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

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.