摘要
高光谱遥感能同时获取地物空间影像和连续且精细的光谱信息,以图谱合一的形式更为精确地描述地物,然而高光谱影像普遍存在同物异谱和同谱异物现象,给准确地物分类带来挑战;激光雷达(light detection and ranging,LiDAR)能够获取地物拓扑信息,可用于构建地表三维模型,但单纯采用LiDAR数据无法准确识别地物。基于以上2点,开展高光谱影像和LiDAR数据的融合研究,采用形态学属性剖面进行特征提取,借助稀疏多项式逻辑回归分类器分类,探讨在不同特征组合下的融合与分类效果;并以黑河中游张掖绿洲农业区精细作物分类为目标,采用航空高光谱影像和LiDAR DSM数据对本文方法进行了应用验证。研究表明,该方法可获得精度更高和稳定性更好的分类结果,高光谱与LiDAR融合的方法分类精度最高可达94. 50%。
Hyperspectral remote sensing can simultaneously acquire spatial images of space and fine spectral information so as to describe the features more accurately.However,when the phenomena of different spectra in the same objects or the same spectra in different objects occur,the classification of hyperspectral images will face a daunting challenge.Light detection and ranging(LiDAR)can obtain the terrain topology information and can be used to construct the surface3D model.However,features cannot be accurately identified by using LiDAR data only.Based on the above two points,the authors carried out a study to fuse hyperspectral images and LiDAR data.Morphological attribute profile was used to extract features,and sparse multinomial logistic regression(SMLR)was used to do classification.The fusion and classification effect in different combinations of characteristics were also investigated.The CASI/SASI aerial hyperspectral image and LiDAR DSM data were used to validate this method based on the Zhangye Oasis agricultural area in the middle reaches of the Heihe River which is a good target for the classification of crop.The results show that the method using hyperspectral and LiDAR data can obtain better classification results with higher accuracy and stability,and the best classification accuracy is94.50%by fusion features based on the extended morphological attribute profile.
作者
杨思睿
薛朝辉
张玲
苏红军
周绍光
YANG Sirui;XUE Zhaohui;ZHANG Ling;SU Hongjun;ZHOU Shaoguang(School of Earth Sciences and Engineering, Hohai University, Nanjing 211100, China;School of Naval Architecture and Ocean Engineering, Jiangsu Maritime Vocational Institute, Nanjing 211170, China)
出处
《国土资源遥感》
CSCD
北大核心
2018年第4期33-40,共8页
Remote Sensing for Land & Resources
基金
国家自然科学基金项目"高光谱遥感影像稀疏深度学习与分类研究"(编号:41601347)
江苏省自然科学基金项目"稀疏图嵌入的高光谱遥感影像小样本分类研究"(编号:BK20160860)
中央高校基本科研业务费"高光谱遥感影像半监督深度学习与分类研究"(编号:2018B17814)
江苏省光谱成像与智能感知重点实验室2018年开放基金项目(编号:3091801410406)
测绘遥感信息工程国家重点实验室2018年开放基金项目(编号:17R04)共同资助