使用偏極化合成孔徑雷達資料萃取臺灣建成區

Abstract

都市人口的急遽增加導致都市的面積快速成長，監測準確的建成區的範圍對於都市規劃等都市分析至關重要。傳統的監測方法，如實地勘查紀錄或人工數化等，所耗費的人力和時間成本高；相較之下，遙感探測(Remote sensing)可以快速獲取大範圍的地表資訊，有助於提升萃取建成區的效率。合成孔徑雷達（Synthetic Aperture Radar，SAR）衛星能夠不受天氣條件因素下進行日夜觀測，而且對於地表的人造結構物也具有高靈敏度。偏極化合成孔徑雷達(Polarimetric SAR，PolSAR)透過記錄電磁波照射地表所產生的不同偏極訊號，與傳統單偏極化的SAR資料相比，可以獲取更多有價值的地表偏極特徵資訊。在都市分析研究中，由於建築物擁有獨特偏極化特徵資訊和散射機制，因此使用PolSAR資料可以有效的萃取建成區。然而，回顧統整過去十年的有關使用遙感探測於都市分析研究相關文獻發現，雖然SAR影像的使用比例較高，但是對於PolSAR資料的應用卻相對較少，而且研究方法多依賴人為決策且具有一定的局限性。 本研究使用日本宇宙航空研究開發機構 (Japan Aerospace Exploration Agency，JAXA)所規劃的ALOS-2/PALSAR-2的L波段SAR衛星的全偏極SAR影像並產製PolSAR資料萃取臺灣北部的建成區，並建立自動化的萃取流程以減少人為決策，並增加方法的泛用性。此外，本研究也將比較使用不同偏極數量、不同偏極化解構法和SAR的不同影像資訊對於萃取成果的影響。 成果顯示使用全偏極SAR影像的偏極特徵資訊在萃取建成區上比起使用雙偏極SAR影像有較佳的成果；且結合模型解構法以及特徵向量和特徵值解構法的偏極特徵資訊與僅使用單一解構法相比，能夠提高成果的準確度，而本研究結合Yamaguchi解構法以及Cloude和Pottier解構法的偏極特徵資訊萃取建成區，成果的準確率為0.88，F-score為0. 87。 本研究的貢獻在於提供可自動化的流程，提高萃取建成區的效率和方法的泛用性，同時，本研究詳細比較了不同條件的PolSAR資料對於萃取成果的影響，並分析了SAR影像與光學影像的萃取成果差異。本研究成果分析可以提供給未來研究者使用PolSAR資料進行都市規劃和分析等領域的應用。

Rapid urban population growth has led to the speedy expansion of built-up areas. Accurate monitoring of built-up areas is vital for the urban environment and urban planning. Compared to traditional monitoring methods such as field survey, remote sensing provide the advantage of quickly acquiring surface information over large areas, thus improving the efficiency of extracting built-up areas. Synthetic Aperture Radar (SAR) satellites can perform day and night observation and have high sensitivity to artificial structures. Polarimetric SAR (PolSAR) can record the different polarized signals reflected from the surface, providing more valuable polarimetric information compared to single-polarization SAR data. Due to the unique polarimetric characteristics and scattering mechanisms of buildings, PolSAR data can effectively extract built-up areas. However, a review of the literature on remote sensing in urban analysis over the past decade reveals that while SAR imagery is frequently used, the application of PolSAR data is relatively limited, often relying on manual decision-making and having certain limitations. In this study, we utilized the fully polarimetric SAR imagery from the ALOS-2/PALSAR-2 L-band SAR satellite to extract built-up areas in northern Taiwan using PolSAR data. We developed an automated extraction workflow to reduce human decision-making and increase the generalizability of the method. Additionally, we compared the impact of different polarization quantities, polarization decomposition methods, and other SAR image information on the extraction results. The results showed that using the polarimetric characteristics information from fully polarimetric SAR imagery had better results in extracting built-up areas compared to using dual-polarization SAR imagery. Furthermore, combining model-based decomposition and eigenvector & eigenvalue decomposition improved the accuracy of the results. In this study, we use the PolSAR data from Yamaguchi decomposition, and Cloude and Pottier decomposition achieved an accuracy of 0.88 and an F1-score of 0.87 in extracting built-up areas. This study provides an automated workflow that enhances the efficiency and generalizability of built-up area extraction. We also conducted a detailed comparison of results using different conditions of PolSAR data and analyzed the differences in extraction results between SAR imagery and optical imagery. These research findings serve as a comprehensive reference for future researchers seeking to utilize PolSAR data in urban planning and analysis domains.