以迭代式像元基礎的亂數基礎分類法在萬大水庫衛星影像崩塌地的判釋研究

The Study of Iterative Entropy-based Classification by Remote Sensing Data: A Case Study of Wan Da Reservoir

10.29417/JCSWC.201303_44(1).0007

萬絢(Shiuan Wan)；雷祖強(Tsu-Chiang Lei)；王文宜(Wen-Yi Wang)

影像分類 ； 亂度基礎分類法 ； 崩塌地 ； Landslide ； Image Classification ； Entropy-based Classification (EBC)

中華水土保持學報

44卷1期（2013 / 03 / 01）

78 - 86

繁體中文

傳統上，大範圍的崩塌地災害多以人工調查，會消耗大量的人力、物力與時間，近年來透過遙測影像資料可以快速判別出崩塌地區。因此本研究利用空間資訊技術（地理資訊系統Geographic Information System；GIS與遙感探測Remote Sensing；RS）獲取地表土地覆蓋之情形。亂數基礎分類法（Entropy-Based Classification, EBC） 是一種典型的分類方法，他的優勢在於透過這個分類器，可以明確找出決策屬性和條件屬性的關係，並於條件屬性建立關鍵的門檻值。然而他的缺點在於若選取的樣本無法具代表性時，分類誤差會大幅上升。因此，我們蒐集了不同的光譜及植生相關的紋理資訊來分析萬大水庫周圍的地區，在研究中，以迭代式交叉驗證法選取了最佳訓練樣本的資料集，並建立以像素的分類（Pixel-Based Classification ; PBC）的分類法則，此法則利用了亂度基礎法來計算，我們針對在敏感地模型中崩塌地的發生與不發生屬性之間的關係進行了研究，以提供崩塌地的影像知識，合理的建構相關的判釋規則。

Traditionally, the in-situ investigation of landslides requires a large amount of manpower and is also very time-consuming. Presently, by applying spatial information (Geographic Information System; GIS and Remote Sensing; RS), we can obtain landform and land cover information systematically. Entropy-Based Classification is a typical classifier which renders relations among system inputs (condition attributes) and outputs (decision attributes). Also, the thresholds of inputs are attained. However, the accuracy of this classifier is determined by the selected data. Accordingly, through remote sensing of image data, we can quickly determine a landslide area by using an effective classifier. This study purposed Pixel-Based Classification (PBC) to extract landslide image information through GIS techniques. More specifically, the image data are used to present the current situation with different spectral and texture information to analyze the area surrounding the Wan-Da reservoir. The pixels with texture information are employed to compute the Entropy-based Classification. In this way, we expect to correlate the image attributes to the occurrence/non-occurrence of the image data. The sensitivity area is analyzed and the relations among occurrence and non-occurrence are studied. The knowledge rules of images on landslides are rationally constructed.

1. 洪藝家(2016)。降雨誘發之坡地崩塌災損評估。長榮大學土地管理與開發學系（所）學位論文。2016。1-102。