Dissertation > Excellent graduate degree dissertation topics show

Spectral Feature Extraction and Matching Resesarch on Hyper-Sectral Remote Sensing Image

Author: PiSheng
Tutor: GuoKe；ChenZuo
School: Chengdu University of Technology
Course: Applied Mathematics
Keywords: Hyperspectral Absorption feature Decision Tree Match
CLC: O433.4
Type: Master's thesis
Year: 2011
Downloads: 154
Quote: 0
Read: Download Dissertation

Abstract

Sustainable and rapid development of China's economic mineral resources and energy demand is enormous. However, our main mineral resources in the current security situation is very serious. Remote sensing technology as an effective earth observation tools, has gone through the full-color photography, color photography, multi-spectral remote sensing, hyperspectral remote sensing different historical stages. From the 1970s onwards, the remote sensing technology has been widely applied to land and resources survey and monitoring areas, and achieved fruitful results. Hyperspectral remote sensing has greatly enhanced remote sensing earth observation capacity and the ability to identify the surface features, and improve the level of quantitative remote sensing technology, remote sensing from the identification of surface features directly identify the stage of development to the surface features using hyperspectral remote sensing technology to detect mineral is one of the main direction of the remote sensing technology applications. Hyperspectral remote sensing image classification method of mineral identification is mainly divided into two types, one is based on the mathematical transformation mathematical statistics image dimensionality reduction, such as principal component analysis. Another mineral spectra form a physical mechanism, the direct use of the characteristics of high spectral resolution of hyperspectral remote sensing data by selecting the absorption spectrum, calculated spectral absorption characteristics of rock and mineral identification, such as spectral feature fitting. Feature matching effectively combine these two types of methods to achieve mineral identification classification from the physical mechanism of the formation of mineral spectrum research diagnosis spectral characteristics of the analysis of surface features, and the use of mathematical statistics theory method to achieve the mapping of mineral identification Fine and classification become the focus of this study. Multi-feature matching the decision trees mineral identification and classification of mapping technology is a knowledge discovery and expression, defined by the rules, the process of establishing and running the decision tree. Existing types of mineral identification decision tree or using only a single feature (such as the spectra of the main absorption peak position characteristics) and the same classification algorithm for mineral identification, or use the different features they use the same algorithm, but these two decision tree in to some extent limit the improvement of the recognition accuracy. To improve the accuracy of mineral identification, the absorption characteristics of the spectrum of the various typical alteration minerals multivariate decision tree: First calculate the spectral absorption characteristics of the study area typical alteration minerals, and these characteristics are expressed in the form of knowledge . Then calculated the spectral characteristics of the image information entropy, selected according to the size of the information entropy spectral absorption index, spectral absorption depth, the class characteristics of the spectral slope, left area combined with the completion of the main absorption characteristics of the mineral itself mineral identification rules defined . Finally, on the basis of the rules defined build and run the decision tree classification mapping results. The result of China University of Geosciences (Wuhan) HUANG Ding-hua and China Land and Resources aerial remote sensing center mapping results, especially epidote and serpentine mapping results agree well. Muscovite, chlorite, calcite, three types of results are slightly different. This method of knowledge representation and rule definition mineral mapping is feasible, effective, and it can be achieved to some extent, the mineral type and distribution of mapping the analog data, also established multi-feature hybrid decision tree classification and recognition tree classification ability. For the quantitative evaluation of the decision tree classification accuracy, this paper will build multi-feature decision tree and SAM method applied to multi-feature decision tree to identify the correct rate of 89.06%, while the SAM method correct rate based on USGS spectral library randomly generated the analog image: The results showed 79.99%. This shows that the decision tree can still be effective in the case of missing feature priori knowledge part of the study area to maintain good robustness. However, the research focused on the typical altered minerals continue to expand other categories of alteration minerals, to improve the accuracy of identification will be the focus of future research.

Related Dissertations

CLC: > Mathematical sciences and chemical > Physics > Optics > Spectroscopy > Spectroscopy
© 2012 www.DissertationTopic.Net Mobile