Geological and Mineral Exploration and Development Bureau,Zhengzhou,Henan 450006,China)

机构地区：[1]乌鲁木齐经纬天地矿业工程技术有限公司北屯市分公司,新疆北屯836000 [2]河南省地质矿产勘查开发局测绘地理信息院,河南郑州450006

出　　处：《测绘标准化》2024年第1期42-47,共6页Standardization of Surveying and Mapping

摘　　要：植被精细化分类对掌握不同植被在维护生态环境安全中的作用具有重要意义。本文选择国产高分一号(GF-1)卫星影像,构建该影像下的基于光谱、纹理、植被指数的多特征空间,并选择ReliefF算法与CFS算法进行特征筛选,最后结合随机森林算法与libsvm模型,研究特征选择对分类精度的影响,并获取植被精细化分类的最佳多特征分类算法模型。结果表明,特征选择能在一定程度上提高分类精度;CFS算法相较于ReliefF算法能更好地简化特征子集的维度,并获取能提高分类精度的更优子集;基于CFS-libsvm的植被精细化分类方法具有较高的分类精度。Fine classification of vegetation is of great significance for understanding the role of different vegetation in maintaining ecological environment security.This paper selects the domestic Gaofen 1(GF-1)satellite image,constructs a multi feature space based on spectrum,texture and vegetation index under the image,selects ReliefF and CFS algorithms for feature screening,and finally combines the random forest algorithm and libsvm model to study the impact of feature selection on classification accuracy,and obtain the best multi feature classification algorithm model for fine classification of vegetation.The results show that feature selection can improve the classification accuracy to a certain extent.Compared with ReliefF algorithm,CFS algorithm can better simplify the dimension of feature subset,and obtain better preferred subset to improve classification accuracy.The refined classification method of vegetation based on CFS-libsvm has higher classification accuracy.

关 键 词：植被精细化分类 RELIEFF算法 CFS算法 特征优选 

分 类 号：P237[天文地球—摄影测量与遥感]