多源遥感数据图谱协同岩石单元分类方法

岩石单元的结构、构造、差异风化和出露状况在遥感图像上综合表现为图形纹理特征即"图"标志,其矿物成分和组合则表现为光谱特征即"谱"标志.传统遥感岩石单元分类以利用其光谱特征为主,图形纹理特征为辅,因此分类精度有限.以新疆维吾尔自治区与甘肃省交界的北山西段为研究区,开展岩石单元图形指数和光谱指数协同分类方法研究.基于Worldview-2全色图像构建的图形指数,能够量化岩石单元的层理、构造、展布形态和微地貌等特征,包括0°和45°定向滤波图像及灰度共生矩阵计算出的同质性和异质性特征图像、熵特征图像;光谱指数基于Worldview-2多光谱图像和ASTER(Advanced Spaceborne Thermal Emission and Reflection Radiometer)短波红外波段图像利用比值、和-差方法构建.多源遥感图像构建的光谱指数其光谱波段涵盖可见光-近红外及短波红外,包括RI(Ratio index)ASTER、SI(Spectral index)ASTER、SIWorldview-2.采用面向对象方法对建立的图谱指数进行多尺度分割,依据不同岩石单元出露规模建立适宜的分割尺度,利用光谱指数自动提取相应岩石信息,实现岩石单元自动分类.结果表明,实验区基于图谱协同方法共划分出17类岩石单元,总体精度达到83.62%,而单独利用Worldview-2和ASTER图像,仅划分出13类和14类岩石单元.提出的图谱协同岩石分类方法可为我国西部高海拔深切割无人区地质调查及找矿工作提供新思路和遥感技术支撑.

Geological Units Classification with Texture-Spectral Synergy of Multi-Sourced Remote Sensing Images

The structural pattern, differential weathering, and outcrop situation of different geological units can be described by the texture information of remote sensing imagery as graphical features. The geological units compositions of differential minerals are shown as spectral cues.With regard to geological units classification, the classification accuracy of most studies is limited since they have mainly utilized the spectral cues from multis-pectral or hyperspectral images to characterize their spectral features, and high resolution imagery to depict the texture information as a supplement. In this study, the texture-spectral indices are established with the multi-sourced remote sensing images of Worldview-2 and ASTER(Advanced Spaceborne Thermal Emission and Reflection Radiometer)and tested for geological units classification in the western section of Beishan Mountain, which are located at the junction of Gansu Province and Xinjiang Uygur Autonomous Region. Based on the panchromatic band of Worldview-2 imagery, the following graphical features that quantify the bedding structure, distribution morphology and micro-topography features of different rock units were extracted: 0° and 45°directional filtering, the textures of homogeneity, dissimilarity and entropy generated from gray level co-occurrence matrix. Based on multi-spectral bands of Worldview-2 imagery and the short-wave infrared bands of ASTER imagery, the spectral indices were established by the band-ratio and addition-difference methods, including RI (Ratio index)_ASTER, SI(Spectral index) _ASTER, and SI_Worldview-2. First, based on the object-oriented approach and the texture-spectral indices, texture-spectral features were used to conduct multi-resolution segmentation to produce numerous geological units with different scales. Second, the geological units were classified at different scales using spectral indices. The results show that: (1) 17 types of geological units were classified with an overall accuracy of 83.62%, based on the proposed method; (2) and only 13 or 14 types of geological units were classified, by using the Worldview-2 or ASTER images, respectively. The outlined approach could provide the theoretical and remote sensing technical support for geological survey and prospecting work in the high altitude with deep-cut unpopulated areas of western China.