道路是各種車輛與行人通行的主要基礎設施之一，研究道路圖資也是 遙測領域所著重的項目之一。本研究利用高解析衛星影像，萃取道路資訊， 並建立初步的道路路網資訊。 本研究開發出一套系統化的流程，可從高解析衛星影像中萃取道路資 訊並重建道路路網。此程序主要可分為兩個階段:第一階段利用多光譜影像 之光譜資訊與灰階共生矩陣(GLCM)演算法解算的紋理資訊，並以支持向量 機分類演算法，進行影像分類萃取出道路像元。第二階段則利用雷登轉換進 行線性特徵萃取，並進行道路像元的線追蹤，進而重建道路路網。 本研究以 Pleiades 高解析度衛星影像為主要資料，並以台北市部份地區 為研究區域。結合前述的兩項主要流程，能有效的偵測測試區域中影像的道 路區域，並量化精度與評估演算法的可靠性。在驗證演算法的成果過程中， 評估了目標道路萃取的完整性與中心線位置準確性。而整個過程又分為三 大步驟:(1)整體的分類正確性 (2)萃取的完整性，以及(3)道路中心線精度。 實驗結果顯示，整體的分類正確性達到 93%，道路路網完整性達到 86%。;Automatic road extraction from remote-sensing imagery plays an important role in many applications. In general, road extraction from remote-sensing imagery can be considered as a classification process in which pixels are divided into road classes and others. This study develops a systematic procedure using high-resolution satellite imagery to extract and reconstruct road network in urban areas with a road width greater than 8 meters. The developed procedure for extracting roads and reconstructing the network can be divided into three steps. The first step; obtaining texture parameters, including contrast, entropy and homogeneity of each pixel using gray level co- occurrence matrix (GLCM). Afterwards, the support vector machine (SVM) is used as a classifier to identify road pixels. Finally, linear feature detection of the road network with radon transform is performed. This study uses high resolution Pleiades satellite imagery which covers an area of 8 km2 of Taipei city, Taiwan. Experiment results show that the proposed procedure and method can achieve an overall accuracy of 93.11% with a kappa coefficient of 89.20% and completeness over 85%. The results indicate the proposed method is efficient to extract road network from high resolution satellite images.
