基于Sentinel-2A影像的森林类型提取研究

Forest Type Extraction Based on Sentinel-2A Images

基于Sentinel-2A遥感影像的光谱特征、指数特征和纹理特征,结合野外调查数据、森林二类调查数据和DEM数据等辅助数据,采用分层多尺度分割提取技术,对研究区主要森林类型进行提取。研究表明:主成分变换前三个分量的加入有助于提高影像分割效果和分类精度;在影像光谱特征基础上,加入指数特征和纹理特征可在一定程度上提高森林类型提取精度;在多元数据的支持下,选取合适的特征和阈值进行分层与分类是一种有效的森林类型分类策略,取得了较高的分类精度,其中杉木、马尾松、竹林和其它林地的总体分类精度达79.80%,Kappa系数为0.725;通过决策树和随机森林两种分类器的优势互补,以分类精度混淆矩阵中的生产精度和用户精度平均值作为先验知识进行多分类器决策级投票融合,相比于单一决策树和随机森林分类器,组合分类器具有更高的分类精度,总体分类精度分别提高了3.32%和2.13%。

Based on the spectral features,index features and texture features of Sentinel-2 A remote sensing images,combined with field survey data,secondary forest survey data,DEM data and other auxiliary data,the main forest types in the study area were extracted by the hierarchical multi-scale segmentation and extraction technology.The study showed that the addition of the first three components of the principal component transform was helpful to improve the segmentation effect and classification accuracy.On the basis of image spectral features,adding index features and texture features can improve the accuracy of forest type extraction to a certain extent.With the support of multivariate data,it is an effective forest classification strategy to select appropriate features and thresholds for stratification and classification,which achieves high classification accuracy.The overall classification accuracy of Chinese fir,Masson pine,bamboo forest and other forest lands was 79.80%,and the Kappa coefficient was 0.725.Through complementary advantages of two classifiers decision tree and random forest,the production accuracy and user average precision of classification accuracy confusion matrix as prior knowledge to carry on the multi-classifier vote decision level fusion,and compared with random forest and single decision tree classifier,combination classification has higher classification accuracy,the overall classification precision improved by3.32%and 2.13%respectively.