基于朴素贝叶斯的高光谱矿物识别

发布时间：2018-03-18 16:25

本文选题：高光谱遥感　切入点：矿物识别　出处：《吉林大学》2014年硕士论文　论文类型：学位论文

【摘要】：高光谱遥感（Hyperspectral Remote Sensing）是在电磁波谱的可见光、近红外、中红外和热红外波段范围内，利用成像光谱仪[1]获取许多非常窄的光谱连续的影像数据的技术[2]。由于高光谱遥感技术在分辨率方面的巨大优势和相关技术的日趋成熟，其应用领域也日益广泛，世界各国对遥感领域的发展都十分重视。矿物识别在地质学中是一个非常重要的课题。也是高光谱遥感技术最能发挥其优势的应用领域之一。所谓矿物识别，是指利用一定的方法和手段能够精确的识别野外获得的矿物样本的种类。高光谱技术的应用，使遥感地质由识别地矿种类发展到识别单矿物以至矿物的化学成分。由于矿物成因的复杂性以及矿物样本的多样性，其反射波谱受到自身化学成分及晶体结构的影像，更受到其他矿物光谱混合其中等因素的影响。另外，由于同一种矿物因形成过程和所处地理条件的影响，其化学成分、晶体结构及光谱特征会仍然有所不同，使其矿物光谱具有地理区域特征；并且由于测量条件的不同，矿物也会随光照强度、背景颜色、风化程度、颗粒大小等因素的原因呈现出不同的光谱特征，这也导致了“同谱异物”和“同物异谱”的现象[5]大量存在，因此目前基于波形匹配的矿物识别方法并不能取得令人满意的效果，在进行矿物识别过程中极易出现混淆和误判现象。数据库技术的发展，使我们有了利用海量数据进行分析的条件。但是，要想从大量的数据中提取有用的信息和规律，并不是意见容易的事情。处理海量数据无论对于软件还是硬件都有着较高的要求。因此人们迫切需要处理海量数据的方法，这导致了数据挖掘技术的诞生和发展。数据挖掘（Data Mining）是指从大量的数据中自动搜索隐藏于其中的有着特殊关系性的信息的过程。作为一个新兴的研究领域，自从20世纪80年代以来，数据挖掘已经取得了显著进展并且涵盖了广泛的应用。与传统的数据分析相比，数据挖掘技术更加关注“是什么”而不再关注“为什么”，即在有大量可靠数据的前提下，我们可以直接获得两种事物之间的相关性，而不必去关注为什么会有这种关联。如今，数据挖掘已经被应用到了众多的领域，例如金融、零售、物流等商业领域以及气象、地质等科研领域。随着数据挖掘领域的不断发展，相应的技术会越来越成熟，，数据挖掘技术会在越来越多的领域发挥出重大作用。高光谱遥感技术的发展，使我们能够利用高光谱遥感数据来处理某些传统的问题。但是由于高光谱遥感拥有海量数据的特点，传统的算法已经不能满足遥感数据处理的需要[4]。而数据挖掘技术的发展为高光谱遥感数据的处理开辟了一个新的方向。通过对海量的高光谱数据进行数据清理、数据集成、数据选择、数据变换、数据挖掘以及模式评估，我们可以挖掘出对我们有用的知识和规律。本文结合高光谱数据的特点与相关特征参数,以朴素贝叶斯、K-均值等分类聚类算法为基础，开发适合于高光谱数据处理及应用的光谱建模、光谱匹配技术，并且采用ENVIi软件自带的高光谱数据库开展高光谱数据挖掘技术研究，探索高光谱遥感数据在矿物识别、矿物特征提取等方面的应用潜力。并且希望能将多种数据挖掘算法结合在一起，形成一个针对高光谱数据的算法体系，并且利用得到的地物波谱数据，充实现有的标准波谱库，推动高光谱遥感领域的研究。
[Abstract]:Hyperspectral remote sensing (Hyperspectral Remote Sensing) is in the electromagnetic spectrum of visible light and near infrared, mid infrared and thermal infrared range, using imaging spectrometer [1] to obtain image data of many very narrow continuous spectrum technology [2]. due to the high spectral remote sensing technology in distinguishing great advantages and related technical aspects of the mature rate and its applications have become increasingly widespread, the world development of the field of remote sensing is very seriously.
Mineral identification is a very important topic in geology. One of the applications of hyperspectral remote sensing technology is the most can exert its advantages. The so-called mineral identification, refers to the use of certain methods and means can accurately obtain the wild type mineral samples. Application of hyperspectral technology, the remote sensing geological identification mineral species developed to identify single mineral and Mineral chemical composition. Because of the complexity of the genesis of mineral and mineral samples diversity, the spectral reflectance is restricted by its chemical composition and crystal structure of the image, more affected by other factors such as the mineral spectral mixing effect. In addition, because of the same minerals due to the formation of the process and effect the geographical conditions, the chemical composition, crystal structure and spectral characteristics will still vary, the mineral spectrum has geographical features; and the measurement conditions Different minerals will be weathered with light intensity, background color, reason, particle size and other factors show different spectral features, which also lead to the same spectrum "and" synonyms spectrum "phenomenon of the existence of a large number of [5], so the current based on waveform matching recognition method and mineral has not been satisfactory results in the process of mineral identification are prone to confusion and misjudgment.
The development of database technology, we have analyzed the conditions of using large amounts of data. However, in order to extract useful information and rules from a large amount of data, is not easy. Massive data processing for both hardware and software have higher requirements. Therefore, there is an urgent need to deal with massive data and this led to the birth and development of the data mining technology.
Data mining (Data Mining) refers to data from a large number of automatic search hidden in the process with a special relationship of information. As a new research field, since 1980s, data mining has made significant progress and covers a wide range of applications. Compared with traditional data analysis, data mining technology pays more attention to "what" and "why", that is no longer concerned in a large amount of reliable data, we can directly obtain the correlation between two things, without having to pay attention to why this association. Now, data mining has been applied to many fields, such as finance, retail, logistics business and meteorological, geological and other research fields. With the continuous development of the field of data mining, the technology will be more and more mature, data mining technology will be more and more The field has played a major role.
The development of hyperspectral remote sensing technology, which enables us to handle some traditional problems by using hyperspectral remote sensing data. But due to the high spectral remote sensing has a mass of data, traditional algorithms can not meet the needs of remote sensing data processing and processing of [4]. data mining technology for the development of hyperspectral remote sensing data has opened up a new the direction of data cleaning, the hyperspectral data of massive data integration, data selection, data transformation, data mining and pattern, we can dig out the knowledge and rules useful to us.
According to the characteristics of hyperspectral data and relevant parameters, to Naive Bayesian, mean K- classification clustering algorithm based on spectral modeling for high spectral data processing and application of spectral matching technique, and using the hyperspectral data library of ENVIi software to carry out the hyperspectral data mining technology research and exploration of hyperspectral remote sensing the data in the application potential of mineral mineral recognition, feature extraction and so on. And I want to be a variety of data mining algorithms together to form a system of algorithm for hyperspectral data, and using the spectral data obtained, enrich the existing standard spectrum library, promote the research field of hyperspectral remote sensing.

【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【参考文献】