應用無人飛機航拍影像協助農業勘災－以香蕉災損影像判釋為例

Application of Unmanned Aerial Vehicle (UAV) Photography on Agricultural Post-disaster investigation - A Case Study on Image Interpretation of Banana Plantation Damages by Disasters

周巧盈(Chiao-Ying Chou) ； 巫思揚(Shi-Yang Wu) ； 陳琦玲(Chi-Ling Chen)

《航測及遙測學刊》 23卷2期 (2018/06) Pp. 83-101

https://doi.org/10.6574/JPRS.201806_23(2).0002

摘要

無人飛行載具(UAV)拍攝技術的快速紀錄、高機動性與高空間解析度影像，能提供農損查報所需之空間輔助圖資，縮短人力勘災的時間並加速災後復耕。本研究應用UAV之高解析度影像，透過地理資訊系統之影像非監督分類、影像分割與數值地表模型(DSM)，針對香蕉園風災後之傾倒情形，進行災損影像判釋技術之發展與不同判釋法之評估。結果顯示，UAV航拍之正射影像能清楚觀測香蕉園的災損範圍與相對災損情形，但DSM對園內的災損傾倒不具有效的判釋能力。影像非監督分類之像元式與物件導向式之災損判釋率比對現地核定之災損率，以災損率達20%為基準（透過30處地籍樣區），前者之災損判釋率為86.7%，後者為96.7%，兩者對於香蕉傾倒災損都具有十分高的判釋能力。雖然後者之災損判釋率高，但所需要的人力、技術與時間成本也都相對較高。因此，針對災後即時且大範圍香蕉園的勘災工作時，本研究推薦影像像元式之非監督分類技術，提供災後應變與勘災的空間輔助圖資，進行香蕉園災後大範圍的災損判釋與評估作業。

關鍵字

地理資訊系統 ； 農業災損 ； 物件導向式影像分類 ； 像元式影像分類 ； 數值地 ； 表模型

並列摘要

In order to improve the reconnaissance, rescue processes and rehabilitation, the characteristics of unmanned aerial vehicles (UAV), including quickly recording, mobility, and high-spatial-resolution images, can provide maps on assisting the agricultural post-disaster investigation. In this study, geographic information system, image unsupervised classification, image segmentation, and digital surface model (DSM) are applied with UAV high-spatial-resolution images. The objectives of this study are to develop image interpretation techniques on banana plantation damages by typhoon disasters and to evaluate the abilities of agricultural damage interpretations among different techniques. As results, the damage region and the situation of banana plantations can be delineated by UAV orthophotos. However, DSM cannot discriminate the degrees of wind-thrown in plantations. In addition, comparing the interpretation damage rates of pixel-based and of object-oriented image unsupervised classifications, with ground-based official approves (according to 30 samples of cadastral applications with the baseline of 20% of damage rate), the damage rate of pixel-based image interpretation is 86.7% and the damage rate of object-oriented image interpretation is 96.7%. Both image interpretations can discriminate the banana plantation damages effectively. Although the latter performs better than the former, it demands higher on techniques, labor forces, and time consuming. Therefore, we suggest adopting the pixel-based unsupervised image classification to provide the maps on assisting in post-disaster operation and investigation for wide-region banana plantations.

並列關鍵字

Geographic Information System ； Agricultural Damage ； Object-oriented Image Classification ； Pixel-based Image Classification ； Digital Surface Model (DSM)