Location:Home > Engineering science > Information and Communication Engineering > Target Detection Techniques by Using Convex-based Unmixing in Hyperspectral Images
Details
Name

Target Detection Techniques by Using Convex-based Unmixing in Hyperspectral Images

Downloads: []
Author
Tutor: ZhangZuo
School: Harbin Institute of Technology
Course: Information and Communication Engineering
Keywords: hyperspectral unmixing,spatial correlation,convex,subpixel-mapping,target detect
CLC: TP751
Type: Master's thesis
Year:  2012
Facebook Google+ Email Gmail Evernote LinkedIn Twitter Addthis

not access Image Error Other errors

Abstract:
Hyperspectral remote sensing has become a frontier technology in the remotesensing field in recent decades. Hyperspectral image processing also has been a veryactive research area. Image-spectrum merging is one of the greatest achievements ofhyperspectral imagery, which image by hundreds of continuous spectral bands.Obviously for Hyperspectral sensor, to get more spectral bands means dividing thewhole electromagnetic that scattered within ground instantaneous field view into morefractions, that lead to each fraction¡¯s energy get smaller. But to get higher spacialresolution need electromagnetic energy more concentrated. So to get higher spectralresolution and to get higher spacial resolution have a contradictory relationship. Underpresent conditions, as the manufacture technique level of hyperspectrophotometer islimited, hyperspectral remote sensing¡¯s high spectral resolution due to low spatialresolution correspondingly. One direct consequence from low spatial resolution is theexistence of mixed pixels. Every mixed pixel in hyperspectral imagery is comprised byseveral different constituent substances. Mixed pixels set an obstacle for target detection.So investigating how to solve mixed pixel problem has an important significance.In this dissertation, firstly, hyperspectral sensing characteristics have been studied,and the physical process that how mixed pixels formed is investigated in principle.Further the writer introduces the relation between convex theory and linear mixingmodel(LMM).Through studying mixed pixel¡¯s forming mechanism, we know manyreasons effect this process. Different cases result in different mixing models. In thisdissertation, we only focus on LMM, and explore the contact between LMM andconvex in theory.Secondly, this text research the work flow of hyperspectral unmixing, and modelan optimization problem base on Craig¡¯s criterion, further transform this problem into aconvex optimization problem. As a result we can solve the unmixing problem byconvex theory. In this part, we did experiments both of simulative and real data, andcompared the performance with NFINDR-FCLS(Fully Constrained Least Squares)algorithm, which demonstrate the improvement of convex based algorithm.Finally, we utilize unmixing result for target detection application. The result ofspectral unmixing is only a mathematical description. For most pixel based targetdetection algorithms the result was hard to utilized directly. So we take advantage of the information given by the unmixing and use sub-pixel mapping technique to enhance thespatial resolution of hyperspectral image. Experiments demonstrate after sub-pixelmapping target detection show better results than directly detect without using theunmixing result.
Related Dissertations
Last updated
Sponsored Links
Home |About Us| Contact Us| Feedback| Privacy | copyright | Back to top