以物件為基礎之無人機影像變遷偵測方法：以都市區土地利用監測為例

Abstract

影像變遷偵測為航遙測領域中用以分析不同時期土地覆蓋或土地利用變化的影像處理方法。隨著航遙測技術的進步，獲得影像的方式越來越多元化，影像解析度越來越高，而變遷偵測的相關方法也隨著不同的應用需求不斷地在演進中。傳統變遷偵測常採用像元式的分析方法，但當影像解析度提升時，以像元為基礎之變遷偵測方法容易產生椒鹽雜訊的問題，且不容易用於土地利用的變遷情況；近年來許多研究開始利用以物件為基礎之影像分析法(OBIA)進行變遷偵測，一般稱之為以物件為基礎之變遷偵測(OBCD)。OBCD除了可以避免椒鹽現象之外，若 進一步引入人為定義的變遷規則將可以用來進行土地利用的變化分析。 都市變遷為影響都市發展的一項重要因素，如何快速了解都市變遷成為都市計畫中的一項重要議題，而最為快速且有效的方法即為影像變遷偵測。本研究利用無人機航拍獲取都市區之高解析度真實正射影像及數值地表高程資料，探討並比較目前常見之影像變遷偵測方法，同時提出一套完整的物件變遷偵測流程，分析可用來進行變遷偵測的多維物件特徵，並引入人為定義的變遷規則以偵測都市變遷，如建物新建及拆除、建物增建、及其他土地利用變化等。此流程執行快速，除了可解決椒鹽雜訊問題外，也避免了大多數將偽變遷誤判為變遷之錯誤。在小區域實驗中，相較於郊區，本文之方法於都市地區可獲得較佳的成果，總體準確度約為98%，Kappa約為0.8；而應用於場景複雜多變的大範圍影像時，亦可獲得98.69%之總體準確度，Kappa值為0.8373，錯誤來源多為將偽變遷誤判為變遷，部份為影像品質不佳導致。

Image change detection is an image processing method used in remote sensing to analyze land cover or land use change in different periods. With the advancement of remote sensing, the image resolution is getting higher and higher, and the change detection methods are constantly evolving for huge application requirements. Traditionally, change detection often uses pixel-based analysis methods, but when the image resolution is improved, the pixel-based change detection (PBCD) method is prone to the problem of salt-and-pepper noise and is not easy to be used for land use change.In recent years, many studies have begun to use object-based image analysis (OBIA) method for change detection, commonly referred to as object-based change detection (OBCD). In addition to avoiding the problem of salt and pepper noise, OBCD can be used to conduct land use change analysis if further introduced artificially defined change rules. Urban monitoring and land use change detection are important issues for urban development and planning. The fastest and most effective method for understanding urban change is image change detection. This study uses UAV aerial photography to obtain high-resolution real orthoimages and Digital Elevation Model (DSM) data of the urban area, then explores and compares the current common image change detection methods. Proposed a complete object base change detection process, which can be used for the analysis of land use change. Used multi-dimensional object features of transition detection and introduce artificially defined rules to detect urban changes, such as new construction, demolition buildings, and other land use changes. In addition to solving the problem of salt and pepper noise, this method is efficient, and it also avoids most mistakes that misjudge the pseudo-change as a change. In the small-area experiment, it showed that urban areas can obtain better results than suburbs, the overall accuracy is about 98% and kappa is about 0.8. When applied to large-scale images with complex and varied scenes, the overall accuracy is 98.69% and the Kappa is 0.8373. The mostly source of errors was falsely judged the pseudo-change objects as change objects, and some errors caused by the poor image quality.