以衛星影像區域物件化分類法參數優化法在萬大水庫崩塌地影像之實證

The Optimal Solution in Regional Based Classification Approach-A Case Study on Landslide Image Classification

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

《水保技術》 7卷2期 (2012/04) Pp. 66-73

https://doi.org/10.30040/JSWCT.201204.0001

摘要

在影像分類中，一般是以逐像元（Pixel-based）為主的分類演算法進行影像資訊之萃取，而逐像元分類方法通常會造成椒鹽效應（Salt and Pepper Effect），使分類結果難以應用，因此需要經過繁雜的後續處理工作後，分類結果方能使用。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）獲取萬大水庫附近地表土地覆蓋之情形，以區域物件化判釋（Region Object-oriented Classification；ROC）進行判釋區域，同時也利用不同的光譜及紋理資訊來分析萬大水庫周圍的地區，我們設計了一個以像素的分類透過亂度基礎法來計算，我們針對在敏感地模型中發生與不發生屬性之間的關係進行了研究，希望藉由此研究可以讓山坡地相關管理單位對於坡地災情復舊重建之參考。

關鍵字

崩塌地 ； 影像分類 ； 區域物件化判釋

並列摘要

In the image classification, it generally used pixel-based classification model to extract information of image. The results of by pixel-based algorithm can induce Salt-and-Pepper Effect. Therefore, this study purposed a region-based model of Region Object-oriented Classification (ROC) to extract landslide image information. The surface information from the Wan Da reservoir area is collected and studied. Region Object-oriented Classification (ROC) is used to classify the landslide area. We collected different spectrum with several texture information to analyze the surrounding area of Wan Da reservoir. Entropy based classification is used as a classifier to determine the landslide/non-landslide area. Various parameters of S (similarity) and A (area) are used and then the best combinations are found. In the parallel study, we developed a pixel based classification through the calculation on the entropy for simple comparison. The relations of occurrence vs. non-occurrence of landslide with regards to attributes of land surface are studied. Thus, this could be of help to manage the recover on the landslide area.

並列關鍵字

Landslide ； Image classification Region Object-oriented Classification (ROC)