安徽省利辛县平原区人工林树种分类研究

快速准确地绘制平原区人工林树种分布对研究人工林的生态水文和社会经济效益具有重要的意义。将资源3号（ZY-3）全色波段分别同ZY-3多光谱、哨兵2号多光谱进行融合，在图像分割基础上提取变量，采用分层优化变量组合的随机森林分类方法对安徽省利辛县人工林树种进行分类，并与分类回归树和随机森林相比较。结果表明：利用分层分类方法，平原区的人工林树种分类精度可以达到92%以上；基于哨兵2号和ZY-3融合的光谱特征变量分类精度比 ZY-3 数据本身的融合提高了2.49% ~ 2.91%；而分别加入纹理变量后，分层分类方法大幅度提高了树种分类精度。因此，基于高分辨率遥感数据的光谱和纹理特征，采用分层分类方法，可以有效提高平原区人工林树种的分类精度。

Man-made Tree Species Classification in Lixin County,Anhui Province

Forest plantations in plain area of China are highly fragmented and dispersed due to flexible planting system. Accurate and timely mapping plantation distribution in plain area plays an important role in plantation management and hydro-ecological functions. In this study， a hierarchical-based classification method， which optimizes variable combinations at each node， was proposed and applied in Linxi County， Anhui Province for mapping tree species in plain area. The object–based classification with combinations of various types of spectral and texture features derived from the fused image of ZY-3 multispectral and panchromatic band （ZY）， and the fused image of Sentinel-2 multispectral bands and ZY-3 panchromatic band （STZY） were conducted. The results showed that the proposed method provides higher classification accuracy than regression tree and random forest approaches for both datasets， especially for dominant poplar tree species. When only spectral features were used， STZY offered better results than ZY， overall accuracies increased by 2.49%~2.91%， indicating the importance of spectral information in classification. When textures were integrated with spectral features into classification procedure， overall accuracies increased by 10.19% and 4.99% for STZY and ZY respectively using the hierarchical-based classifier， implying that texture features are essential for classification. Thus， the proposed hierarchical-based classifier with optimized variable combinations is an effective method in tree species mapping in plain area using spectral and texture features derived from high spatial-resolution data.