摘要
植被光谱变异性广泛存在于遥感图像当中,本文尝试通过PROSAIL辐射传输模型来描述植被端元变异性,并提出一种光谱解混方法,实现逐像元地估计植被变异性端元。具体地,面向植被—土壤背景两端元的场景,在非负矩阵分解框架下,利用PROSAIL辐射传输模型从机理上描述植被端元的变异性,并通过两组神经网络来分别实现辐射传输模型的反演与正算,从而更高效地拟合植被端元,最终得到一种能逐像元求解变异性植被端元的光谱解混算法。由于该方法求解了植被端元的空间变异光谱,因此,能够对植被参数遥感反演的尺度效应进行纠正。为此,本文进一步以LAI尺度效应为例,通过无人机图像实验来验证该方法的有效性。实验结果得出,经过光谱解混方法处理后,该方法能较准确地估计植被端元,并能使LAI尺度效应均方根误差RMSE能够从0.2151降低到0.0896,有望提升遥感植被信息提取的精度。
As ground features are affected by various factors, the problem of endmember variability will occur. Endmember variability greatly affects the accuracy of spectral unmixing results. This study is based on the vegetation-soil binary scene, and spectral unmixing is performed under the framework of NMF(Nonnegative Matrix Factorization). The PROSAIL model is used to describe the variability of vegetation endmembers from the mechanism so that the results of spectral unmixing have clear physical meaning. To improve efficiency, we set up two neural networks for model calculation and model inversion. In this way, the spectral unmixing algorithm can obtain the endmembers of the vegetation pixel by pixel, which can more accurately describe the variability of the vegetation endmember.In addition, there is diversity in the spatial resolution of remote sensing data products. The problem of the scale effect is widespread and is a key issue in the field of remote sensing. Analysis of the reason is largely due to the mixed pixels that universally exist. The method studied in this paper describes the variability in the vegetation endmember. A spectral unmixing algorithm that can describe the variability of vegetation endmembers pixel by pixel is obtained. This result can be used to invert vegetation parameters. Therefore, this study attempts to correct the scale effect of remote sensing products by considering the spectral unmixing method of vegetation endmember variability.This paper takes the LAI scale effect as an example. The effectiveness of the method is verified by an Unmanned Aerial Vehicle(UAV)image experiment. Three subimages were selected, and then they were resampled to two different levels of spatial resolution for experimentation. Among them, exponential function fitting is performed through the simulated spectrum of the PROSAIL model, and the relationship model is constructed to invert the LAI. The experimental results show that(1) the spectral unmixing method that uses the PROSAIL model to describe the variability of vegetation endmembers can obtain higher unmixing accuracy;(2) after using this spectral unmixing method, the Root Mean Square Error(RMSE) of the LAI scale effect is significantly reduced, and it has a certain effect on the correction of the LAI scale effect. This can improve the remote sensing scale effect problem to a certain extent.In summary, the spectral unmixing method has a certain effect on the correction of LAI scale differences and can improve the problem of the remote sensing scale effect to a certain extent. However, the research has the following issues that need further consideration:(1) This article only considers the vegetation-soil binary scene, but this method has not verified the multiple endmember scene.(2) The variability of soil and other background endmembers can be further considered.(3) The model can be further optimized to reduce the difference between the PROSAIL model spectrum and the real image spectrum, thereby improving the accuracy of unmixing.(4) In this paper, the evaluation is carried out by the method of upscaling. In the future, more complicated practical factors can be considered for evaluation. Simultaneously,real images collected at different flight altitudes can also be used for evaluation.
作者
韦钦桦
罗文斐
唐凯丰
WEI Qinhua;LUO Wenfei;TANG Kaifeng(School of Geography,South China Normal University,Guangzhou 510631,China)
出处
《遥感学报》
EI
CSCD
北大核心
2023年第2期456-470,共15页
NATIONAL REMOTE SENSING BULLETIN
基金
高分辨率对地观测系统重大专项(编号:11-Y20A40-9002-15/17)
广东省自然资源厅科技项目(编号:GDZRZYKJ-ZC2020003,GDZRZYKJ2020004)。
