基于线性混合光谱模型的遥感岩性信息提取研究

发布时间：2018-06-24 09:24

本文选题：高光谱遥感 + 线性混合光谱分解　；参考：《成都理工大学》2017年硕士论文

【摘要】：混合像元分解一直是遥感学科研究热点问题之一,线性混合光谱模型是解决混合像元分解的有效方法,根据分解算法限制条件不同可以分为无约束条件的混合光谱分解、部分约束条件的混合光谱分解以及完全约束条件的混合光谱分解。混合像元分解在土地利用分类、农作物面积提取和植被覆盖度提取研究较多,但在岩性提取方面研究较少,大部分岩性提取都是传统基于逐个像元进行,因此本文利用Hyperion高光谱数据,以西藏境内达孜县与墨竹工卡县之间的交界处作为研究区域,开展基于线性混合光谱分解的岩性提取。取得的主要研究成果如下:(1)对EO-1卫星Hyperion高光谱数据开展数据预处理。通过波段去除、条纹及坏线修复、大气校正、影像裁剪等数据预处理过程获得研究区真实反射率图像。(2)岩石波谱数据获取与分析。通过设计野外岩石采集路线,利用ASD光谱仪在室内获得岩石波谱数据,开展波谱数据处理,建立研究区岩性波谱库。在此基础上进行岩性波谱分析,查明了同一岩性的风化面反射率值与新鲜面反射率值大小差异但两者波谱特征曲线却基本一致的规律。(3)通过分析多种端元提取算法,优选适于岩性分类的端元提取算法。PPI、SMACC、N-FINDR三种端元提取算法应用广泛且易于实现,利用三种算法对研究区进行端元提取,并采用传统SAM方法开展岩性分类,分析比较三种方法提取的端元光谱及岩性分类效果。研究表明:N-FINDR端元提取法要优于PPI法和SMACC法,PPI法受人为因素的限制提取的端元信息不够完整,SMACC法提取的端元精度偏低较为不稳定。(4)分析线性混合光谱模型下的三种丰度反演算法,选出最优算法进行岩性提取。通过探讨三种丰度反演算法:无约束条件的最小二乘法(UCLS)、部分约束条件的最小二乘法(SCLS)、完全约束条件的最小二乘法(FCLS),利用三种算法对研究区进行丰度反演,得到三种反演结果的RMSE值分别为0.0435、0.0378、0.0223,并通过分析评价得到线性混合光谱分解算法的岩性提取精度,总体精度为87.71%,Kappa为79.82%。研究表明:完全约束的最小二乘法的精度最高,且与传统SAM岩性提取相比,基于完全约束条件的线性混合光谱分解能进一步提高岩性提取精度。
[Abstract]:Mixed pixel decomposition has always been one of the hot issues in remote sensing. Linear mixed spectral model is an effective method to solve mixed pixel decomposition. The mixed spectral decomposition of partial constraint condition and the mixed spectral decomposition of complete constraint condition. Mixed pixel decomposition is widely studied in land use classification, crop area extraction and vegetation coverage extraction, but less in lithology extraction. Therefore, using Hyperion hyperspectral data and the boundary between Dazi County and work card County in Tibet as the study area, the lithology extraction based on linear mixed spectral decomposition is carried out. The main research results are as follows: (1) preprocessing the Hyperion hyperspectral data of EO-1 satellite. The true reflectivity images of the study area are obtained by the data preprocessing processes such as band removal, fringe and bad line restoration, atmospheric correction and image clipping. (2) Rock spectral data acquisition and analysis. The rock spectrum data were obtained by ASD spectrometer and the lithology spectrum database was established by designing the rock collection route in the field and using the ASD spectrometer to obtain the rock spectrum data in the laboratory. On the basis of the analysis of lithology spectrum, the difference of reflectivity between weathered surface and fresh surface of the same lithology is found out, but the characteristic curve of wave spectrum is basically the same. (3) by analyzing various endmember extraction algorithms, The three End-element extraction algorithms, PPISMACCN-FINDR, which are suitable for lithology classification, are widely used and easy to realize. The three algorithms are used to extract endcomponents from the study area, and the traditional SAM method is used to carry out lithology classification. End-component spectra and lithologic classification results obtained by the three methods were analyzed and compared. The results show that the ratio N-FINDR endcomponent extraction method is better than PPI method and SMACC method. The end component information extracted by SMACC method is not complete enough. (4) three abundance inversion algorithms based on linear mixed spectral model are analyzed. An optimal algorithm is selected for lithology extraction. Through discussing three abundance inversion algorithms: unconstrained least squares (UCLS), partial constrained least squares (SCLS), complete constrained least squares (FCLS), the abundance inversion of the study area is carried out by using three algorithms. The RMSE values of the three inversion results are 0.0435 / 0.0378/ 0.0223.The lithology extraction accuracy of the linear mixed spectral decomposition algorithm is obtained by analysis and evaluation. The overall accuracy is 87.71 and Kappa is 79.82. The results show that the precision of the fully constrained least square method is the highest, and compared with the traditional SAM lithology extraction, the linear mixed spectral decomposition based on the complete constraint condition can further improve the accuracy of lithology extraction.
【学位授予单位】：成都理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P627

【参考文献】