透過您的圖書館登入
IP:18.189.170.17
  • 期刊

利用物件式導向進行崩塌地種類判釋、復育追蹤-以高雄市寶來地區為例

Object-based Classification for Detecting Landslides and Vegetation Recovery-A Case at Baolai, Kaohsiung

摘要


本研究設計半自動化崩塌地判釋系統,對高雄寶來地區崩塌地,進行坡地災害潛勢預估。分別使用eCognition軟體和ENVI軟體,以階層式分類方式和支持向量機進行自動化判釋,並輔以專家判釋檢核,判釋崩塌地類型。利用直接相減法,以2009年及2015年影像的常態化差值植生指標(NDVI)運算,得到該崩塌地六年間增減情況。此外,本研究另以SHALSTLB模式,使用NDVI、坡度、岩性、事件的降雨量,自動模擬,取模式適配性最高的前1%筆資料,繪製崩塌地潛勢圖。研究發現使用上述兩種軟體進行物件式導向判釋,整體精度均可以超過90%,Kappa值超過80%,但兩者在河道與崩塌地的判釋上,仍出現部分瑕疵,原因可能是地物邊界混淆。從2009到2015年,崩塌地已縮小50%面積,但是在河道攻擊坡仍呈現不穩定,甚至擴大崩塌的趨勢。SHALSTLB的模擬顯示最有可能發生崩塌地的位置,多集中在舊崩塌地冠部及坡腳處,但是在崩塌地面積的推估上出現低估現象。

並列摘要


This study used object-oriented analysis to classify landslides at Baolai village by using Formosat-2 satellite images. We used multiresolution segmentation to generate the blocks and hierarchical logic to classify five types of features. We then classified the landslides and used univariate image differencing to observe the vegetation recovery after 6 years. We used the SHALSTB model to integrate landslide susceptibility maps. This study used the extreme example of 2009 typhoon Morakot, in which precipitation reached 1991.5 mm in 5 days, and selected a 1% sample with the highest modified success rate to produce the highest landslide susceptible area. Both software programs exhibited high overall accuracy and kappa values. Because of boundary confusion, there were some flaws in calculation. From 2009 to 2015, the landslide area decreased 50%. However, the river bank remains unstable because of the ongoing erosion process. The landslide susceptibility maps indicated that the old landslide area was susceptible to landslides in an extreme event; however, we underestimated the landslide area.

並列關鍵字

Object-oriented classification landslide Baolai Village FS

延伸閱讀