水稻分蘖期无人机高光谱影像混合像元特征分析与分解  

Characteristic Analysis and Decomposition of Mixed Pixels From UAV Hyperspectral Images in Rice Tillering Stage

作　　者：于丰华[1,2] 赵丹 郭忠辉 金忠煜 郭爽 陈春玲 许童羽[1,2] YU Feng-hua;ZHAO Dan;GUO Zhong-hui;JIN Zhong-yu;GUO Shuang;CHEN Chun-ling;XU Tong-yu(College of Information and Electrical Engineering,Shenyang Agricultural University,Shenyang 110866,China;Liaoning Agricultural Information Engineering Technology Research Center,Shenyang 110866,China)

机构地区：[1]沈阳农业大学信息与电气工程学院,辽宁沈阳110866 [2]辽宁省农业信息化工程技术研究中心,辽宁沈阳110866

出　　处：《光谱学与光谱分析》2022年第3期947-953,共7页Spectroscopy and Spectral Analysis

基　　金：辽宁省博士启动基金计划项目(2020-BS-131);国家重点研发计划项目(2016YFD0200600)资助。

摘　　要：开展水稻无人机高光谱解混,获取水稻植株的高光谱反射率信息,对于提高水稻理化参量的反演模型精度具有重要意义。目前大多基于高光谱遥感影像自身数据进行解混,运用算法模型进行高光谱数据解混,将高光谱图像和可见光图像进行优势互补,提出一种基于无人机高清影像与高光谱遥感影像融合的稻田无人机高光谱解混方法,解决单一数据局限性问题,增强光谱数据对地物的描述能力。为了更好的计算端元丰度,将同一目标区的高清数码正射影像与无人机高光谱遥感影像利用经纬度信息进行空间配准,使得不同传感器获得的图片在几何位置上对齐,通过SVM分类器的监督分类方法对可见光的数码正射影像进行地物分类,利用地物分类的结果对应高光谱的一个像元,从而得到一个像元内的端元丰度。设相邻区域内的水体端元是相同的,利用线性解混模型(LSMM)对相邻区域的混合像元进行解混,最终获取水稻高光谱反射率信息。结果表明对两种图片进行空间配准丰富了数据源信息,有利于像元的端元丰度计算,其中水稻端元丰度在70%以上解混效果最好,丰度在50%以上解混效果一般,丰度在30%以下解混效果较差;选择监督分类方法进行地物分类,精度达到99.5%,面向对象方法分类精度为98.2%,监督分类方法优于面向对象分类方法;最终得到的混合像元分解反射率高于原混合像元反射率,减少了水体混合部分对光谱数据的影响,使得分解后水稻的光谱反射率更加准确,为水稻理化参量无人机成像高光谱遥感反演提供更加准确的科学依据。Conducting the unmanned aerial vehicle(UAV)hyperspectral unmixing of rice and obtaining the hyperspectral reflectance information of rice plants is of great significance for improving the accuracy of the inversion model of rice physical and chemical parameters.Most of the current research is based on the data of hyperspectral remote sensing images themselves for demixing.That is,unmixing of hyperspectral data is carried out by using algorithm model.In this study,the advantages of hyperspectral images and visible spectral images were complemented,and a hyperspectral unmixing method for UAV in rice field was based on the fusion of UAV high-definition images and hyperspectral images remote sensing images was proposed.This method solved the problem of the limitation of single data and enhanced the description ability of spectral data for ground objects.In order to better calculate the endmember abundance,the high-definition digital orthophotos of the target area were spatially aligned with the UAV hyperspectral remote sensing images,so that the pictures obtained by different sensors were aligned in geometric positions.The supervised classification method of the SVM classifier was used to classify the digital orthophoto of visible light,and the result of the classification was used to correspond to a pixel of the hyperspectrum to obtain the endmember abundance within a pixel.Suppose the endmembers of the water body in adjacent areas were the same,the linear unmixing model(LSMM)was used to unmix the mixed pixels in the adjacent area and finally the hyperspectral reflectance information of rice was obtained.The results showed that the spatial registration of the two images enriches the data source information,which was beneficial to the endmember abundance calculation of the pixels.Among them,the unmixing effect of rice endmember abundance above 70%was the best,the unmixing effect of abundance above 50%was general,and the unmixing effect was poor when the abundance was below 30%.Use the supervised classification method

关 键 词：高光谱 无人机 端元提取 水稻 混合像元 

分 类 号：S252.9[农业科学—农业机械化工程]