【作者】 杨佳佳；

【导师】 姜琦刚；

【作者基本信息】 吉林大学 ， 地质资源与地质工程， 2012， 博士

【摘要】 岩石和矿物都是地质作用的产物，矿物是岩石的基本组成单位。矿物是由地质作用形成的天然单质或化合物，不同的矿物及矿物组合组成性质不同的各类岩石。矿物不仅包含不同的化学成分，而且具有一定的内部结构，随着外界条件的改变，其成分、结构会发生变化，形成新的矿物。岩石是由一种或多种矿物组成的，具有一定结构构造的集合体，广泛分布于地壳中，是地质作用发展的历史记录。通过对岩石、矿物的研究可以探索和了解地壳的发展、变化及运动规律，为找矿及地质研究提供重要依据。矿产赋存于岩石中，伴随着地质作用而呈现出一定的规律，表现为一定的矿产类型总与一定的岩石类型有着依赖关系。展开对岩矿信息的地质研究工作能够为探测、查明、开产矿产资源提供地质资料，在矿产预测和开发中显示出岩矿研究工作的经济意义及实用价值。遥感地质是综合应用现代遥感技术来研究地质规律，进行地质调查和资源勘察的一种方法。它从宏观的角度，着眼于由空中取得的地质信息，以各种地质体对电磁辐射的反应作为基本依据，结合其他各种地质资料的综合应用，以分析、判断、识别一定地区内的地质情况。近年来，随着多、高光谱遥感技术的发展，光谱分辨率的提高，遥感技术在岩矿信息识别和地质矿物识别填图等领域中的应用有了广泛的前景。岩矿光谱特征与其物化属性的关联分析是遥感岩矿信息识别与提取的基础。这方面的研究大体包括从两方面进行：一是从岩石矿物特征光谱为切入点，结合岩石矿物的化学成分、晶体结构等物化知识进行相关性分析以提取岩矿信息；二是以物理模型为主，结合矿物光谱知识进行岩矿信息提取。目前，遥感技术用于岩矿信息识别及地质填图主要依赖于地物的光谱特征，是直接利用岩石矿物的光谱特征进行岩矿识别，定量、半定量提取岩矿信息。过度地依赖岩石、矿物的光谱信息往往忽略了矿物中化学成分与矿物、矿物与岩石之间的关系。大气辐射对地物波谱特征的影响、混合光谱造成的光谱漂移、变异等对单个波形的影响，导致多环节引入波谱特征对岩矿信息提取结果有较大的干扰。本文基于遥感技术，结合地质研究方法提取青海省格尔木辖区内乌图美仁地区岩矿信息。分为两步进行：一是在该区区域地质背景下，基于遥感图像灰度及纹理特征，采用图像增强的一些方法，扩大图像中不同岩性之间的灰度及纹理差别，突出地物目标信息，使之易于目视解译，在此基础上建立感兴趣区，利用监督分类的方法对图像进行分类，达到区分不同岩性的目的；二是在此基础上，对不同的分类地区进行层层剥离。从对遥感波谱特征的分析出发，建立矿物中氧化物与光谱特征的光谱数学物理模型，借助标准矿物计算（CIPW）的岩石化学计算方法，按照岩矿光谱信息—氧化物—矿物—岩石的研究方法，对该区岩石、矿物进行识别并定量、半定量提取岩矿信息。本次研究从不同于以往的角度出发为遥感技术用于岩矿信息识别、提取矿物含量和化学成分等提供了一种新的思路。主要研究成果和创新点如下：1、研究基于不同源遥感数据的图像融合增强方法。以ASTER数据和QucikBird-2数据为遥感数据源，实现了对多光谱图像和高分辨率图像的融合，得到了多光谱高分辨率影像，增强了多光谱图像的空间分辨率。实验表明，HIS加小波融合的方法在提高分辨率的同时，在最大熵值及光谱特征保持的定量评价下都取得较好的效果。增强后的图像在应用于感兴趣区建立以及图像分类时，可以解决由于高分辨率图像光谱分辨率不足而导致的解译、分类效果不佳、纹理粗糙等问题，空间及光谱分辨率增强后的融合图像提高了遥感图像整体质量和综合分析精度，分类效果明显优于融合前图像。2、基于LS-SVM分类技术的高分辨率多光谱图像分类方法研究。以最小二乘支持向量机（LS-SVM）分类技术为技术手段，基于融合后的高分辨率多光谱图像，提取图像中与岩性相关的纹理、形状、光谱信息，选取过程中以图像的光谱和纹理为主要特征信息，同时以J-M距离、转换分类度为依据选取最优特征空间，采用因子分析变换降维对特征空间进行压缩，使特征信息实现最优化，然后对已知样本进行训练，建立分类模型，评价模型精度。最后利用该模型对研究区进行岩性划分，并进行分类后处理。结果表明：加入纹理等信息后的LS-SVM分类模型更加利于岩性的判别；基于LS-SVM的分类方法在遥感图像岩性识别中表现良好。3、利用支持向量机回归方法改进了氧化物质量分数定量反演的精度。矿物中氧化物的反演不是一个简单的线性问题。在对比了一些线性回归算法与支持向量机回归算法之间的回归效果之后，提出了基于支持向量机的回归模型，并针对这个模型从核函数方面进行了改进，改进后的回归模型提高了氧化物的反演精度。实验表明，在提供同样数量样本的情况下，该方法反演结果更加准确，应用效果更好。4、将CIPW岩石化学计算方法引入到遥感地质矿物定量识别中，提出“岩矿光谱信息—氧化物—矿物—岩石”的岩矿信息提取方法。将CIPW岩石化学分析方法引入到遥感地质工作中来，把遥感图像中的每一个像元作为一个岩石样本，以典型岩石、矿物的诊断性波谱特征为基础，建立矿物中氧化物与光谱特征的光谱数学物理模型，对每个像元计算其化学成分，然后进行CIPW标准矿物计算，得到标准矿物定量信息，最后按照岩矿光谱信息—氧化物—矿物—岩石的研究思路进行计算，得到了该区定量、半定量岩矿信息提取结果。5、根据地质成岩理论建立岩性识别规则，提取岩性信息。岩石中矿物或矿物组合按一定的规律共同存在，反映着统一的形成条件，依一定的地质作用而共同组合。岩石由于成因的不同和外在条件的改变，所包含的矿物也会发生变化。从地球化学的观点来看，同一地层内的不同岩性所含的化学元素是多样的，各元素的含量又是非常不均匀的，从而导致其氧化物、矿物含量也有所不同。再从岩石学的角度来看，不同的岩性必然是由种类繁多及多种组合矿物形成。基于地质理论，在已经圈定的岩性分类基础上，以不同岩石中氧化物含量、矿物含量差异为基准，建立岩性识别规则，按照层层剥离的原则，对研究区内的岩性进行分类识别。更多还原

【Abstract】 Rocks and minerals are the products of geological processes. Mineral is the basicunit of rock. The mineral is a simple substance or compounds formed by geologicalprocesses. Different minerals or the combinations of them compose various types ofrocks of different nature. The minerals contain different chemical compositions andhave certain internal structures. As the external conditions change, the compositionsand structures also make some changes that may lead to the formations of newminerals. Rock is the assemblage of one or more kinds of minerals aggregating with acertain texture and structure. It is widely distributed in the earth’s crust and is therecord of the history of geological developments.Through the study of rocks and minerals, both the exploration and theunderstanding of crust development, changes as well as the laws of movements can beachieved so that bring about important inferences for prospecting and geologicalstudies. Minerals host in the rocks, along with geological processes and show certainregularity of the dependency of the type of rocks on the types of the minerals.Launching the geological study of the rock and mineral information can providegeological data and material for exploring, probing, and mining the mineral resourceswhich shows the economic significance and practical value of the rock and mineralresearch work in the mineral prediction and development.Remote sensing in geology is a technique that comprehensively applies modernremote sensing techniques to study the geological rules as well as carry out geologicalsurvey and resources exploration. It focuses on the geological information obtainedfrom the air from a macro perspective and takes the reactions of a variety of geologicbodies of electromagnetic radiation as a fundamental basis. This technique iscombined with a variety of other integrated applications of geological data so that itcan be used to analyze, to judge, and to identify certain areas of geological situation.In recent years, with the development of multi-spectral and hyper-spectral remotesensing technology as well as the improvement of spectral resolution, the applicationsof remote sensing technology have good prospects of development in several fieldssuch like the identification of rocks and minerals, the identification of minerals andmapping and so on.Associated analysis of spectral features of rocks and minerals together with theirmaterialization properties is the basis of the identification and extraction of rocks andminerals via the means of remote sensing. The research in this area can be generally divided into two aspects: one is to take the characteristic spectra of rocks and mineralsas a starting point and carry out relevance analysis by taking the chemicalcomposition of rocks and minerals, the crystal structure or some other physical andchemical knowledge into consideration to extract the rock and mineral information;another is to put emphasis on the physical model and extract the information of rocksand minerals with the knowledge of spectra of them. Currently, remote sensingtechnology in the identification of information of rocks and minerals as well as thegeological mapping are mainly dependent on the spectral characteristics of the surfacefeatures. It directly makes use of the spectral characteristics of the rocks and mineralsto identify them and extracts their information quantitatively or semi-quantitatively.Excessive reliance on the spectral information of rocks and minerals often results inthe overlook of the relationship between the chemical composition of minerals andmineral or the mineral and the rocks. The influence on the characteristics of objectspectrum that is caused by the atmospheric radiation and the influence on the singlewaveform caused by the spectral shift or the mutation that is brought about by themixed spectrum may result in the repeated introductions of the spectral characteristicsand may lead to greater interference of the results of rock and mineral informationextraction.Based on remote sensing technology, this paper combines itself with geologicalresearch methods to extract the information of rocks and minerals within Wutumeirenin the area of Golmud in Qinghai Province. The study can be divided into two steps:First, in geological background of this region that mentioned above, the author adoptsthe methods of image enhancement based on the gray level and texture features of theremote sensing images and expand the differences in the gray level and texturefeatures between various lithological characters in order to give prominence to theinformation of the surface features. This makes it easy for the visual interpretation andestablishment of the regions of interest. Besides, this paper uses the method ofsupervised classification to classify the images to meet the purpose of distinguishingdifferent lithological characteristics. Moreover, it also separates the regions ofdifferent classifications.From the analysis of the spectral characteristics viewed by the techniques ofremote sensing, the author establishes the mathematical and physical models of therelationships between the oxides in the minerals and the characteristics of thespectrum. Via using the chemical computing method of Norm Mineral Calculation(CIPW) for the rocks, according to the method of spectral information–oxide–mineral–rock, this paper carries out the identification of the rocks and mineralsin the certain region as well as the extraction of the information of the rocks andminerals quantitatively or semi-quantitatively.This research carries out investigation from the angle that is quite different fromthe ones taken in the previous studies and provides a new way of thinking about the identification of the information of rocks, the content of the mineral content and thechemical composition, etc.The main research results and innovations of this research are as following:1.The investigation of the method of the integration of image enhancementbased on different remote sensing data of multiple sources.Taking the ASTER data and QucikBird-2data as the data sources, this paperrealize the target of the integration of multi-spectral images and high resolutionimages and gets the images with the good qualities of both the two kinds of images sothat it improves the spatial resolution of multispectral images. The experiments showthat the HIS wavelet fusion method can improve the resolution; at the meanwhile, itcan also achieved good results in the quantitative evaluations of the maximum entropyas well as the maintenance of spectral features.When applied to the establishment of the regions of interest and the classificationof images, the enhanced images can solve the problems of the low quality ofinterpretation and classification, the coarseness of the texture that are caused by theinsufficient spectral resolution of the high-resolution images. This in turn brings aboutthe improvement of the quality of the integrated image and the precision of thecomprehensive analysis. The classification based on these images is better than othersbased on the images without the integration.2.The classification methods based on the LS-SVM classification techniqueswith high-resolution multi-spectral image.Using least squares support vector machine (LS-SVM) classification techniques,based on merged high-resolution multi-spectral image, texture, shape and spectralinformation of image with the relevance of lithology were extracted. Spectral andtexture of image were treated as the main feature information in the process ofselecting. Based on JM distance and conversion of classification degree, the optimalfeature space was selected. Feature space was compressed in factor analysis totransform the dimensionality reduction so that feature information achieved optimaleffect. Training the known samples, establishing classification model and evaluatingthe accuracy of the model followed. Finally, the study area was dealt with lithologypartition and classification post-processing by the model. The results showed that:LS-SVM classification model with the texture and other information were moreconducive to lithology discrimination; the LS-SVM classification method performedwell in the remote sensing images lithology identification.3.Support vector machine regression method improves the accuracy ofquantitative retrieval of oxide weight percent.The retrieval of the mineral oxide is not a simple linear problem. Aftercontrasting the regression effect between some linear regression algorithms andsupport vector machine regression algorithm, support vector machine regressionmodel was proposed and improved in terms of kernel function. The improved regression model improved retrieval accuracy of the oxide. The experiments show thatthe retrieval result of the method is more accurate and its application effect is better inthe case of the same number of samples.4.The CIPW rock chemistry calculation method is introduced into quantitativeidentification of remote sensing geological mineral and put forward a″rock andmineral spectral information-oxide-mineral-rock″method of extracting rock andmineral information.The CIPW rock chemical analysis method is introduced into the remote sensinggeological work, regarding each pixel of remote sensing images as a rock sample.Based on typical rock and diagnostic spectral characteristics of the mineral, aspectral mathematical physics model of oxide in the mineral and the spectralcharacteristics of mineral were established. The chemical composition of each pixelwas calculated. Then proceed to the CIPW Standard mineral calculation to get thestandard mineral quantitative information.At last, calculating according to the rock and mineral spectral information-oxide-mineral-rock research idea, we will get quantitative and semi-quantitative rockand mineral information in this area.5.According to the theory of geological diagenesis, lithology identification ruleswere established to extract lithology information.Minerals and mineral assemblages according to certain rules exist in the rock. Itreflects the uniform formation conditions. They combine according to a certaingeological effect. Due to different causes of formation and the changes of externalconditions, the mineral which rocks contain will change. From the geochemical pointof view, the chemical elements contained in different rocks in the same stratum isdiverse and the content of each element is very uneven, resulting in the contents ofoxide and mineral are also different. From the petrological point of view, differentlithology must be formed by a wide range and variety of combinations of minerals.Based on geological theory, the delineated lithology classification is seen as abenchmark. Based on the content of oxide in different rocks and the difference ofmineral content, the lithology identification rules is established. According to theprinciple of layer stripping identification, the lithology in the study area can beclassified.更多还原