数学与系统科学研究院

计算数学所学术报告

报告人：        Yanfei Wang

Institute of Romote Sensing Applications, CAS

报告题目 Land Surface Parameter Retrieval via Kernel-driven BRDF Model & Other Related Inversion Problems in Remote Sensing

Abstract: The essence of the quantitative remote sensing is the inversion. There are many different kinds of inverse problems in remote sensing. In this talk, we focus on the land surface parameter retrieval by solving the kernel-based BRDF model. Generally speaking, the description of a solvable physical process should be overdetermined as described in Proposition 3 of [Verstraete et al., 1996]. However such a requirement is hardly satisfied even in the coming EOS era. Therefore, in order to solve the BRDF inversion problem, Li et al (1998, 2001) utilizes a priori knowledge to convert the problem into a over-determined system to find its least squares error solution. So, from the computational view, both of them actually solve an overdetermined system. In this talk, we investigate the robust estimation of the land surface albedos by direct solution of the kernel-driven BRDF model. Our method is based on the deeply investigation of the spectrum of the linear driven kernel, then we develop a direct method, i.e., a numerically truncated singular value decomposition method, which can alleviate the difficulties in numerical computation when the discrete kernel is badly conditioned. This method can always find a set of suitable BRDF coefficients even for poor sampled data. Numerical performance is given for the widely used 18 data sets among the 73 data sets [Li et al., 2001].

We also present some other inversion problems in remote sensing, which is still under study.

报告时间：2004年10月26日  下午3：00-4：00

报告地点：科技综合楼三层报告厅