面向对象的旱区植被遥感精细分类研究

Fine vegetation classification of remote sensing in arid areas based on object-oriented method

针对旱区植被分类尺度过大、种群无法准确提取的问题,该文提出了面向对象的CFS-RF分类模型,即利用CFS算法对先验样本数据集进行特征优选,结合随机森林构建分类规则,完成分类过程。以新疆阿勒泰为研究区,利用GF-2数据,通过CFS、ReliefF两种不同特征选择方法和J48、SVM、RF 3种分类算法构造出6种面向对象分类方案来实现小尺度植被种群提取。结果表明,经过特征选择,上述分类方案的精度和效率均得到了提升。其中,CFS-RF算法最优,总体精度达到92.41%,Kappa系数为0.90,更适用于旱区植被遥感精细分类。

Aiming at the problem that the classification scale of vegetation in arid areas was too large and the species could not be accurately extracted,this paper proposed an object-oriented algorithm named CFS-RF,CFS algorithm was used to optimize the features of prior sample data sets,Combining with random forest,the classification rules were constructed and the classification process was completed.In this paper,taking Altay,Xinjiang province as the research area,using GF-2data,six object-oriented classification schemes were constructed by combining two different feature selection methods,CFS and ReliefF,with three classification algorithms,J48,SVM and RF,to achieve the extraction of small-scale vegetation species.The result showed that through feature selection,the accuracy and efficiency of the above classification scheme had been improved,and the CFS-RF algorithm was the best,the overall accuracy reached 92.41%,and the Kappa coefficient was 0.90,it was more suitable for fine vegetation classification using remote sensing in arid areas.