基于光学与雷达遥感影像协作的多云雾区域农作物信息提取研究  

Extraction of Crop Information in Cloudy Areas based on Optical and Radar Remote Sensing Images

作　　者：周兴霞 王颖洁 杨攀 ZHOU Xingxia;WANG Yingjie;YANG Pan(The Third Institute of Photogrammetry and Remote Sensing,Ministry of Natural Resources,Chengdu 610100,China)

机构地区：[1]自然资源部第三航测遥感院,四川成都610100

出　　处：《遥感技术与应用》2024年第2期362-372,共11页Remote Sensing Technology and Application

基　　金：四川省重点研发项目(2022YFS0449);四川省自然资源科研项目(Kj-2022-(3))联合资助。

摘　　要：及时掌握农作物分布信息,对农业结构调整和国家粮食安全保障具有重要的战略意义。然而,多云雾区域受气候条件制约,难以获取全时序的光学影像,无法构建涵盖农作物完整生长季的物候特征数据集,农作物信息提取受到限制。因此,本研究以西南多云雾区域重要的产粮地之一——四川省广汉市为例,基于SAR影像(Sentinel-1)和光学影像(Sentinel-2)数据协作,构建了研究区冬季农作物典型时相光谱特征和夏季农作物SAR时序特征,采用面向对象决策树分类方法,提取了研究区农作物时空分布信息并进行精度分析。结果表明:①广汉市农作物以粮油作物为主,主要包括小麦-水稻、油菜-水稻、土豆-大豆、土豆-玉米4种一年两熟的轮作种植模式。②水稻、大豆、玉米3种夏季农作物SAR时序特征差异明显,光学-雷达遥感数据协作提取冬季-夏季农作物类型及空间分布为多云雾区域农作物信息提取提供了新的思路。③面向对象的决策树分类方法分类总体精度和Kappa系数分别为85.49%和0.81,分类结果保持了大面积作物的完整性,避免了分类结果的“椒盐现象”。Rapid and accurate extraction of crop type,spatial and temporal distribution is of great significance for agricultural structure adjustment and national food security.However,there are few optical remote sensing im‐age of cloudy areas,thus crop monitoring is limited.To make up this shortage,spectral signature of winter crops and SAR time series characteristics of summer crops were proposed based on the Sentinel-2 and Sentinel1 data for high-accuracy crop mapping.The Guanghan County,an important grain-producing region in south‐west China,was studied.The object-oriented decision tree classification method was explored for spatial and temporal distribution extraction of crops in study area,and the classification accuracy was verified.The results shows that:(1)the main crops in Guanghan County are grain and oil crops,and the major crop rotation pat‐terns are wheat-rice,rape-rice,potato-soybean and potato-corn;(2)the SAR time series characteristics of rice,soybean,corn show clear differences,extracting the types and distribution of winter-summer crops based on the optical-SAR remote sensing images provides a new idea for crops monitoring by remote sensing images in cloudy areas.(3)The overall accuracy and Kappa coefficient of object-oriented method reach 85.49%and 0.81,which can maintain the integrity of large area crops,and avoid salt and pepper noise.

关 键 词：光学遥感 雷达遥感 多云雾区域 面向对象 农作物信息 

分 类 号：S127[农业科学—农业基础科学] TN958[电子电信—信号与信息处理]