结合时间序列Sentinel-1数据和面向对象的湿地信息提取方法

基于光学影像的遥感技术受云雨等天气条件影响较大,而合成孔径雷达(SAR)由于具有穿透能力可以很好的克服这一缺陷。本文以黑龙江流域扎龙湿地为研究区域,采用时间序列C波段双极化(VV、VH)Sentinel-1数据,结合面向对象的图像分析技术对扎龙湿地进行分类。对比分析了5种机器学习算法得出随机森林算法的精度最高,总体精度为88.43%,Kappa系数为0.8646,其中沼泽的制图精度达到84.68%,用户精度达到89.47%。使用Sentinel-1数据对扎龙湿地进行湿地信息提取的最佳时相为5月、7月和8月。

Time Series Sentinel-1 Data and Object-oriented Wetland Information Extraction Method

Remote sensing technology based on optical image is greatly affected by weather conditions such as cloud rain,and synthetic aperture radar(SAR)can overcome this defect well because of its penetrating ability.In this paper,the Zhalong wetland in Heilongjiang River Basin was used as the research area.The time series C-band dual-polarization(VV,VH)Sentinel-1 data was used,and the object-oriented image analysis technology was combined to classify Zhalong wetland.The five kinds of machine learning algorithms are compared and analyzed.The random forest algorithm has the highest accuracy,the overall accuracy is 88.43%,and the Kappa coefficient is 0.8646.The mapping accuracy of the swamp reaches 84.68%,and the user precision reaches 89.47%.The best phase for wetland information extraction from Zhalong Wetland using Sentinel-1 data is May,July and August.