| 摘要: | 使用特徵進行衛星影像目標偵測時，經常會受到雜訊因子的干擾而影響目標判釋的結果。然雜訊當中並非全屬無使用價值的資訊，當目標被雜訊訊號所干擾影響時，其實目標訊號亦隱藏於雜訊當中，為有效萃取相關目標資訊，並完成判釋影像中的目標，本篇論文中，我們提出了新的雜訊因子去除演算法，將雜訊周遭的像素資訊均納入特徵值與共變異數的計算，能夠精準的描述出雜訊訊號並予以消除，以降低影響並成功運用於電子光學衛星影像。相關演算法經再推廣於超多頻譜遙測影像亦驗證非常有效，可偵測出特徵及目標物，對國防軍事、環境監測與國土規劃和民生用途均具實際應用價值。
There are a lot of noise factors which will influence the target detection of satellite imagery as utilizing feature extraction. However, those noises are not all categorized to be useless information. As the targets are interfered by noises, those target information are also hidden inside the noise signals. In order to extract the target information and achieve the completely interpretation of targets, there is a novel noise effect removing approach has been proposed in this paper. By using the noise effect removal algorithm, those neighboring pixels around the noises are manipulated with the calculation of eigenvalue and covariance. Through the manipulation, the noise signal could be exactly described and eliminated to decrease its impacting effect. The proposed algorithm not only has been testified on the EO-1 imagery, but also testified on some hyperspectral imagery. Experimental results demonstrate the feasibility and validity of our proposed method in detecting those feature and targets. It could be applied to those areas such as national defense, environment surveillance, land planning, commercial application, and etc. |
