中国科学院遥感与数字地球研究所-高光谱遥感应用技术研究室推荐内容
【2009博】基于光谱诊断和目标探测的高光谱岩矿信息提取方法研究
|
【作者】 李庆亭 【导师】 童庆禧；张兵；蔺启忠 【 学位年度 】 2009 【论文级别】 博士 【关键词】 高光谱遥感，光谱混合模型，光谱诊断，目标探测，蚀变信息提取 【Key words】 hyperspectral remote sensing, spectral mixing model, spectral diagnosis, target detecting, alteration information extraction 【中文摘要】 高光谱遥感最先是由地质学家们在研究矿物和岩石的光谱特性时提出来的，地质应用可谓是高光谱遥感应用最早也是最成功的领域，但其并非已经很成熟和完善了，仍然在很多方面不能够满足地质应用的实际需求，如基于光谱诊断性特征的岩矿精细识别、地化信息的定量提取和地质小目标的探测和提取等，仍要求人们深化对光谱作用的机理及规律的认识，发展针对具体应用需求的技术和方法，进而提高地质遥感的技术能力。 本文总结了高光谱遥感地质研究现状，重点总结了光谱产生的机理和高光谱遥感岩矿信息提取的研究进展。在此基础上研究了岩石和矿物的光谱混合模型，进而分析了蚀变矿物组合的光谱特征，研究信息识别和提取的方法，建立了基于蚀变矿物组合混合光谱的岩矿信息提取方法，并做具体的应用示例。最后针对蚀变岩识别和提取进行目标探测算法研究，并根据地质应用的特点建立合适的目标探测策略，应用于高植被覆盖区和裸露区的蚀变信息提取。论文的主要研究内容和成果如下： （1）在对线性混合模型和非线性混合模型研究的基础上，提出了基于Hapke和Shkuratov混合理论的光谱模拟模型，其模拟结果明显优于线性混合模型的模拟结果，并得到实验验证。针对不透明矿物提出单次散射反照率的近似求取方法，使Hapke模型适用于含有不透明矿物的矿物混合物（岩石或土壤）的光谱模拟。所提出的混合光谱模拟模型将有助于建立岩石和矿物的二向反射率光谱库，并可应用于月球或其他行星的光谱模拟。 （2）利用高空间分辨率的全色相机和多光谱成像光谱系统，结合ASD光谱仪，设计了混合光谱测量实验，为光谱混合模型研究提供了思路，此实验思路可以进一步应用于航空平台的实验。结合线性混合模型和非线性混合模型，针对遥感数据混合像元特点，提出低空间分辨率的遥感图像混合像元光谱模拟模型，可以应用于图像混合光谱模拟，如接触蚀变带的图像光谱模拟。 （3）在对典型围岩蚀变的矿物组成及其光谱特征分析的基础上，针对岩矿信息特别是蚀变信息的提取，提出了面向目标的提取思路，即目标光谱模拟----光谱特征分析和光谱匹配----权重函数判决的三步提取策略，并成功地应用于Cuprite矿区蚀变带的提取，实验证明目标的光谱模拟和权重函数综合判决可以大大提高识别的精度和可信度。利用地面实测光谱，基于蚀变带组合的光谱进行了包古图地区光谱地质剖面图的制作，研究了如何从光谱剖面升级到光谱地质剖面的方法，结果证明在光谱分辨率和信噪比满足的情况下，可以实现蚀变组合的精细识别和填图。针对ASTER遥感数据光谱和空间分辨率都比较低的特点，提出光谱匹配和彩色合成相结合的岩矿信息提取方案，克服了光谱比值和光谱变换的缺点，岩体和蚀变信息得到了增强。 （4）针对地质目标提取的特点，集中研究了目标已知而背景未知的目标探测算法，包括正交子空间投影算法（OSP）、约束能量最小化算法（CEM）、ACE和AMF探测算法和椭圆形轮廓分布探测算法（ECD），深入分析了其进行目标探测的原理和算法效能的影响因素，并提出改进策略，即引入综合权重函数加权自相关矩阵（协方差矩阵）的思想，使目标探测算法不仅能够识别小目标，也能够识别大目标。通过不同类型研究区的地质小目标的探测实验，优选出了适用于不同背景下地质目标提取的目标探测算法，结果表明对于地质小目标的提取，应首选ACE、AMF算法，其次为CEM。
【Abstract】 Hyperspectral remote sensing was created by the geologist when they studied the spectral features of rocks and minerals and its application in geology is earliest and most successful. Even though, the technique still is not mature and perfect because its short in satisfying some need in modern geology application. The accurate recognition of rocks an minerals based on the diagnosis spectral features, the quantificationally extraction of geochemistry information and the extraction of small geological objects all need to deep the understanding of the mechanisms and principles of spectroscopy of rocks an minerals. The technology and method for the unique application should be developed for the improving of the geological remote sensing technology. The art of geological study by hyperspectral remote sensing was introduced in the paper, especially the spectroscopy of rocks and minerals and its information extraction methods. Then the spectral mixing models were studied, the spectral features of altered mineral groups were analyzed and the information extraction methods were investigated. The rock and mineral information extraction method was set up for alteration recognition and several tests were practiced. At last, the target detection algorithms are studied for the recognition of alteration, the proper geological object detecting strategy was set up and was practiced for the alteration information extraction in vegetation covered area and no-vegetation area. The main content and result is introduced as follows: (1) The spectral simulating model which integrating the Hapke model and Shkuratov mixing model was provided based on the study in the linear and nonlinear spectral mixing models. The spectral simulating result using the model provided in the paper was better than the traditional linear model and was tested by the experimentation. The approximation for single scattering albedo of opacity minerals in Hapke model was provided to get the spectral simulating method adapted to the spectral simulating of mineral mixtures which containing opacity minerals. The model is useful for the establishing of the bidirectional reflectance library of rocks and minerals and can be applied in the spectral simulating of the moon and other planets. (2) The mixed spectral experiment was practiced which integrating the high spatial resolution camera, multispectral imaging system and ASD FieldSpec FR2500. This idea gived the supporting for the study of spectral mixing model and can be applied in the aero platform in large scale in further study. A spectral simulating model for remote sensing with low resolution was provided, it integrated the linear and nonlinear model and can used for the spectral simulating of mix-pixel such as the image spectra of alteration zone. (3) Based on the analysis of mineral component and spectral features in typical alteration wall rock, object-oriented information extraction method was provided for the extraction of rocks, minerals and alteration, which contained three steps: the object spectra simulating, spectral feature analysis and spectral matching, weight function decision. This method was used for alteration zone in Cuprite area successfully. The method was also used for the execution of the geological spectral histogram in Baogutu area taking the spectral of alteration group as reference. The accurate recognition and mapping can be realized if the spectral resolution and signal to noise is enough high. The information extraction method which integrating of spectral matching and color component can greatly enhance the rock body and alteration in the image,despite the spectral and spatial resolution of ASTER data is low. (4) Several target detecting algorithms with objects known and background unknown considering the characteristics of geological objects were studied, such as Orthogonal Subspace Projection(OSP), Constrained Energy Minimization(CEM), Adaptive Coherence/ Cosine Estimator(ACE), Adaptive Matched Filter(AMF),Elliptically Contoured Distributions(ECD). The principle of the detectors and the factors effecting detecting efficiency were analyzed and the improved strategy was provided. The integrated weight function correlation matrix (covariance matrix) was considered and the improved algorithms can detect the big object bedides the small object. Several target detecting algorithms were selected by the detecting test of small geological object with different background in vegetation covered area and no-vegetation area, the efficiency of ACE and AMF is best and CEM is better.
(责任编辑：admin) |
