结合Sentinel影像与特征优选的山地城市不透水面提取

Extraction of Impervious Surface in Mountainous City Combined withSentinel Images and Feature Optimization

不透水面是影响山区生态环境的重要因素。多源遥感数据融合是不透水面提取的重要方法,但容易造成分类特征冗余,需要进行特征优选。现有研究尚缺乏对不同特征优选方法在不透水面遥感提取中的对比分析。首先使用Sentinel-1和Sentinel-2影像等数据提取特征,然后分别基于JM距离、随机森林(RF)模型和ReliefF算法进行特征优选,并基于RF分类模型评价不同特征子集的分类精度,最后提取山地城市昆明市2020年的不透水面。结果表明,纹理特征相比光谱和地形特征重要性更高,基于ReliefF算法的特征子集分类效率和精度最高,模型训练时间相比JM距离与RF分别减少了84s和16s,总体精度为0.95,Kappa系数达到0.79,为基于多源数据的地物遥感提取提供了参考。

Impervious surface is an important factor affecting the ecological environment of mountainous areas.Multi-source remote sensing data fusion is an important method for impervious surface extraction,but it is easy to cause classification features redundancy and feature opti⁃mization is needed.Existing researches lack comparative analysis of different feature optimization methods in remote sensing extraction of im⁃pervious surface.Firstly,we used Sentinel-1 and Sentinel-2 images extracted features.Then,features were optimized based on Jeffries-Ma⁃tusita distance,Random Forest(RF)model and ReliefF algorithm respectively,and the classification accuracy of different feature subsets were evaluated based on RF classification model.Finally,the impervious surface of mountainous city Kunming in 2020 was extracted.The re⁃sults show that texture features were more important than spectral and topographic features.The feature subset based on ReliefF algorithm had the highest classification efficiency and accuracy,and the training time of the model was reduced by 84 seconds and 16 seconds respectively compared with JM distance and RF,with an overall classification accuracy of 0.95 and a Kappa coefficient of 0.79.This article provides a refer⁃ence for remote sensing extraction of ground objects based on multi-source data.