臺灣博碩士論文加值系統

合成孔徑雷達（synthetic aperture radar, SAR）具有對大氣高穿
透力及全天候的特性﹐目前已成為一相當重要的遙測工具。而全偏極合成
孔徑雷達（polarimetric SAR）更有著可任意合成各種偏極的優點﹐利用
這個特性可獲取更豐富的目標資訊。 合成孔徑雷達有著斑駁
（speckle）雜訊特性﹐它會影響目標資訊的判識﹐因此在利用合成孔徑
雷達資料之前﹐斑駁必須先濾除。濾除speckle的方法已經發展出很多種
﹐但皆不適用於全偏極合成孔徑雷達。李氏濾波器（Lee Filter）
[Lee,1981]在各偏極影像 （spatial domain）上利用統計的方法濾除雜
訊﹐這種方法的缺點是相位的雜訊並沒有 被濾除﹔而Polarimetric
Whitening Filter[Novak and Burl,1990]及Optimal Weighting
Filter[Lee, et al., 1991]則是利用各偏極間的統計相關特性來濾波﹔
MMSE Vector Speckle Filter[Lee, et al.,1991]是利用偏極及空間域（
spatial domain）之間的相關性來濾波﹐以上三種皆會引起各偏極之間
cross-talk的問題。最近Lee[1997]依全偏極合成孔徑雷達的特性發展出
一濾波器稱為Polarimetric SAR Speckle Filter。本研究的目的就是利
用這個濾波器來濾除斑駁以研究其對於地表分類的影響。並以糢糊動 態
學習類神經網路（Fuzzy Dynamic Learning Neural Network）[Tzeng
and Chen,1997]作為分類器。研究的成果顯示經由Polarimetric SAR
Speckle Filter濾除斑駁之後的 確提升了分類的正確率。並由HH/VV的振
幅比及HH-VV的相位差進一步評估其對雜訊濾除 的效果﹐結果發現其的確
能濾除雜訊﹔另外﹐濾波後的HH-VV﹑VV-HV及HV-HH相關係數 幾無改變
﹐這表示這濾波器並沒有破壞各偏極之間的關係﹐也就是保留了全偏極合
成孔 徑雷達的特性。

The polarimetric SAR is a powerful, yet useful data for remote
sensing due to its variable polarizations. The main defect of
SAR data is that it is inherently accompanied with
speckle. Speckle makes SAR data more difficult to interpret so
that speckle must be filtered before it can be better applied in
some applications. There are several speckle filtering schemes,
but none of them is suitable for polarimetric SAR. Until
recently a newly-developed SAR speckle filter, called
Polarimetric SAR Speckle Filter [Lee, et al.,1997] was
proposed. The main purpose of this study is to evaluate the
filtering effects on classification of polarimetric SAR. The
classifier applied in this study is a Fuzzy Dynamic Learning
Neural Network [Tzeng and Chen, 1997], an excellent
classifier for multiband image classification. Observations of
inter-channel (HH, HV, VV) amplitude ratio and phase
difference indicate that the new filter effectively suppresses
the speckle noise, while preserves the polarimetric
information. Classification accuracy was substantially improved.