二向性反射分布函数的先验知识耦合式融合方法研究

The key to solving the ill-posed inversion problem of remote sensing is to supplement enough a priori knowledge. The fusion of a priori knowledge is an effective way to raise knowledge content and reduce the inversion ill-posedness. Traditional decision fusion method considers each function model as the black box and ignores the coupling relationships between models. Traditional decision fusion method cannot realize knowledge fusion. To solve these problems, this project focuses on studying the coupled fusion method of a priori knowledge of BRDF. First, the universal representation framework of a priori knowledge of BRDF is built; second, the coupling relationships of a priori knowledge are analyzed and represented; third, the fusion method of a priori knowledge of BRDF is studied based on the sufficient consideration of coupling relationships; finally, the method that evaluates the fusion effectiveness and the strategy that revises the model are proposed. There are two innovations: first, the analysis and representation method of the coupling relationships of a priori knowledge of BRDF is proposed; second, the fusion method of a prior knowledge of BRDF is proposed based on the analyses of coupling relationships. The study of this project has important theoretical significance and application value in terms of revealing the coupling relationships of a priori knowledge of BRDF, realizing the effective fusion of a priori knowledge of BRDF, and reducing the adverse effect of ill-posed inversion.

解决遥感病态反演问题的关键在于补充足够的先验知识,先验知识的融合是增加反演知识含量、降低反演病态性的有力手段。传统决策融合方法将各功能模型视为黑箱子,对模型间耦合关系重视不足,无法实现知识融合。针对上述问题,本项目着重研究BRDF先验知识耦合式融合方法这一科学问题。首先,构建BRDF先验知识的通用表示框架；其次,分析与表示多种先验知识的耦合关联；然后, 在充分考虑BRDF先验知识耦合关联的基础上,研究BRDF先验知识融合方法；最后,制定先验知识融合有效性的检验方法及模型调整策略。创新点为：(1)提出BRDF先验知识的耦合关系分析与表示方法；(2)提出基于耦合分析的BRDF先验知识融合方法。本项目在充分揭示BRDF先验知识耦合关系,实现BRDF先验知识高效融合,降低病态反演的不利影响等方面具有重要的理论意义和应用价值。

解决遥感病态反演问题的关键在于补充足够的先验知识。多源先验知识的融合是增加反演知识含量的有力手段，但传统决策融合方法对模型间耦合关系重视不足,难以实现反演先验知识的高效融合。本项目从BRDF反演先验知识的通用表示框架、耦合分析、融合方式、检验调整等方面入手，对BRDF先验知识耦合式融合方法进行研究，取得了一系列创新性研究成果：(1)提出基于多核学习的先验知识通用表示框架,该框架一方面能够通过多源特征抽取多层面反演知识; 另一方面, 能够将多类型BRDF反演模型知识参数化表示,根据参数类型构建反演知识库。(2)提出面向多种BRDF先验知识耦合关系的双层hashing编码分析框架,该框架既能充分考虑知识参数的耦合关系又能实现反演知识的局部匹配,为多类型反演知识的有效融合奠定基础。(3)提出基于Boosting的多源遥感反演知识局部耦合式融合方法,使得多源遥感反演知识通过交替互补的形式逐步实现融合。本项目在充分揭示BRDF先验知识耦合关系,实现BRDF先验知识多层面高效融合,降低病态反演的不利影响等方面具有重要的理论意义和应用价值。