基于高分二号遥感影像的树种分类方法

为推广国产高分数据在森林树种分类方面的应用，以北京市延庆区八达岭国家森林公园主要区域的6期高分二号影像为数据源，在分层分类的基础上，利用支持向量机递归特征消除、C5.0决策树、FSO 3种特征优选方法，从4种特征维度下实现面向对象的支持向量机和随机森林的森林树种分类，最终取得总体精度平均为83.65%，特定树种生产者精度介于93.75%（山杏）和38.10%（刺槐）之间，特定树种用户精度介于100%（华北落叶松）和44.74%（榆树）之间的良好结果。结果表明：C5.0决策树耗时最短（0.01 h）且其所选特征应用于分类总体精度最高（86.90%）；在不同特征维度下支持向量机分类的总体精度比随机森林平均高出3.28%；支持向量机和随机森林均对特征维度不敏感，但良好的特征优选结果仍会对支持向量机的分类效率（最高提升86.98%）和随机森林的分类精度（最高提升9.22%）产生较大影响。

Classification Method of Tree Species based on GF-2 Remote Sensing Images

In order to promote the application of Chinese Gaofen data in the classification of forest tree species, The six GF-2 images of the main area of Badaling National Forest Park in Yanqing District, Beijing were used as the data source, we used support vector machine-recursive feature elimination, C5.0 decision tree and feature space optimization three feature optimization methods to accomplish the object-oriented Support Vector Machines (SVM) and Random Forest (RF) forest tree classification from four feature dimensions on the basis of the hierarchical classification. we can achieve good classification results that the average Overall Accuracy of the study was 83.65%, the Producer's Accuracy of specific tree species was between 93.75% (Apricot) and 38.10% (Locust), and the Use's Accuracy of specific tree species was between 100% (North China Larch) and 44.74% (Elm). The results showed the C5.0 feature selection took the shortest time(0.01 h) and features selected by it could be applied to the highest classification accuracy (86.90%). Under different feature dimensions, the Overall Accuracy of SVM classification was 3.28% higher than the RF.SVM and RF were both insensitive to feature dimensions, but good feature optimization results will still have a large impact on the classification efficiency of SVM(Highest improvement was 86.98%) and the classification accuracy of RF(Highest improvement was 9.22%).