利用光譜解混技術於亞像元變遷偵測與識別:崩塌地變遷應用
現今大多變遷偵測的方法只以單一像素為處理單位,但受限於衛星影像空間解析度,一個像素本身常包含一種以上的成份物質。為了從影像取得更精密的資訊,我們研究了一些光譜解混技術,如獨立成份分析法、非負矩陣分解法、非監督式完全限制最小平方法、和頂點成份分析法,並利用它們偵測亞像元等級的變遷。本篇論文即為一研究特例報告,利用亞像元變遷偵測應用於偵測崩塌地的變遷。我們先利用光譜解混技術將多光譜影像的豐度特徵莘取出來,再合併原有崩塌地地理上的特性(例如坡度這個特徵),進行事後分類比較型的變遷辨識。實驗結果顯示,亞像元變遷偵測超越以往單一像素的變遷偵測,可以提供更多的資訊。 Most of change dete...
Main Authors: | , |
---|---|
Other Authors: | , , |
Format: | Thesis |
Language: | English |
Published: |
2006
|
Subjects: | |
Online Access: | http://ir.lib.ncku.edu.tw/handle/987654321/20952 http://ir.lib.ncku.edu.tw/bitstream/987654321/20952/1/ |
Summary: | 現今大多變遷偵測的方法只以單一像素為處理單位,但受限於衛星影像空間解析度,一個像素本身常包含一種以上的成份物質。為了從影像取得更精密的資訊,我們研究了一些光譜解混技術,如獨立成份分析法、非負矩陣分解法、非監督式完全限制最小平方法、和頂點成份分析法,並利用它們偵測亞像元等級的變遷。本篇論文即為一研究特例報告,利用亞像元變遷偵測應用於偵測崩塌地的變遷。我們先利用光譜解混技術將多光譜影像的豐度特徵莘取出來,再合併原有崩塌地地理上的特性(例如坡度這個特徵),進行事後分類比較型的變遷辨識。實驗結果顯示,亞像元變遷偵測超越以往單一像素的變遷偵測,可以提供更多的資訊。 Most of change detection algorithms for multi-temporal images are performed in unit of pixels Due to the limit of spatial resolution a pixel is in many cases a mixed pixel that consists of more than one ground cover types We reviewed several spectral unmixing techniques such as independent component analysis (ICA) non-negative matrix factorization (NMF) unsupervised fully constrained least squares linear unmixing (UFCLSLU) and vertex component analysis (VCA) We employed the spectral unmixing techniques to explore subpixel information and to detect subpixel-scale changes Furthermore we demonstrated an application of subpixel change detection to detection of landslide expansions The abundance feature extracted from multispectral images by spectral unmixing was incorporated with the slope feature into the process of landslide change identification based on the post-classification comparison procedure Our result shows that the subpixel change detection method can provide more detailed information about landslide changes than pixel-based change detection algorithms 1 Introduction 1 1 1 Background and Motivation 1 1 2 Organization 4 2 Related Work 5 2 1 Linear mixing model 6 2 2 Independent Component Analysis (ICA) 6 2 3 Non-negative Matrix Factorization (NMF): 10 2 4 Unsupervised fully constrained least squares linear unmixing algorithm (UFCLSLU) 12 2 5 Vertex Component Analysis (VCA) 14 2 6 Change detection based on Simple Differencing 16 2 7 Change detection based on Significance and Hypothesis Tests 17 2 8 Change detection based on integrating spectral feature and texture feature 18 2 9 Change Identification by Post-classification Comparison 19 2 10 Adaptive Bayesian Contextual Classification Based on Markov Random Fields 19 3 Proposed Approaches 22 3 1 Subpixel Change Detection 25 3 2 Subpixel Change Identification 26 4 Experiments 28 4 1 Datasets 28 4 2 Simulated data 31 4 3 Real data 35 4 3 1 Comparison of Change Detection Methods 38 4 3 2 Comparison of Change Identification Methods 41 5 Conclusions 49 6 References 50 |
---|