EGI OpenIR
新疆卡拉塔格地区 HySpex 高光谱影像蚀变信息填图
Alternative TitleMapping Alteration Minerals Using HySpex Hyperspectral Data in the Kalatage District, NW China
VATANBEKOV FURKAT
Subtype硕士
Thesis Advisor周可法
2020-06-30
Degree Grantor中国科学院大学
Place of Conferral北京
Degree Discipline工学硕士
Keyword蚀变提取 野外验证 反射光谱 三角翼 低空探测 Alteration extraction field verification endmember reflection spectrum delta wing low-altitude-aircraft
Abstract遥感是通过分析给定设备或传感器收集的数据来获取对象目标,区域或现象信息的艺术或科学,该技术与被调查的对象。 区域或现象没有直接的物理接触,因此遥感技术具有广泛的覆盖范围。 成像高光谱传感器可以通过数十到数百个波段对地面物体进行成像,并在获取地面特征的空间信息时获取地面特征的连续光谱信息, 这大大提高了识别地面物体的能力。本研究使用的 HySpex 成像高光谱数据具有较宽的光谱响应范围 (400-2500 nm) , 较窄的谱带宽度 (2.8nm/5.45 nm) 和较高的空间分辨率可以更有效地应用于矿物蚀变信息的提取。如何通过遥感图像分类技术从遥感图像中提取有效信息是遥感地质科学应用的热点和难点之一。随着高光谱遥感技术的进步,卫星和机载遥感由于分辨率和效率要优于地面点测量, 数据应用在地质学和矿物学领域越来越广泛和适合。本研究区玉带斑岩型铜矿位于卡拉塔格西部, 断层以南, 矿床主要由侵入性,喷出性和沉积岩组成,包括流纹岩斑岩,斑岩性石英闪长岩和辉长岩,它们以储量的形式出现在该地区的西部,并被堤坝侵入。 研究区以斑状石英闪长岩为主体,并与矿化有密切的时空联系。 研究区的行政区划属于新疆哈密市五宝乡,西邻吐鲁番市。地貌多为低丘陵宽谷低丘准平原地区,平均海拔为 500-600 米,最高海拔为 693 米,相对高度差为数十米。本研究的目的是通过利用超低空高精度快速探测技术平台(三角翼搭载 HySpex 高光谱传感器),以玉带斑岩型铜矿区为试验区,开展低空高光谱提取蚀变信息的方法技术试验。为了使研究结果可信和准确,我们采集了研究区的岩矿样品。根据遥感高光谱影像提取的蚀变信息图,选择样品的收集路线,并记录和确定每一个样品采集的具体位置。通过地面便携式高光谱仪(ASD) 进行了测量和分析,并将结果与 USGS 光谱库进行比较。研究小组采集的 HySpex 高光谱数据是最原始的 DN 值数据,不仅没有坐标系统,而且还有较大的几何变形和大气影响。因此,通过辐射校准、 几何校正和大气校正对数据进行预处理, 为提取蚀变信息做好基础。 首先, 使用 HySpexRAD 软件将数据转换为与辐射单位成比例的格式、 然后使用 PARGE 软件进行基本格式转换,几何校准,并结合 NovAtel 的 SPAN-CPT 仪器记录的姿态数据(POS)进行地形校正。最后,使用基于遥感应用程序基于 MODTRAN4辐射转换开发的 ATCOR 软件进行大气校正。此外,通过 ENVI 5.1 经典版本、 ArcGIS和 CorelDraw 等处理软件来处理和分析图像。通过最小噪声分离(MNF)变换和像素纯度指数(PPI)结合来增强地质背景和 HySpex高光谱数据的信息提取分析。考虑到研究区内感兴趣区域的地质条件重点研究了黄铁矿和绿泥石两个蚀变矿物。经过现场勘查, 我们发现大量的黄钾铁矾暴露在探槽周围和广泛分布在矿化区域。 探槽附近的山顶上出现少量的绿泥石,其余部分则覆盖着厚厚的盐碱壳。我们使用 SAM 算法来比较端元频谱向量与 n 维空间中每个像素向量之间的角度, 较小的角度表示与参考光谱的匹配度更高,而比弧度指定的最大角度阈值远的像素不进行分类。对于图像的定性分类,为每个端元定义一个小角度的角度阈值, 该阈值可提供最佳结果,并用于分类。 SAM 的分类结果为58.03%, Kappa 系数指数为 0.49。对蚀变分布的分析表明,大多数黄钾铁矾的面积分布在探槽周围, 沿探槽呈线性分布。大部分亚氯酸盐都位于的东北部的山坡, 上其余的亚氯酸盐散落在山沟周围。研究结果表明: SVM 分类器非常适合高光谱数据,因为它不假设类的特定统计分布, 该分类器的任务是在 n 维空间中分离类。 通常情况下,这些类不是线性可分离的,因此我们应用了核分数径向基分数(RBF 核)。 SVM 分类的结果显示出 69.25%的准确性, Kappa 系数为 0.59。在手动创建分类的同时,可以纠正倾斜误差,从而提高分类精度。此外, SVM 和 SAM 之间的差异是原始斜率。对于 SAM和 SVM结果对比, SAM结果的总体分类精度分别提高了近 4.3%至 59.57%; SVM则提高了 69.25%。蚀变矿物提取的结果与现场验证结合,结果表明: 支持向量机方法提取的蚀变矿物信息更加集中,而 SAM提取的信息则更加分散。在蚀变方面,从前述方法获得的蚀变矿物信息与实地观测到的蚀变分布相一致。 结果显示与 SAM相比,使用 SVM识别变质矿物的准确性更高且更敏感。超低空 HySpex高光谱探测平台可在空间分辨率处于亚米级的情况下获得连续且完整的光谱目标曲线, 可以为研究目标之间的定量和半定量参数提供技术支持, 并且证明该技术对找矿是有效。 通过本研究结果证明,利用超低空HySpex 高光谱平台探测可以有效地识别蚀变矿物信息,并用于建立光谱与矿物元素含量之间的定性和定量关系。 下一步研究, 我们将在大面积成矿带应用该技术,以探测岩石的类别和矿物组成。
Other AbstractIn mineral exploration, many techniques have been applied to detect the presenceof mineralization and related altered rocks. Remote sensing technology has a range ofadvantages over wide coverage, high efficiency, short cycle, and strong economicbenefits, which is of more critical importance to pathfinders. Among all theseadvantages, the applications of remote sensing, especially hyperspectral technology arevery crucial. Because of, its advanced techniques and methods for the exploration andmapping of minerals, it exhibits significant diagnostic spectral features throughout theelectromagnetic spectrum, which allows for detecting the chemical composition andrelative abundance of deposits of the minerals in the rocks or within a geologicalformation. The HySpex imaging hyperspectral data used in this study has a widespectral response range (400-2500 nm), a narrow bandwidth (2.8 nm / 5.45 nm) andhigher spatial resolution, which effectively applied to minerals extraction of alterationinformation. How to extract effective information from remote sensing images throughremote sensing image classification technology is one of the hot spots and difficultiesin the application of geological sciences in the field of remote sensing. With theadvancement of hyperspectral remote sensing technology, satellite, and airborne remotesensing are superior to ground point measurements due to their resolution and efficiency,and data applications in the field of geology and mineralogy are becoming more andmore extensive and suitable.The jade belt porphyry copper deposit in this study area located in the western partof Karatag, NW Chania. The deposit is mainly composed of intrusive, effusive, andsedimentary rocks, including rhyolite porphyry, porphyry quartz diorite, and gabbro inthe form of reserves they appeared in the western part of the region. The study area isdominated by porphyry quartz diorite mineralization. The administrative division of thestudy area belongs to Wubao Township, Hami City, in the Xinjiang province, and isadjacent to Turpan City in the west. The landforms are mostly low hills, wide valleys,and low hills and quasi-plain areas, with an average altitude of 500-600 meters, amaximum altitude of 693 meters and a relative height difference of tens of meters. Thepurpose of this study is to carry out a low-altitude hyperspectral method of extractingalteration information by using the ultra-low altitude high-precision rapid detectiontechnology platform (the delta wing is equipped with HySpex hyperspectral sensors)and the jade belt porphyry copper mine area as the test area. For more accuracy of thisstudy, we were collected the field data and analyzed it through Analytical SpectralDevices (ASD), and the results were compared with the USGS spectral library.Although through the comparative study, the sample data sets with better quality wereselected to extract the alteration of the study area. According to the anomaly map of thealteration information, first of all, the extraction of information from images data wasdone, then selection of the routes for the samples collected was planned and the specificlocations for the samples collection were determined. During data acquisition, apowered delta wing (Airborne XT91), imaging spectrometer, POS (Position andOrientation System), and control system were collected data from low-altitude flightsin the study area. The power delta wing is a self-powered, high-flying aircraft, which ismade from the most advanced high-tech material.To make the research results credible and accurate, we collected rock samples fromthe study area. According to the alteration information map extracted from the remotesensing hyperspectral image data, were selected the sample collection route, record anddetermine the specific location of each sample. The measurement and analysis werecarried out by the ground portable hyper spectrometer (ASD), and the results werecompared with the USGS spectral library.The HySpex hyperspectral data collected by the research team is the most originalDN value data, not only has no coordinate system but also has large geometricdeformation and atmospheric influence. Therefore, the data were pre-processed throughradiation calibration, geometric correction, and atmospheric correction to obtain a goodfoundation for extracting alteration information. First, use HySpex RAD software toconvert the data to a format proportional to the radiation unit, then use PARGE softwareto perform basic format conversion, geometric calibration, and combined with theattitude data (POS) recorded by NovAtel's SPAN-CPT instrument for terrain correction.Finally, the ATCOR software developed based on the remote sensing application basedon MODTRAN4 radiation conversion was used for atmospheric correction. Besides,the image processed and analyzed by processing software such as ENVI 5.1 classicversion, ArcGIS, and CorelDraw.Through the combination of Minimum Noise Separation (MNF) transformation andPixel Purity Index (PPI) to enhance the information extraction analysis of the geologicalbackground and HySpex hyperspectral data.Considering the geological conditions of the area of interest in the study area, twoaltered minerals jarosite and chlorite were studied. After an on-site investigation, wefound that a large amount of jarosite was exposed around the trench and widelydistributed in the mineralized area. A small amount of chlorite appeared on the top ofthe mountain near the trench and the rest of them was covered with a thick saline-alkalishell. We use the SAM algorithm to compare the angle between the endmemberspectrum vector and each pixel vector in n-dimensional space. A smaller angle indicatesa higher degree of matching with the reference spectrum, while pixels farther than themaximum angle threshold specified by radians are not sort. For the qualitativeclassification of images, a small angle threshold is defined for each endmember, whichprovides the best results and used for classification. The classification result of SAM is58.03%, and the Kappa coefficient index is 0.49. The analysis of the alterationdistribution shows that the area of most jarosite is distributed around the trench andlinearly distributed along the trench. Most of the chlorite is located on the hillside inthe northeast, and the remaining chlorite is scattered around the trenches.The research results show that the SVM classifier is very suitable for hyperspectraldata because it does not assume a specific statistical distribution of classes and the taskof this classifier is to separate classes in n-dimensional space. Normally, these classesare not linearly separable, so we applied kernel fraction radial basis fraction (RBFkernel). The results of the SVM classification showed 69.25% accuracy, and the Kappacoefficient was 0.59. While manually creating classifications, it is possible to correcttilt errors, thereby improving classification accuracy. Besides, the difference betweenSVM and SAM is the original slope. For the comparison of SAM and SVM results, theoverall classification accuracy of SAM results was improved by nearly 4.3% to 59.57%,and the accuracy of SVM was improved by 69.25%. The SVM classifier is well suitedfor the hyperspectral data because it does not assume a specific statistical distributionof classes. The task of this classifier is to separate the classes in n-dimensional space.In our case, the classes are not linearly separable, therefore we have applied kernelfraction Radial Basis Fraction (RBF kernel). The results of the SVM classificationdisplayed 69.25% accuracy with Kappa indices of 0.59. While the classification wasmanually created slope mask, which corrected the inclination errors, to improvement inclassification accuracy. Moreover, the difference between SVM and SAM is the original slope. The classification accuracy of the results was improved from 4.3% to59.57% for SAM, and from to 69.25% for SVM.The results of altered mineral extraction are combined with field verification. Theresults show that the information of altered minerals extracted by the Support VectorMachine method is more concentrated, while the information extracted by SAM is morescattered. In terms of alteration, the altered mineral information obtained from theaforementioned method is consistent with the alteration distribution observed in thefield. The results show that using SVM classification to identify metamorphic mineralsis more accurate and more sensitive than the SAM classification method.The results of this study prove that the detection of the ultra-low altitude HySpexhyperspectral platform can effectively identify altered minerals information and usedto establish a qualitative relationship between the spectrum and the content of mineralelements. In the next step of research, we will apply this technology in large-scalemetallogenic belts to detect the type and mineral composition of the rocks.
Subject Area地球探测与信息技术
Language英语
Document Type学位论文
Identifierhttp://ir.xjlas.org/handle/365004/15392
Collection中国科学院新疆生态与地理研究所
研究系统
Affiliation中国科学院新疆生态与地理研究所
First Author Affilication中国科学院新疆生态与地理研究所
Recommended Citation
GB/T 7714
VATANBEKOV FURKAT. 新疆卡拉塔格地区 HySpex 高光谱影像蚀变信息填图[D]. 北京. 中国科学院大学,2020.
Files in This Item:
There are no files associated with this item.
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[VATANBEKOV FURKAT]'s Articles
Baidu academic
Similar articles in Baidu academic
[VATANBEKOV FURKAT]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[VATANBEKOV FURKAT]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.