KMS XINJIANG INSTITUTE OF ECOLOGY AND GEOGRAPHY，CAS
| 基于随机森林算法的成像高光谱蚀变信息识别技术研究 | |
| 赵杰 | |
| Subtype | 硕士 |
| Thesis Advisor | 周可法 |
| 2017-05-01 | |
| Degree Grantor | 中国科学院大学 |
| Place of Conferral | 新疆乌鲁木齐 |
| Degree Discipline | 理学硕士 |
| Keyword | 随机森林算法 高光谱遥感 蚀变信息识别 random forest hyper-spectral remote sensing alteration minerals information extraction |
| Abstract | Remote sensing technology has been applied to the field of geology since itappeared, and the alteration mineral information extraction is one of the mostimportant applications. How to extract the alteration mineral information quickly andaccurately is a difficult problem. With the development of science and technology, theresolution of the sensor is improved. Therefore it is possible to obtain the continuousspectrum of ground objects, which makes it possible for people to distinguishdifferent alteration minerals.According to the characteristics of the alteration information extraction, scholarshave proposed a variety of methods, such as spectral angle mapper, spectralabsorption index and spectral feature fitting. With the development of machinelearning, decision tree, artificial neural network, support vector machine, randomforest and other algorithms have been introduced into the application of remotesensing classification. The random forest is a new algorithm. Compared to thedecision tree and artificial neural network, which is more stable. The accuracy ofrandom forest is higher than that of the support vector machine, and the computationspeed is faster. It has great application potential in the extraction of hyperspectralinformation. In the process of general classification, most of the training samples areobtained from the image by a priori knowledge and visual discrimination. But theextraction of alteration mineral information is different from the general remotesensing image classification, and the conventional sample extraction methods are notapplicable to this application. Therefore, it is a problem how to extract trainingsamples accurately. In order to successfully achieve the application of random forestalgorithm in the alteration information extraction, we need to solve the problem. Inview of the above problems, a systematic study was carried out. The main researchcontents and results are as follows:（1）In this study, The performance of random forest is investigated and theclassification results are compared with the support vector machine under differentconditions. The results shows random forest has obvious advantages in operation speed and better performance, when the number of training samples used to establishclassification model by the two classifiers is similar. Therefore, random forest cansave a lot of time under the condition of ensuring the classification accuracy, and hashigh application value in the remote sensing classification.（2）A novel method is proposed by combining random forest algorithm withspectral angle mapper. In the experiment, we select training samples by spectral anglematching, then construct a random forest classification model, and carry out theexperiment using the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) dataat Cuprite, Nevada. The results of the study showed that this method is valuable foridentification of alteration information.（3）HySpex hyper-spectral data were applied in practice. In the key areas, theexperiment was carried out to identify and extract the alteration information. Afterfield verification, the method is proved to be effective and has strong practicalsignificance. |
| Other Abstract | 遥感技术是地质找矿领域一项重要的技术手段，其中非常重要的一方面是应用遥感技术提取蚀变矿物相关的信息，并为成矿预测提供信息支持，然而，快速精确地进行蚀变矿物信息提取一直是个难点。遥感技术的飞速发展使得传感器的波普分辨率不断提高，因此使获取地物的连续光谱成为可能，利用成像高光谱来区分不同的蚀变矿物成为热点。在高光谱数据中提取蚀变信息的方法越来越多，逐渐发展出了光谱角匹配、光谱吸收指数、光谱特征拟合等多种方法，其中光谱角匹配技术是应用最广泛的方法。随着统计学习理论的发展，决策树、人工神经网络、支持向量机、随机森林等方法也被引入遥感分类研究中，本文将对随机森林算法进行深入的应用研究。随机森林算法作为一种比较新的算法，与决策树算法和人工神经网络算法相比，具有稳定性强，运算速度快和不易造成过学习的特点，而精度与支持向量机相当，因此，随机森林算法在成像高光谱蚀变信息提取方面具有很大的潜力。遥感谱蚀变矿物信息提取不同于一般的遥感图像分类，以先验知识从影像上获取训练样本的方式在蚀变矿物信息提取中并不适用，如何精确地提取蚀变矿物信息训练样本是随机森林算法的一个难点。针对以上问题本文做了系统研究，主要研究内容与结果如下：（1）本文系统研究了随机森林算法的基本原理和方法，分析了随机森林算法的在遥感地物识别与分类中的精度和运算效率，并从识别类精度和时间成本两个方面对算法性能进行了评价。实验表明随机森林算法能够在保证精度的条件下节约大量时间，在成像高光谱遥感数据处理方面具有很大的潜力。（2）针对随机森林算法在高光谱蚀变信息识别中的适用性问题，提出了一种将随机森林算法与光谱角匹配技术相结合的识别方法，实验过程中以光谱角匹配技术选取训练样本，克服了算法在蚀变信息样本提取中的难点，然后构建随机森林识别训练模型，并利用 Cuprite 铜矿区的高光谱影像进行了实验，验证了算法的适用性。（3）在已构建的蚀变信息识别方法的基础上，利用卡拉塔格玉带铜矿HySpex 成像高光谱数据，在实验区的重点区域开展了蚀变信息的识别提取和应用，经过实地验证，表明该方法效果较好，具有很强的应用实践意义。 |
| Subject Area | 地图学与地理信息系统 |
| Language | 中文 |
| Document Type | 学位论文 |
| Identifier | http://ir.xjlas.org/handle/365004/14915 |
| Collection | 研究系统_空间对地观测与系统模拟研究室 |
| 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 | 赵杰. 基于随机森林算法的成像高光谱蚀变信息识别技术研究[D]. 新疆乌鲁木齐. 中国科学院大学,2017. |
| 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