KMS XINJIANG INSTITUTE OF ECOLOGY AND GEOGRAPHY，CAS
| 基于成像高光谱的蚀变信息识别及尺度影响研究 | |
| Alternative Title | Research on Alteration Minerals Identification and the Effect of Spatial Scale on Imaging Hyperspectral |
| 梁丹迪 | |
| Subtype | 硕士 |
| Thesis Advisor | 周可法 |
| 2019-06-30 | |
| Degree Grantor | 中国科学院大学 |
| Place of Conferral | 北京 |
| Degree Discipline | 工学硕士 |
| Keyword | 高光谱遥感 混合调制匹配滤波 支持向量机 蚀变矿物填图 空间尺度 Hyperspectral Remote sensing Mixed Tuned Match Filtering Support Vector Machine Alteration Minerals Mapping Spatial Scale |
| Abstract | 高光谱遥感技术又称为成像光谱遥感技术，它是一种将遥感影像和光谱探测融合于一体的技术。高光谱遥感影像具有精细的光谱分辨率，提供了地物更多的光谱细节信息，在蚀变信息识别方面独具优势。在岩矿高光谱遥感影像中，单个像元的光谱曲线表征对应地物的诊断性光谱特性，可以帮助人们识别岩矿类型和组分，实现矿产资源的大面积勘查。有关岩矿信息识别与矿物填图的研究也一直是高光谱遥感在地质领域的主要研究方向。Hyspex 成像光谱仪获取的高光谱遥感数据， 具有高光谱分辨率和高空间分辨率的特点，是一种较新的高光谱数据。 目前针对 Hyspex 空中飞行获取的数据前人研究较少， 在高光谱遥感信息识别方面缺乏完整的方法技术体系。 本研究使用 Hyspex 获取的成像高光谱数据做蚀变信息提取，分别采用传统的基于光谱特征的线性光谱混合模型的匹配调制滤波法和基于机器学习算法的支持向量机法做蚀变填图，对比分析两种方法蚀变填图结果的可靠性和精度。在蚀变填图的基础上， 研究不同空间尺度的成像高光谱影像数据对蚀变信息提取总体精度以及类别精度的影响。本研究的主要工作和成果如下：（1） 使用光谱仪采集样品光谱，对其做包络线去除、一阶导数和二阶导数运算。在此基础上，分别分析和总结 USGS 岩矿光谱标准库中矿物的光谱吸收特征和样品的光谱吸收特征， 为本文之后开展蚀变矿物填图相关的研究奠定理论基础。（2） 对野外获取的成像高光谱遥感影像做预处理，包括辐射定标、几何校正、大气校正、 配准、 拼接裁剪等，消除大气、外界自然环境和飞行过程中主客观因素等的影响，得到遥感影像真实的反射率信息， 获取蚀变信息提取的基础影像。（3） 采用 MTMF 和 SVM 两种方法做蚀变填图， 针对填图结果做野外验证，对比分析两种方法蚀变填图结果的可靠性和精度。通过野外勘察，从提取的蚀变信息分布的大体位置和蚀变信息的类别等方面，表明 Hyspex 成像高光谱遥感对玉带矿区蚀变填图的结果是可靠的。 同时， 对两种方法的蚀变填图结果做定量评价，得到 MTMF 和 SVM 方法蚀变填图的总体精度分别为 57.14%和 72.86%，SVM方法提取结果的精度明显高于 MTMF 方法的。因此，本研究采用 SVM 方法对Hyspex 高光谱数据做蚀变填图效果更佳。（4） 在蚀变信息提取的基础上，进一步研究 Hyspex 高光谱遥感影像的空间尺度效应。采用最邻近重采样法对 Hyspex 数据做空间尺度扩展，采用 SVM 方法对各空间分辨率的影像做蚀变填图和野外验证。 当空间分辨率从 1.5m 降到6m，总体精度分别下降 5.71%、 8.58%和 7.15%。 1.5m 到 2m 空间分辨率下蚀变填图的总体精度相对变化不大，而从 2m 到 4m 空间分辨率下的总体精度下降较大，其蚀变填图结果受空间分辨率的影响较大。从定性和定量的角度综合分析来看，如果需要大面积探测时，兼顾探测效率（获取面积大、时间短、成本低）和精度要求，可选择 2m 空间分辨率的影像所对应的航高（1400m）为最佳飞行高度。 |
| Other Abstract | Hyperspectral remote sensing technology, also known as imaging spectral remotesensing technology, is characterized by integrating remote sensing image with spectraldetection. Hyperspectral remote sensing images have fine spectral resolution, whichcould provide more spectral detailed information of the ground object and thereforehave unique advantages in quantitative analysis of material composition. In thehyperspectral remote sensing images containing geology information, a single spectralpixel is able to manifest diagnostic spectral characteristics of rocks and minerals.Hence it could help identify minerals and their chemical components, and conductlarge-scale exploration of mineral resources. Research on identification of rocks andminerals, and mineral mapping, has also been one of the main directions ofhyperspectral remote sensing in geology.The remote sensing image obtained by the Hyspex imaging spectrometer, withfine spectral resolution and spatial resolution, is a more advanced, relatively nichehyperspectral data. At present, especially for mineral mapping and identificationresearch on the methods, technical processes and spatial scale effects of Hyspeximaging hyperspectral data has been in urgent need. In this study, the hyperspectraldata obtained by Hyspex were used to extract detailed information of mineralalteration. This study applied mixed tuned match filtering algorithm based on thelinear spectral hybrid model of the traditional spectral modulation, and support vectormachine algorithm based on the machine learning on the hyspex image to obtain thedistribution map of alteration minerals in the YuDai copper deposit in the Kalatagdistrict. The reliability and accuracy of the results of minerals mapping by using twomethods were analyzed, respectively. Then, based on the minerals mapping, studies ofthe effects of Hyspex imaging hyperspectral data on the overall accuracy andclassification accuracy of the minerals identification and mapping were carried out.Several achievements have been reached in our research:（1） The spectra of rocks sample were collected by the ASD spectrometer, andrespectively use the continuum removal, first derivative and second derivativemethods to analyze. On the basis of, the spectral absorption characteristics of theminerals in the USGS rock ore spectral standard library and the spectral absorptioncharacteristics of the rocks sample were analyzed, which laid a theoretical foundationfor the research related to alteration minerals mapping after the paper.（2） The pre-processing of Hyspex imaging hyperspectral data collected in thefield was finished, including radiometric calibration, geometric correction,atmospheric correction, registration, etc. The real reflectivity of remote sensingimages were obtained, and to ensure the quality of research data.（3） MTMF and SVM algorithm were employed to conduct the alterationmapping. The results were tested in the field. The reliability and accuracy of twomethods were analyzed. Through field verification, the locations of the alterationminerals and their alteration types were approved to match relatively well to theresults obtained from the Hyspex images. Moreover, the results of the alterationmapping using two methods are quantitatively evaluated. The overall accuracy of thealteration mapping by the MTMF and SVM methods is 57.14% and 72.86%,respectively. The accuracy of the SVM method is significantly higher than that of theMTMF method. Therefore, SVM method is considered more feasible to identify thealteration minerals.（4） Based on the result of minerals mapping, the spatial scale effect of mineralmapping in imaging hyperspectral data was studied further. Hyspex data was spatiallyscaled by nearest neighbor resampling method, and then the SVM method were usedto perform alteration mapping and field verification analysis for each spatialresolution image. After analysis, when the spatial resolution is reduced from 1.5m to6m, the overall accuracy decreases by 5.71%, 8.58% and 7.15%, respectively. Theoverall accuracy of the altered mapping under 1.5m to 2m spatial resolution isrelatively small, while the overall accuracy decreases from 2m to 4m spatialresolution, and the results of the alteration mapping are greatly affected by spatialresolution. From the perspective of qualitative and quantitative analysis, if large-area detection is required, taking into account the detection efficiency (large acquisitionarea, short time, low cost) and accuracy requirements, the altitude (1400m)corresponding to the image with 2m spatial resolution can be selected. |
| Subject Area | 地球探测与信息技术 |
| Language | 中文 |
| Document Type | 学位论文 |
| Identifier | http://ir.xjlas.org/handle/365004/15323 |
| 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 | 梁丹迪. 基于成像高光谱的蚀变信息识别及尺度影响研究[D]. 北京. 中国科学院大学,2019. |
| Files in This Item: | There are no files associated with this item. | |||||
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.
Edit Comment