基于Sentinel-2A的太行山区土地覆被分类方法研究

以太行山区为研究对象,基于Sentinel-2A遥感影像数据,采用基于像元和面向对象分类两种策略,定量分析不同特征组合模式下,最大似然法(ML)、贝叶斯(Bayes)、支持向量机(SVM)、决策树(Decision Tree)以及随机森林(RF) 5种分类方法在该区域地表土地覆被信息分类中的表现差异。结果表明:(1)基于像元的RF分类器取得了最高精度,仅使用光谱特征参与分类和使用光谱、红边、指数特征参与分类的总体精度分别为96. 85%和96. 64%。(2)红边和指数特征的加入能够对各分类器分类精度产生不同程度的影响,即使基于像元的RF和面向对象的CART决策树总体精度有所下降,但降幅均在0. 5%左右,其他分类器精度均有所提升。

Land Use/Cover Classification in Taihang Mountain Area Based on Sentinel-2A Imagery

Based on Sentinel-2A remote sensing image data, this paper took Taihang Mountain Area as the research object to quantitatively analyze the different performance of five classification methods of maximum likelihood (ML), Bayes, support vector machine (SVM), decision tree, and random forest (RF) in the region under different feature combination modes, which adopting two strategies of pixel based and object-oriented classification. The results show that (1) the RF classifier based on pixel achieves the highest accuracy, while the overall accuracy of only using spectral features and using spectral, red edge and exponential features is 96.85% and 96.64%, respectively.(2) The addition of red edges and exponential features can have different degrees of impact on the classification accuracy of each classifier. Even if the overall accuracy of the pixel-based RF and object-oriented CART decision trees decreases, but the decline is about 0.5%. The accuracy of other classifiers has been improved.