在本研究當中,主要是利用時間序列雷達影像針對台灣西海岸沙洲建立一套高精度的自動化特徵萃取流程。首先對原始影像進行前置處理以校正衛星成像過程中所產生的系統性偏差,處理的項目包含正射化、幾何校正、輻射校正、多觀點處理、斑駁濾波等。再結合多尺度分析概念與相關性分析減少繁瑣的人工參數選擇過程,使偵測出來的初步成果具有一致性。再搭配一系列數學形態學運算使萃取的水線結果具有與真實水線相近似的線段連續性,並且將運算過程當中所可能產生的精度誤差控制在最小的範圍之內,而得到一個品質與精度皆佳的自動化水線萃取成果。 經過比較不同特徵萃取方法後,可以發現本研究方法的結果影像不僅雜訊較少,同時原本水線破碎不平滑的情況也獲得改善。而我們使水線特徵變清晰容易辨識後,接下來便對流程中每個步驟的效果進行檢驗以改良自動化流程架構中不足的地方。驗證過程包含初期-斑駁濾波效果,中期-形態學運算精度偏差量,及後期-各測區水線萃取成效,並整合該區鄰近潮位站所記錄之潮高資料,輔以地真資料(水深地形圖資和GPS點位資料)來對水線萃取成果進行定量化分析,可以發現本研究精度誤差約在1個像元左右。 在本研究的最後利用1996年與2005年間雷達影像偵測台灣西部沿海各沙洲的變遷情況,發現芳苑附近沙洲正漸漸淤積成海埔地,而外傘頂洲跟海豐島洲則產生南移的現象,並且面積正逐年減少當中。 In this study, we proposed a waterline extraction algorithm using a time series of satellite synthetic aperture radar images over the west coast of Taiwan. First, the data preprocessing, including ortho rectification, geometric correction, radiometric correction, multi-looking, speckle filtering etc.., was devised. Then, a general idea of multi-scale, correlation analyze, edge detection, and mathematics morphology operations were applied to automatic delineate the waterlines. Delineated the waterlines which were later calibrated to the same tidal height if necessary. Verifications were conducted by GPS ground truth and ancillary data from bathymetric maps. It is found that the accuracy of the extracted waterline is about 1 pixel. Finally, preliminary observations of waterlines changes in the period of years of 1996 and 2005 were provided. Among them, Fang-Yuan Sandbar pronouncedly recess itself into a reclaimed land, while Wai-san-ting Sandbar and Hai-feng-dao Sandbar, shrinking in size, move southwardly.