利用水平集方法擷取水下特徵－以東沙環礁及曾母暗沙為例

Abstract

光學遙感衛星影像提供一種水下特徵資訊有用且方便的來源，特別是水深較淺的區域，主要是因為光能夠穿透水層的這個特性，因此可以反射水底訊息給衛星接受器，而光能穿透水的特性會受到波長影響，因此本研究將探討多光譜影中的每個波段能所提供的水下特徵的資訊量多寡，並利用水平集方法進行影像分割擷取水下特徵。 本研究選取了兩個區域，第一個區域為東沙環礁，東沙環礁由東沙島、潟湖以及周圍環礁所組成，潟湖水深範圍約為0公尺至20公尺，而外圍環礁水深則約為0公尺至於3公尺，其研究資料為WorldView-2衛星影像，於2013年10月所拍攝。第二個研究區域為曾母暗沙，研究資料為兩幅Landsat 8衛星影像，水深約為17公尺至25公尺，這兩個研究的區的水皆透徹，不過與第一個研究區域不同的是曾母暗沙為一個完整的水下特徵，而東沙環礁內的水下特徵較多。 東沙環礁的衛星影像經過影像前處理後，選取兩個水深不同的區域對每個波段以及NDWI與主軸轉換影像進行影像分割並分析成果。評估標準為越多的水下特徵被擷取出來，代表能提供水下資訊的能力越好，根據此評估標準，主軸轉換後的第一主成分波段的成果最好，接著是NDWIWV2，代表波段組合比單一多光譜可提供更多的水下資訊。而在曾母暗沙的例子中，將分割成果的區塊邊界與人工數化邊界比較，其中四個波段中C/A與blue波段成果最好，而在波段組合中，NDWI與coastal、red、green以及blue波段進行主軸轉換後，得到三個主成分波段，其中以第一主成分波段成果最好。經與水深比較，曾母暗沙之邊界與30公尺等深線接近。

Optical remote sensing satellite provides a useful and convenient source for identifying underwater features particularly for shallow water area. This capability varies with wavelength. In this study, the capability of finding underwater features with different bands of multispectral image. There are two study sites in this study. The first site is Dongsha atoll, which is composed of Dongsha island, lagoon, and surrounding reefs. The water depth ranges from zero to less than three 3m at outer ring and 20m in the lagoon. The images were acquired with WorldView-2 in October of 2013. The second site is Zengmu shoal, which is completely an underwater feature. The image used is a scene acquired with Landsat 8. The water depth ranges from 17 to 25 meter. The water clarity is high in both areas. The difference of these two study sites is that Dongsha atoll contains more underwater features than Zengmu shoal. For the Dongsha case, two sub areas are selected. Besides the original bands, NDWI and principle component transformation image are analyzed as well. The assessment is made with the number of segments identified. The more segments identified would be taken as providing more information. From the result, the first principle component performs the best, and then is NDWIWV2. It means that band combination provides more information than a single band. For the Zengmu shoal case, the boundary from segmentation is compared with manually digitized. Among the spectral bands, C/A and blue performs the best. And the first component of the Principle Components from Coastal, Blue, Green, Red bands, performs the best. The boundary of Zengmu shoal is found to be close to the -30m contour line.