應用支援向量機及紋理分析於九份二山崩塌地變遷之評估

Using Support Vector Machine and Texture Analysis for Landslide Change Assessment in the Chiufanershan Area

林文賜(Wen-Tzu Lin) ； 廖紹安(Shao-An Liao)

《水保技術》 4卷1期 (2009/01) Pp. 1-8

https://doi.org/10.30040/JSWCT.200901.0001

摘要

921集集大地震造成中部山區大規模崩塌，本研究以南投縣國姓鄉九份二山為樣區，利用地震後之SPOT衛星影像分析崩塌地資訊。藉由影像相減法分析崩塌地光譜特性，輔以支援向量機及紋理分析資訊，評估崩塌地之分類精度，作為崩塌地災害治理之參考。分析結果顯示支援向量機結合紋理分析，其Kappa值可從73.2%提升至88.4%，得知紋理分析可降低影像雜訊及增加地表特徵分離度，有效提升萃取崩塌區位精度。九份二山崩塌面積由地震初期(1999/9/27)之210.8公頃，於九年後(2007/5/7)減少為72.6公頃，約65.6%崩塌地復原，歷經九年時間，植生陸續入侵崩塌地，九份二山崩塌區位有逐年復原之趨勢。

關鍵字

崩塌地 ； 支援向量機 ； 紋理分析

並列摘要

Massive landslides, caused by the catastrophic Chi-Chi earthquake on September 21, 1999, occurred at Chiufanershan area in Nantou County. In this study, multi-temporal SPOT satellite images were chosen for landslide change analysis. First, image subtraction method was employed for analyzing landslide spectral characteristics. The landslide sites were extracted using support vector machine and texture analysis. The results can be used as the references of disaster assessment in the landslide area. The analyzed results indicate that support vector machine combined with texture analysis can obviously increase the classified Kappa value, from 73.2% to 88.4%. It indicates that texture information can effectively isolate the surface characteristics among land covers. The analyzed results show landslide areas have decreased from 210.8 ha on September 27, 1999 to 72.6 ha on May 5, 2007, about 65.6% of area restored, indicating that the sites of landslide have been gradually restored over nine years of natural vegetation succession.

並列關鍵字

Landslide ； Support vector machine ； Texture analysis