基于稀疏表示的热红外高光谱数据岩性分类研究

发布时间：2018-04-14 12:23

  本文选题：热红外高光谱 + 稀疏表示　； 参考：《中国地质大学(北京)》2017年硕士论文

【摘要】：高光谱遥感图像包含丰富的光谱信息,对地物具有更强的分辨能力,但也带来了很多复杂的问题,如维数过多和数据不确定性等。传统的遥感图像处理方法较难满足当前遥感应用的需求,有必要针对具体应用需求探索专门的遥感图像处理方法。论文以“热红外高光谱矿化蚀变矿物提取方法研究与应用示范(地大北京)”项目为依托,开展了热红外高光谱遥感图像的分类及应用研究。针对热红外高光谱数据,引入了模式识别领域当前先进的方法——稀疏表示作为技术手段,综合考虑空间维和光谱维的信息,提出了一种邻域加权的稀疏表示分类方法,并以岩性分类为例,在甘肃柳园研究区进行了应用。主要研究内容与成果如下:1.以TASI数据为例,从热红外辐射传输过程出发,研究了热红外高光谱遥感图像的大气校正和温度与发射率分离预处理方法,分别研究了MODTRAN模型、ASTER-TES方法和ISSTES方法。采用MODTRAN模型和ASTER-TES方法对TASI数据进行了预处理,反演得到了研究区的TASI数据地表发射率产品。2.系统研究了稀疏表示的理论与方法,建立了基于稀疏表示的高光谱数据分类模型。对稀疏表示问题优化模型和稀疏表示分类模型分别进行了展开研究。其一,根据稀疏表示分类方法和热红外高光谱遥感图像的特性,提出了一种邻域加权的稀疏表示分类方法(SRCWN)。其二,引入了K-SVD作为类别字典构建的方法,基于类别字典将未知像元进行稀疏表示。其三,对稀疏表示结果进行各类别重构误差计算,以重构误差最小化规则确定未知像元所属类别。该方法以稀疏表示分类法为基础,充分考虑了热红外高光谱遥感数据的光谱特性、邻近空间信息和数据的稀疏性,可以更有效地对地物像元类别进行区分。在甘肃柳园研究区,采用本文方法开展了岩性分类应用,得到了研究区高光谱遥感岩性分类图。3.结合测试数据对各分类方法进行了对比评价,本文方法在总体精度和Kappa系数上较SAM、SVM和SRC均有一定的提升。结合野外验证资料从总体和局部的角度分别评价了TASI数据的岩性分类应用情况,总体评价表明本文方法的分类结果与实际情况基本符合,局部表现较传统SAM法类别边界更为清晰。
[Abstract]:Hyperspectral remote sensing images contain rich spectral information and have a stronger ability to distinguish ground objects, but also bring many complex problems, such as excessive dimension and data uncertainty.The traditional remote sensing image processing method is difficult to meet the needs of the current remote sensing application. It is necessary to explore a special remote sensing image processing method for the specific application needs.In this paper, the classification and application of thermal infrared hyperspectral remote sensing images are studied based on the project of "extraction method and application demonstration of thermo-infrared hyperspectral mineralized altered minerals (Beijing)".For thermal infrared hyperspectral data, a neighborhood weighted sparse representation classification method is proposed by introducing the current advanced method of pattern recognition-sparse representation as a technical means, considering the spatial dimension and spectral dimension information synthetically.Taking lithologic classification as an example, it is applied in Liuyuan research area of Gansu province.The main research contents and results are as follows: 1.Taking TASI data as an example, the atmospheric correction and preprocessing methods of temperature and emissivity separation for thermal infrared hyperspectral remote sensing images are studied, and the MODTRAN model ASTER-TES method and ISSTES method are studied respectively.The MODTRAN model and ASTER-TES method are used to preprocess the TASI data, and the surface emissivity product of TASI data in the study area is obtained by inversion.The theory and method of sparse representation are studied systematically, and the classification model of hyperspectral data based on sparse representation is established.The optimization model of sparse representation problem and the classification model of sparse representation are studied respectively.Firstly, according to the characteristics of sparse representation classification method and thermal infrared hyperspectral remote sensing image, a neighborhood weighted sparse representation classification method is proposed.Secondly, K-SVD is introduced as a method to construct class dictionaries, which sparse represents unknown pixels based on category dictionaries.Thirdly, the reconstruction error of each class is calculated for the sparse representation result, and the unknown pixel belongs to a class is determined by the minimum rule of reconstruction error.Based on sparse representation classification, the spectral characteristics of thermal infrared hyperspectral remote sensing data are fully taken into account, and the sparsity of adjacent spatial information and data is considered.In the study area of Liuyuan, Gansu Province, the lithologic classification was carried out by using this method, and the hyperspectral remote sensing lithologic classification map .3in the study area was obtained.Based on the test data, the classification methods are compared and evaluated. The overall accuracy and Kappa coefficient of this method are better than that of SAMSVM and SRC.The lithologic classification and application of TASI data are evaluated from both the overall and local aspects combined with field verification data. The overall evaluation shows that the classification results of this method are basically consistent with the actual situation.The local performance is clearer than the traditional SAM method.
【学位授予单位】：中国地质大学(北京)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P627
，

本文编号：1749301