自動化的從衛星影像圖攫取空間特徵的方法在地理資訊研究中一直是重要課題。傳統上,攫取空間特徵物最原始的方式是人為方式從數位影像攫取特徵物到地理資訊系統。有鑑於此,本研究使用層集多重分類法攫取影像區域邊界,降低人力上的需求,並使用監督式的最佳統計分類法驗證層集多重分類法的可行性,從套疊結果與河道面積比較上得知,層集多重分類法可有效攫取出河道區域並可彌補最佳統計分類法所不足之處。研究區域為南投竹山陳有蘭溪,其衛星影像拍攝年代分別為1999年、2002年及2007年,使用層集多重分類法的分類套疊結果顯示1999至2002年間河道區域為相近的,而2002年至2007年河道區域是有擴大的趨勢;而1999年與2007年河道面積分別為4361.48公頃與6463.74公頃,亦顯示河道區域是呈現擴大的趨勢,因此證明層集多重分類法可有效的攫取區域邊界。
The method of automatically extracting spatial features from satellite images is an important topic within geographic information application research. Traditionally, extracting spatial features manually is the primary method of transforming geographic data from analog to digital formats. Several sophisticated algorithms have been proposed to automatically complete this process. By using gradient methods which compare differences between a pixel and its neighbors, the proposed algorithms can extract edge, and other characteristics, of the given image. For those algorithms that employ gradient methods, the large number of differences between the two regions makes it difficult to automatically extract spatial features and hence some manual input is still required. To overcome this difficulty, we employ a multilayer level set method to isolate regional boundaries such that they are implicitly represented by several nested level lines according to pre-specified level values. This method provides numerical stability and quick convergence. By energy minimization, the initial curves automatically extend closer to the regional boundaries after each iteration. The multilayer level set method is able to not only extract the regional boundaries from the given image but also generate an optimal piecewise continuous image such that each approximation sub-region is homogeneous. In this paper, a SPOT satellite image is used to evaluate the performance of the multilayer algorithms. The spatial features are extracted separately for each different band contained in the multispectral image. They are then combined together to produce the full delineation of extracted features. Using these numerical results, the river and its surrounding areas in the given images can be extracted and shown in different layer values. With careful choice of parameters, the multilayer level set method demonstrates robustness in extracting spatial features from images.