Comparative Evaluation of Satellite Remote Sensing Techniques for Land Cover Change Detection

隨著遙感探測技術及相關科學近年來的迅速發展，衛星影像、航空照片等數位影像的取得日漸容易，其在科學研究上之應用亦日益廣泛。在變遷偵測上，遙測影像可提供大區域、即時性的資訊，並且減少現地探勘、航空照片判釋等傳統方法所需的人力和物力。故若欲在短時間內調查大面積地區之災情，具有不受天候及地形限制的數位影像變遷偵測為另一可供選擇的方式。 多光譜遙測影像變遷偵測方法種類繁多，效果及特性並不相同。為評估不同變遷偵測技術間之差異，本研究選擇陽明山國家公園地區為研究範圍，將SPOT多光譜衛星影像以一般克利金法定位後，透過尖峰對齊之NDVI指標灰階值組體圖差方法決定變遷百分比及其對應變遷門檻值，以不同多光譜遙測影像變遷偵測技術進行變遷偵測，將所得結果與研究區域內相近時期可取得之航空照片進行比對，評估各變遷偵測技術掌握實際變遷情形之能力。此外不同時期之衛星影像間因大氣環境之差異而存在著一定程度的輻射差異並影響變遷偵測之結果，故需透過輻射校正以縮減其誤差。研究中在部分變遷偵測前先透過不同之輻射校正技術進行輻射校正以消除影像間的輻射差異，依其結果評估不同輻射校正方法對變遷偵測之影響，以供未來研究時輻射校正技術選擇之參考。With the rapid progress of remote sensing techniques and related sciences, it’s easier to acquire different kinds of digital images like satellite images and air photos, which is widely used for scientific researches. For land cover change detection, remote sensing image could provide data for a vast area instantaneously and thus resulted in reduced time and effort cost than conventional ways of investigation like field survey and image interpretation of air photos. Remote sensing is also an alternative to get the full view of a disaster over extensive areas in a short time avoiding the limitation of weather and landforms. For multi-spectral images change detection, there are numerous techniques with different properties and effectiveness. To assess the difference between those techniques, the Yangmingshan National Park was chosen as the research area. The SPOT (Le System Pour l’Observation de la Terre) satellite images of the research area are carefully georeferenced with ordinary Kriging method. Change percentage from which the correspondent change threshold of each method could be derived was determined by the difference of peak-adjusted NDVI histogram of the images of two dates. The places where changes occurred were found by each change detecting technique and then compared with the air photos available in the same area around the same time period. Then the effectiveness of applied techniques could be compared. Furthermore, inevitable radiometric difference exists between images of different time period because of the necessary difference of atmosphere conditions, and results of the change detection would be affected by the radiometric difference. We could reduce the radiometric difference and increase the accuracy of the change detection by a procedure known as relative radiometric normalization. In this study, different radiometric normalization techniques were applied to make the adjustment before the change detection, and the effectiveness of the different normalization techniques could be compared based on the results of subsequent change detection.目錄 I 圖目錄 III 表目錄 V 中文摘要 VI 英文摘要 VII 第一章 前言 1 第二章 文獻回顧 2 第三章 研究區域與資料 5 3-1研究區域 5 3-2研究資料 7 3-2.1遙測衛星影像 7 3-2.2相片基本圖 7 3-2.3航空照片 7 第四章 影像校正 13 4-1遙測基本原理 13 4-2影像畸變 15 4-3幾何校正 15 4-3.1半變異元模式 15 4-3.2一般克利金空間推估模式 16 4-3.3交叉驗證 21 4-4輻射校正 22 第五章 變遷偵測 28 5-1衛星影像地理定位 28 5-1.1影像對影像控制點選取 28 5-1.2克利金空間推估與交叉驗證 35 5-2多光譜變遷偵測 35 5-2.1變遷百分比及變遷門檻值決定 35 5-2.2多光譜變遷偵測方法 36 5-2.3方法選取與結果驗證 42 第六章 結果與討論 43 6-1一般克利金影像校正 43 6-1.1初期轉換與空間結構分析 43 6-1.2影像校正結果 47 6-2變遷偵測 47 6-2.1輻射校正 47 6-2.2變遷百分比 53 6-2.3偵測結果 58 6-2.4方法評估 58 第七章 結論與建議 73 參考文獻 7