應用彩色航照以物件導向影像分析方法進行崩塌地及鄰近地物之分類

Application of Aerial Ortho-Imagery with Object-Based Image Analysis to Classify Landslide and Neighbor Surface Features

鄭傅謙(Fu-Chian Jeng)

指導教授 ： 李鴻源

國立臺灣大學/工學院/土木工程學研究所/碩士(2016年)

https://doi.org/10.6342/NTU201602964

摘要

近年來受到全球暖化的影響，使得颱風、洪水等極端氣候出現的頻率增加且強度增強，造成山區每逢颱風、雨季的侵擾，常有山崩、土石流等災情發生，進而導致人民生命財產的損失。因此如何建立一套完善、具有可操作性的規範，以做為分析環境變遷與災害防治的知識基礎，便是獲得大量高精度資料之後所需面臨的問題。本研究根基於現有知識，建立簡明之綜合應用方法與規範，期望能以明確的流程提升遙測資料應用於災防政策之具體成效。
在遙感探測領域當中，過去已經有許多研究藉由衛星影像進行崩塌地的自動偵測，然而受限於解析度的不足造成精度亦受局限。本研究將以常見的彩色正射航照為主要資料並搭配數值地形模型(DTM)進行影像的自動分類，除了光譜所提供的資訊之外，還結合空間資訊來進行分析，協助區分光譜不易分辨的地物。在影像分類方法上，採用物件式影像分析法將像元區塊化，除了能減少高解析度影像在進行分類時常產生之椒鹽效應的問題，物件式分析也更接近人類視覺上對物體的認知，能夠藉由地物的紋理、形狀以及在空間中的相對位置等特性來進行影像的分類。研究結果顯示，影像判釋整體精確度能達到87%以上，且具有穩定、泛用、快速的特性，藉由半自動化影像分析流程，能夠大大減少人工程序，以增加影像大量處理的能力。

關鍵字

彩色正射航照 ； 物件式 ； 影像判釋 ； 半自動化

並列摘要

Recently affected by global warming, it leads to typhoons, floods and other extreme weather occurring more frequency and seriously. During the typhoon and rainy season, it occurs the landslides, debris flow and other disasters more frequently, which making people lose of life and property. Therefore, how to establish the sound and controllable specification as a basis to analyze environmental change and disaster prevention is the problem required to face after getting a large number of high resolution remote sensing data. In this study, based on present knowledge, it builds a simple application of methods and specification. We expect to enhance the application to disaster prevention policies performance.
In the past, analysis satellite images by using remote sensing technique has been a major method for detecting landslide among the field of remote sensing. Due to the insufficient degree of spatial resolution, the accuracy of landslides detection is restrained. This study will mainly use the common orthorectified aerial photographs, and digital terrain model (DTM) to automatically classify the images. These provide not only the information of the spectrum, but also combine spatial information to analyze and assist to spectrum which not easy to distinguish objects. In the image classification method, applying Object-based Image Analysis (OBIA) is to process aerial photographs. Besides reducing classify high-resolution images arising the problem of salt and pepper effect, the object are closer to human visual perception. It can be texture, shape, relative position in space and other features to classify images. The results showed that the overall accuracy of the image interpretation can be up to over 87%, meanwhile it is stable, generic, fast. The semi-automated image analysis processing can greatly reduce the time of artificial processing and increase the capacity of handling the large number of images.

並列關鍵字

Orthorectified aerial photographs ； Object-based ； Image interpretation ； Semi-automated

參考文獻

13. 董炤巖 (2010) 以物件導向分類法進行SPOT衛星影像之崩塌地萃取。碩士論文，屏東科技大學森林系。

16. 鄭智仁 (2010) 應用空載光達與正射影像以物件導向影像分析法進行自動化崩塌地偵測。碩士論文，國立成功大學衛星資訊暨地球環境研究所。

8. 莊心凱 (2012) 結合地貌主題圖層及物件式影像分析方法應用於山區氾濫原及周邊區域特徵判釋。碩士論文，國立台灣大學土木工程學研究所。

3. 王韻皓、陳連晃 (2011) 農業委員會林務局農林航空測量所業務介紹。地籍測量:中華民國地籍測量學會會刊，30(1)，38-46。

18. 譚克平 (2008) 極端值判斷方法簡介。臺東大學教育學報，19(1)，131-150。