對於國家的社會經濟發展，具有建全與完善維護的道路網絡是重要且不容忽視的國家基礎設施。甘比亞近年正在迅速發展，開拓新的道路與改善既有的道路網絡是主要的建設項目之一。因此，本研究目的為研發有效的測繪機制，利用高解析度的影像資料建置最新的道路網絡圖資，並針對甘比亞最活耀的地區進行實例測試。 相較於傳統以像元運算為基礎的演算法，地理物件導向影像分析法(GEOBIA)能更有效的萃取出道路區域。因為其能藉由影像分割和地表分類兩個主要步驟，自高解析影像中描繪出具代表性的物件輪廓，並更易於整合至向量式的GIS平台。 影像分割是道路萃取不可或缺的處理步驟，本研究採用多尺度分割法(Multiresolution Segmentation)切割影像區塊，而其關鍵的尺度參數(scale parameter) 決定了每項切割物件異質性的最大允許量。使用傳統的試誤法雖能求解此參數，但缺乏學理支持且較主觀，故開發高自動化程度與客觀的尺度參數選定方法即為研究重點之一。因此，本研究採用影像中物件的區域變異量(local variance)自動化的選擇前述參數，此方法也進一步延伸應用在道路的幾何特性估測。道路的幾何特性評估可驗證道路尺度參數的推算是否能有效的偵測出道路輪廓。 透過影像分割後產生的物件，可藉由土地覆蓋的分類法則演算法區分其屬性類別。為了確保分類法則演算法的強健性，以分離度分析 (separability analysis)進行分類法則的篩選與物件分類後的量化評估。 結合前述的兩項主要流程，能成功且高效率地自測試案例影像中偵測出道路區域，並以多樣的量化精度評估驗證了演算法的可行性。在廣泛的評估過程中，評估了重點道路萃取的正確性與準確性。而整個評估的過程主要分為三大類: (1)寬度的正確性 (2)萃取的完整性，以及(3)整體的分類正確性。整體的道路區域萃取平均精度超過90％。;The significance of an adequate and a well-maintained road network infrastructure in the socio-economic development of any country cannot be overstated. The Gambia is rapidly developing, with major emphasis placed on establishing new road networks as well as improving the quality of the existing ones. This study’s objective was to perform an efficient mapping exercise of one of the most vibrant regions of The Gambia to establish an up-to-date road network map using high-resolution images. Geographic object-based image analysis (GEOBIA) was used to extract the roads. It was preferred to traditional pixel-based methods because of its ability to delineate readily usable objects from high-resolution imagery. Additionally, its results can be easily integrated with vector based geographic information system platforms for further analysis. It is conventionally composed of two major processes: Segmentation and Classification. Segmentation plays an integral part in GEOBIA, and this study employed the Multiresolution Segmentation algorithm. This algorithm’s pivotal parameter, the scale parameter—which is responsible for controlling the objects’ maximum allowed heterogeneity—is conventionally selected through trial-and-error procedures that can be tedious and subjective. This study, however, adopted an automated methodology in which the selection of the mentioned parameter is based on the local variance of the image objects. This study also extended the adopted methodology to take into account the roads’ distinct geometric properties to derive the scale parameter that best segments them in the image. The objects that were created by the segmentation algorithm were classified by developing rule sets that assigned the objects to their respective land cover classes. The separability analysis of the object classes with regard to their features was performed using a robust automatic process to ensure an efficient classification rule set development process. Combining the aforementioned two major processes ensured the realization of an efficient and objective road extraction process in the study site. An extensive assessment process was performed to evaluate the accuracy and precision of this study’s highlighted road extraction approach. This assessment process was divided into three major categories: (1) width accuracy, (2) extraction completeness, and (3) overall classification accuracy. The highlighted road extraction approach registered an average accuracy of over 90%.
