纹理感知的碎屑岩图谱协同遥感分类方法

Due to the complicated composition of clastic rock, it has been a difficulty of the lithology accurate classification using remote sensing image. In the remote sensing image, the bedding structure of the clastic rock is shown as the stable textural features, which is “graphic” symbol. The components of the clastic rock is shown as the “spectrum” symbol. Based on the Geo-informatic Tupu idea, the subject will establish clastic rock remote sensing classification method from two perspectives which are texture and spectra features. The main contents are as follows: ①calculate the litho-stratigraphic spatial texture information by using the high-resolution images and DEM data, then, we use association rules mining the frequent items between rock unit and the spatial structure textural feature, establish the graphic symbol of the clastic rock recognition, build the graphic-rock unit recognition model;②take advantage of spectral matching and correlational analysis method to associate the measured rock spectrum and the hyper spectral image spectrum, establish spectral symbol of the typical rock recognition, bulid the spectral-rock mapping model;③divide high resolution image by using graphic features, merge the two models of graphic-rock unit and spectral-rock, then determine the pattern and method by feature matching, complete the lithology confirmation by spectral and map symbol, finaly realize automatic rock classification. Taking clastic rock for experimental verification in the south tianshan mountains area, where exposed rich and complete clastic rock without vegetation. This research explores the new ideas for remote sensing lithology classification, at the same time, it provides theoretical and technical support work to remote sensing geology survey and prospecting work in the western difficult areas.

碎屑岩因物质成分复杂，其精准分类一直是遥感岩石分类的难点之一。遥感图像上碎屑岩的层理构造表现为图形纹理特征即“图”标志，物质成分则表现为光谱特征即“谱”标志。课题以地学信息图谱思想为基石，基于纹理和光谱两个特征研究图谱协同岩石分类方法。主要内容有：①利用高分遥感数据和DEM数据量化岩石单元的图像几何空间纹理信息，基于关联规则挖掘岩石单元与纹理特征间的频繁项集，建立典型岩石识别的图形标志，构建图形-岩石识别模型；②利用光谱匹配、相关分析关联实测岩石光谱和高光谱图像光谱，建立典型岩石识别的光谱标志，构建岩石-光谱映射模型；③依据图形特征分割高分图像，融合图形-岩石和岩石-光谱模型，基于特征匹配确定图谱协同模式和方法，完成岩性确认，实现碎屑岩自动分类，以无植被覆盖碎屑岩出露丰富的南天山地区进行实验验证。本研究为遥感岩石分类开辟新思路，同时为西部艰险复杂区遥感地质调查及找矿工作提供理论和技术支撑。

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。在目前的技术条件下，一种遥感数据源很难同时具有高空间分辨率和高光谱分辨率特征。中等空间分辨率遥感数据由于覆盖了碳酸盐矿物、粘土矿物、铁氧化物矿物等矿物的诊断光谱区间，广泛应用于矿物、岩石识别，但受限于空间分辨率，混合像元现象明显，严重制约了中分数据的岩性分类精度。高空间分辨率遥感数据提供了岩石地层表面丰富的空间结构信息，但高分数据覆盖的光谱区间往往较窄，难以满足大多数特征吸收谱段位于短波红外、热红外区间的矿物、岩石信息提取。因此，如何协同不同来源的遥感数据进行矿物岩石识别及岩石地层划分成为遥感地质的重要研究内容。项目分别以新疆维吾尔自治区与甘肃省交界的北山西段、新疆维吾尔自治区乌恰县康苏镇及其附近区域、阿克陶县克孜勒乡塔木地区为研究区，以Worldview-2、Hyperion、ASTER、Landsat-8 OLI为多源遥感数据源，开展岩石单元图形指数和光谱指数协同分类方法研究。利用灰度共生矩阵量化Worldview-2图像的空间信息，利用异源数据融合协同Worldview-2、Hyperion、ASTER、Landsat-8 OLI图像不同波谱区间光谱信息优势，基于面向对象分类、模糊分类、决策树分类方法分别划分三个研究区岩石地层。试验结果表明，项目提出的图谱协同方法岩石分类精度在三个研究区均达到80%以上。研究结果表明，项目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。