基于盲源提取的高光谱矿物识别方法与机制研究

The research of this project focuses on the key scientific problems which restrict the accuracy of current hyperspectral mineral identification. The problems include the uncertain diagnosis absorption characteristics of mineral spectrum, mixed pixels in hyperspectral imagery and the highly correlated minerals. The blind signal processing algorithm which has been successfully applied in the information extraction is adopted in this project. We propose a new framework of the hyperspectral mineral identification based on the blind source extraction algorithm (BSE). Firstly, the information model of the mineral spectrum diagnosis absorption characteristics is going to be established, so that highly related mineral spectrum signal could be distinguished by the model. The information model will have a better robustness for the mineral spectral drift and spectral mutation phenomenon. Secondly, by using the information model of a mineral target, the blind source extraction algorithm will be drived to extracting an estimated signal under the condition of assuming it is contained in an mixed pixel. Finally, a quantitative evaluation of the correlation beteen estimated signal and the mineral spectrum in spectral library will be defined. According to the correlation value, we can determine whether the estimated signal in mixed pixel is the mineral. Based on the researches above all, theory and methods of the BSE-based hyperspectral mineral identification will be proposed systematically to improve the identification accuracy.

利用高光谱遥感技术进行矿物识别对我国矿产资源勘查有着极为重要的现实意义。本项目围绕当前制约高光谱矿物识别精度的关键科学问题展开研究，这些问题包括矿物波谱诊断吸收特性的不确定性、高光谱混合像元以及不同矿物波谱高度相关问题。 项目采用近年来成功用于信息提取的盲信号处理算法，提出了基于盲源提取的高光谱矿物识别新框架。首先，构建矿物波谱诊断吸收特性的信息模型，使得通过该模型能够区分高度相关的矿物波谱信号，并且对矿物光谱漂移、光谱变异等现象有较好的鲁棒性。其次，利用该信息模型驱动盲源提取算法，在假设混合像元中含有目标矿物的情况下提取出估计信号。最后，建立该估计信号与波谱库矿物波谱相关度的定量评价，通过相关度值来判断该估计信号是否为疑似目标矿物。从而建立系统的基于盲源提取的矿物识别理论与方法，进一步提高基于高光谱遥感技术的矿物识别的精度。

利用高光谱遥感技术进行矿物识别对我国矿产资源勘查有着极为重要的现实意义。本项目围绕当前制约高光谱矿物识别精度的关键科学问题展开研究，这些问题包括矿物波谱诊断吸收特性的不确定性、高光谱混合像元以及不同矿物波谱高度相关问题。项目采用近年来成功用于信息提取的盲信号处理算法，提出了基于盲源提取的高光谱矿物识别新框架。首先，构建矿物波谱诊断吸收特性的信息模型，使得通过该模型能够区分高度相关的矿物波谱信号，并且对矿物光谱漂移、光谱变异等现象有较好的鲁棒性。其次，利用该信息模型驱动盲源提取算法，在假设混合像元中含有目标矿物的情况下提取出估计信号。最后，建立该估计信号与波谱库矿物波谱相关度的定量评价，通过相关度值来判断该估计信号是否为疑似目标矿物。从而建立系统的基于盲源提取的矿物识别理论与方法，进一步提高基于高光谱遥感技术的矿物识别的精度。项目研究结果表明，与传统的限制能量最小化、自适应余弦估计、匹配滤波、广义最大似然估计、自适应匹配滤波、正交子空间投影等方法比较，本项目的方法在亚像元检测的检测率和虚警率的综合效率上有一定的提升。