利用深度學習神經網路進行衛星影像的崩塌地辨識

Detecting Landslides in Satellite Images Using Deep Learning Neural Networks

陳映融(Ying-Jung Chen) ； 繆紹綱(Shaou-Gang Miaou) ； 徐于軒(Yu-Hsuan Hsu) ； 林映丞(Ying-Cheng Lin)

《中華水土保持學報》 53卷1期 (2022/03) Pp. 25-34

https://doi.org/10.29417/JCSWC.202203_53(1).0003

摘要

本研究提出可不受地形與地表高度限制，而能快速準確地進行疑似崩塌地分析的系統。首先取得同一位置但不同時期的兩張衛星影像並分析其NDVI值的變化，在變化較大之處，以影像處理法偵測出影像中的變異點，最後使用Faster R-CNN深度學習網路判定變異點是否為崩塌地。本系統以農委會水土保持局BigGIS巨量空間系統的崩塌地資料為基準驗證系統的效能，所得精確率高達92.2 %。本系統還可輸出崩塌地的外圍輪廓，方便後續分析與運用。

關鍵字

崩塌地 ； 衛星影像 ； NDVI ； 深度學習神經網路

並列摘要

This study proposes a system for quickly and accurately analyzing suspected landslides without terrain and surface height restrictions. Two satellite images are obtained in the same location but at different times, and the changes in their NDVI (Normalized Difference Vegetation Index) values are analyzed. When large changes occur, image processing methods are employed to detect image territorial variations, and the Faster R-CNN (Region-based Convolutional Neural Network), a deep learning network, is used to determine whether the territorial variation is a landslide. The performance of this system was evaluated using landslide data from the Big Geospatial Information System and the Soil and Water Conservation Bureau, Council of Agriculture; the resulting precision was 92.2 %. The system also outputs the outer contour of the landslide area to facilitate subsequent analysis and application.

並列關鍵字

Landslide ； Satellite Image ； NDVI ； Deep Learning Neural Network

參考文獻

李欣輯、蘇文瑞 (2017)，「山區聚落坡地災害風險評估及植生復育之研究－子計畫：坡地災害之防災效益量化評估 (I)」， 國家災害防救科技中心研究報告。(Lin, H.C., and Su, H.J. (2017). “Benefits Quantitative Evaluation of Slopeland Disaster Prevention and Mitigation (I).” National Science & Technology for Disaster Reduction, Taiwan, ROC. (in Chinese))

吳建忠 (2007)，「衛星影像自動化分類應用於土地利用調查及變遷之研究」，國立成功大學地球科學系專班碩士論文。(Wu, C.C. (2007). Study on Land Utilizes and Changes by Satellite Image Automation Classification. Master Thesis, National Cheng Kung University, Taiwan, ROC. (in Chinese))

吳振發、詹士樑 (2003)，「常態化差異植生指數應用於都市綠地品質管制之探討」，台灣土地研究，6(2)，17-42。(Wu, C.F., and Chan, S.L. (2003). “Normalized Difference Vegetation Index Approach for Monitoring Urban Green Space Quality.” Journal of Taiwan Land Research, Taiwan, ROC. 6(2), 17-42. (in Chinese))

余軒誼 (2020)，「運用機器學習理論建構崩塌地潛勢以臺東縣太麻里溪流域為例」，國立臺北教育大學社會與區域發展學系碩士班。(Yu, H.I., (2020). Establish a Landslide Potential Prediction Model with Machine Learning Method – A Case Study of Taitung Taimali Watershed Master Thesis, National Taipei University of Education, Taiwan, ROC. (in Chinese))

林穎東、張國楨、楊啟見 (2018)，「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」，中華水土保持學報，49(2)，98-109。(Lin, Y.T., Chang, K.C., Yang, C.J., (2018). “Object-based Classification for Detecting Landslides and Vegetation Recovery-A Case at Baolai, Kaohsiung.” Journal of Chinese Soil and Water Conservation, 49(2), 98-109. (in Chinese))