以物件導向分類法進行SPOT衛星影像之崩塌地萃取

Extracting Landslide Areas from SPOT Satellite Image by Object-Oriented Classification

10.6346/NPUST.2010.00196

董炤巖

崩塌地 ； 紋理 ； 物件導向 ； 太麻里集水區 ； 莫拉克颱風 ； Landslide ； Texture ； Object-oriented ； Taimali River Watershed ； Typhoon Morakot

屏東科技大學森林系所學位論文

2010年

碩士

陳朝圳

繁體中文

台灣地殼運動活躍，斷層構造所造成的地形分布錯綜複雜，地質破碎且河流短、山坡坡度陡峭，導致地質侵蝕作用劇烈，加上山坡地超限利用日益嚴重，在氣候變遷之極端氣候衝擊下，颱風、豪雨及地震的相互作用，常造成大面積的崩塌，使得土壤嚴重流失、植被減少，對人民的生命財產造成嚴重危害。由於崩塌區位分布遼闊且零散，如何於災害發生後能快速獲得災情資訊，利用遙測技術取得崩塌地位置及進行崩塌地監測，已成為重要議題。本研究之主要目的為探討如何以物件導向分類法，以衛星影像之波譜特徵指標組合，建立快速準確的崩塌地萃取方法，以利於未來崩塌地之監策與管理。研究方法係以莫拉克颱風重災區之太麻里溪集水區崩塌地為研究範圍，利用物件導向分類模式，利用影像的光譜值以邊界偵測法進行影像分割，並以光譜及數位地形(Digital Terrain Model, DTM)資訊，設定每個特徵物件門檻值，以樹狀圖的方式進行階層式之影像區塊分類，以提升崩塌地萃取之準確度。研究結果顯示，邊界偵測法對SPOT衛星DN值影像，可有效地進行均質的影像分割，利用物件導向分類法，以亮度指數配合常態化差異植生指標(NDVI)，進行崩塌地分布範圍的萃取，並以坡度30%將崩塌地分類濾除非植生部分之建物及河道，其分類結果之Kappa值為0.81，總體精度為95%。為提升崩塌地分類準確度，本研究加入GLCM二階角動量之紋理特徵值，濾除河 道與崩塌地之混合像元，使崩塌地分類結果之Kappa值提升為0.88，總體精度為97%。

Steep terrains, geological broken and rapid rivers were major erosion activity which severed the geological of Taiwan. Typhoon, torrential rain and earthquake usually caused landslides happened also damaged people property, land soil loss and decrease vegetation. As the distribution of landslides were vast and scattered, how to quickly use the classified methods of remote sensing for catching the location and boundary of landslides after disaster that has become an important issue. In order to monitor the change of landslide and guide policy of management in future, the purpose of this study attempt to use the spectral index and object-oriented classification method to retrieval landslide distribution and to monitor landslide change. In this study, we used Taimali watershed landslide areas after typhoon Morakot as a study area. The object-oriented classification method consists of two steps: image segmentation (spectrum algorithm detecting edge) and object classification. The classification was based on spectrum feature and Digital Terrain Model (DTM) information, using tree diagram method to hierarchical classified objects. The results showed edge detecting method was efficient in segmentation SPOT image. Combined brightness index and NDVI in object-oriented to extract the distribution of landslides, the results showed Kappa value of classification and overall accuracy were 0.81 and 95% separately. The threshold of slope was 30% promoted distinguishing non-vegetation, buildings and river land use. Subsuming the angular second moment of textures into analysis, the Kappa value and overall accuracy of classification were increased to 0.88 and 97% separately.

1. 鄭傅謙(2016)。應用彩色航照以物件導向影像分析方法進行崩塌地及鄰近地物之分類。國立臺灣大學土木工程學系學位論文。2016。1-103。