KMS XINJIANG INSTITUTE OF ECOLOGY AND GEOGRAPHY，CAS
| 基于 HySpex 成像高光谱数据的蚀变矿物填图技术研究 | |
| 唐超 | |
| Subtype | 硕士 |
| Thesis Advisor | 周可法 |
| 2018-06-01 | |
| Degree Grantor | 中国科学院大学 |
| Place of Conferral | 新疆乌鲁木齐 |
| Degree Discipline | 工学硕士 |
| Keyword | 蚀变填图，高光谱遥感，光谱匹配 极限学习机 HySpex Alteration Map Hyperspectral Remote Sensing Spectral Matching Extreme Learning Machine HySpex |
| Abstract | 成像高光谱传感器可以通过数十至数百个波段对地物成像，在获取地物空间信息的同时获取地物连续光谱信息，极大提升了对地物的识别能力。本研究使用的 HySpex 高光谱数据具有光谱响应范围宽、波段宽度窄和空间分辨率高等特点，能更加有效地应用于矿物蚀变信息提取。如何通过遥感图像分类技术从遥感影像中提取有效的信息是遥感地质科学应用研究的热点和难点。成像高光谱影像数据量大、数据冗余且处理困难，需要引入新算法以提升数据处理效率。目前遥感影像分类方法可分为两种类型，第一类是基于光谱曲线的处理方法，根据光谱特征识别矿物。第二类是基于机器学习的处理方法，随着统计模式识别与机器学习领域中新算法不断涌现，决策树、人工神经网络、支持向量机和随机森林等方法也被引入到高光谱图像处理中。机器学习算法中，极限学习机相对其他神经网络具有训练速度快、泛化能力强等优点。因此，其在高光谱蚀变信息提取方面具有很大的潜力。本研究将极限学习机引入遥感图像识别中以提高蚀变信息提取的速度。然后对光谱特征匹配算法和极限学习机的蚀变提取结果进行对比分析，得到最优的蚀变提取结果。最后，本研究应用极限学习机和光谱特征匹配算法进行了深入的蚀变矿物填图技术应用研究，取得以下主要成果：（1）成功搭建了超低空遥感探测平台。本部分的主要工作包括仪器安装、调试、人员调度安排和空地协同等工作。在此过程中进行了包括航线设计、仪器控制和参数厘定等工作，同时在研究区开展了飞行试验，采集到丰富的 HySpex高光谱影像数据。通过几何校正、大气校正、几何精校正、影像裁剪和拼接等预处理工作，获得了高质量的成像高光谱影像数据，为蚀变信息提取提供基础数据。（2）基于极限学习机的研究，构建了蚀变矿物信息提取模型。通过研究极限学习机的原理，探讨了不同隐含层节点数量等参数对模型的影响。同时建立了基于 ELM 算法的蚀变信息提取模型。通过了本部分的研究，开发了极限学习机蚀变提取程序，实现了基于 ELM 模型的蚀变矿物信息提取。（3）基于光谱特征匹配算法研究，构建了蚀变矿物信息提取模型。通过研究蚀变矿物样品的光谱特征，确定了样品光谱特征位置。据此实现了基于光谱匹配算法的蚀变矿物信息提取。（4）模型应用效果对比。通过对本研究提出的蚀变矿物信息提取模型的应用效果进行野外验证，对比分析了基于 ELM 和光谱匹配模型的提取精度，剖析了两种方法在蚀变矿物信息提取中应用效果产生差异的原因，在此基础上，提出了适合本研究区的蚀变矿物信息提取技术方案。 |
| Other Abstract | The imaging hyperspectral sensor can image ground objects through tens tohundreds of bands, and acquire the continuous spectral information of the groundfeatures at the time of acquiring the spatial information of the ground features, whichgreatly improves the ability to recognize the ground objects. The HySpexhyperspectral data used in this study has a wide spectral response range, narrow bandwidth and high spatial resolution, and can be more effectively applied to theextraction of mineral alteration information. How to extract effective informationfrom remote sensing images through remote sensing image classification technologyis a hotspot and difficulty in the application of remote sensing geological science.Imaging hyperspectral image has a large amount of data which cause processingdifficulties. It is need to introduce new algorithms to improve data processingefficiency.At present, remote sensing image classification methods can be divided into twotypes. The first type of processing methods is based on spectral curve identifyingminerals. The second category is based on machine learning. With the continuousemergence of new algorithms in statistical pattern recognition and machine learning,Lots of methods, such as decision trees, artificial neural networks, support vectormachines, and random forests have also been introduced into hyperspectral imageprocessing. In machine learning algorithms, extreme learning machines have theadvantages of faster training and generalization than other neural networks. Therefore,it has great potential in extracting information from hyperspectral alterations. In thispaper, extreme learning machine has introduced into remote sensing imagerecognition to improve the speed of information extraction. Then, the spectral featurematching algorithm and the extreme learning machine's alteration extraction resultswere compared and analyzed to obtain the optimal alteration extraction results. Finally,this study applied extreme learning machine and spectral feature matching algorithm to conduct in-depth research on the application of alteration mineral mappingtechnology, and achieved the following main results:（1） Successfully built an ultra-low altitude remote sensing platform. The maintasks of this section include instrument installation, commissioning, personnelscheduling and airspace coordination. In this process, we included route design,control and parameter determination. At the same time, flight experiments werecarried out in the study area, and a wealth of HySpex hyperspectral image data wascollected. Through geometrical correction, atmospheric correction, geometricprecision correction, image cropping and splicing, and other preprocessing work,high-quality imaging hyperspectral image data was obtained, providing basic data foralteration information extraction.（2） Based on the study of the extreme learning machine, an altered mineralinformation extraction model was constructed. By studying the principle of extremelearning machine, the influence of parameters such as the number of nodes indifferent hidden layers on the model is discussed. At the same time, an alterationinformation extraction model based on ELM algorithm was established. Through theresearch of this part, the extreme learning engine alteration extraction program wasdeveloped and the alteration mineral information extraction based on the ELM modelwas realized.（3） Based on the study of spectral feature matching algorithm, a model foraltered minerals information extraction was constructed. By studying the spectralcharacteristics of altered mineral samples, the spectral feature locations of the sampleswere determined. Based on this, the alteration mineral information extraction based onspectrum matching algorithm was realized.（4） Comparison of model application effects. Through field verification of theapplication effect on the altered mineral information extraction model presented inthis study, the extraction accuracy based on ELM and spectral matching model wascompared and analyzed, and the reasons for the differences in the application effectsof the two methods in were analyzed. Based on this, we proposed a technologicalscheme is suitable for extracting altered mineral information in the study area. |
| 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/365004/14988 |
| Collection | 研究系统_荒漠环境研究室 |
| 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 | 唐超. 基于 HySpex 成像高光谱数据的蚀变矿物填图技术研究[D]. 新疆乌鲁木齐. 中国科学院大学,2018. |
| 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