類神經網路與遙測技術應用於坡地崩塌判釋之研究

Application of Artificial Neural Networks and Remote Sensing Techniques in Image Classification of Landslides

張敬烽(Ching-Feng Chang)

指導教授 ： 陳怡睿

長榮大學/管理學院/土地管理與開發研究所/碩士(2008年)

摘要

近年來由於遙測技術的迅速發展，再加上影像分析之軟硬體功能不斷地提升，此方面技術已廣泛被應用於坡地災害調查、土地利用變遷偵測等領域。在影像分類的技術層面上類神經網路(Artificial Neural Networks)也逐漸地受到重視，同時也可以結合空間屬性資料輔助資訊以增進影像判釋能力。本研究首先將坡地崩塌與非崩塌區域，利用地理資訊系統以人工圈繪方式以面與繪製網格兩圖層為訓練樣區；並運用遙感探測技術結合類神經網路分類方法，進行坡地崩塌災害的監測與判釋，並探討類神經網路應用於衛星影像分類方法之適用性與準確度。
本研究以台南縣為研究範圍，以多光譜SPOT衛星影像，運用遙測軟體ERDAS IMAGES 計算其常態化植生指數(NDVI)之值域，作為因子之一，並運用地理資訊系統之空間分析(Spatial Analyst)功能，以數值地型模型(DTM)計算出訓練樣區之坡度、坡向，再利用圖資(地質圖、斷層圖、水系圖、道路圖)萃取出岩層特性與量測出距斷層、水系、道路等之距離及雨量、高程等資料。本研究將運用「主成份分析法(Principal Component Analysis)」，一方面盡可能考量每一個影響坡地崩塌因素與崩塌間之關係，另一方面亦盡量簡化分析計算過程，且作為類神經網路因子之輸入，藉以提高其判識準確度。本研究將比較類神經網路技術與知識庫分類方式，兩者所得影像分類結果，以做為未來其他崩塌地相關研究之參考。

關鍵字

遙感探測 ； 類神經網路 ； 坡地崩塌 ； 主成份分析

並列摘要

Remote sensing is used extensively in the investigation of slop hazard and land use change. Recently, the performance of software and hardware for image classification has been improved constantly. In this research, the regions with and without landslide in SPOT satellite images were marked by hand in the circumstance of geographical information system (GIS) as training sample. The artificial neural networks incorporated with spatial attribute data from remote sensing were used for image classification. The suitability and accuracy of the proposed method were examined.
The research area was in Tainan county. The range of normalized difference vegetation index was calculated by ERDAS IMAGES and the corresponding values were used as one of input factor for artificial neural networks. The slope and aspect of training region were computed from digital terrain model by spatial analyst of GIS. The distances to fault, river system and road were obtained from general maps. Besides, terrain characteristic and precipitation were also used as input of the networks. To consider all factors affecting landslide, to simplify analysis and to improve the performance of classification, factor analysis of all factors affecting landslide used by literatures are carried out by principal component analysis. Finally, the results were compared with those obtained by applying knowledge-based classification.

並列關鍵字

remote sensing ； artificial neural networks ； landslide ； principal component analysis

參考文獻

[25] 陳繼藩、黃誌勇(2005)，應用類神經網路於SPOT 自然色影像之

與航照資料應用於崩塌地辨識之探討，航測及遙測學刊，第八

[23] 陳莉、魏曉萍、王泰盛(2004)，監督式分類方法於遙測影像判釋

[12] 吳教安(2006)，福爾摩沙衛星二號遙測影像應用於河川砂石資源

[28] 劉守恆(2002)，衛星影像於崩塌地自動分類組合之研究，國立成