建立森林模型對於森林管理及經濟發展極具重要性，而傳統建立模型方式需耗費大量人力、時間及經費，故現以光達資料及高解析影像提升效率。本研究主要目的為融合光達資料及高解析影像偵測單株立木區塊，進而建立三維植生覆蓋模型。本研究主要分為四階段，第一階段為資料前處理，將影像正射化與網格化光達資料做空間套合，第二階段進行樹林區域偵測，首先利用以區塊為單元之影像分割法將區塊分割後，再利用影像知識庫分類來進行辨識分類出樹木區域。第三階段進行單株立木萃取，利用所得到之樹木區域，以影像分水嶺技巧以及局部最大點搜尋法來萃取單株立木。第四階段利用樹高及數值地型模型等資訊進而將模型參數化得到樹模型，並建立出三維植生覆蓋模型。研究中使用兩組測試資料於台灣地區，一組測試資料於芬蘭，台灣地區之測試資料屬性分別為都市區以及果園，地點則分別位於新竹以及台中，芬蘭測區則是由國際航測及遙測學會提供之樣區資料，由研究結果顯示，由自動化萃取單株立木區塊，國內測試區之正確率約90%，高程精度則優於0.6公尺。芬蘭測區自動化萃取單株立木區塊之正確性約為80%，高程精度為1.12公尺。 Three-dimensional forest model is important to ecosystem management. Traditional ground investigation requires vast amount of manpower, resources, costs, and time. The objective of this investigation is the modeling for 3-D forest canopy using LIDAR data and high resolution images. The proposed scheme comprises three major steps: (1) data preprocessing, (2) vegetation detection, and (3) tree crown extraction, and(4) Parameter modeling. The data preprocessing includes spatial registration of LIDAR and high resolution images, derivation of above ground surface from LIDAR data, and generation of spectral index from high resolution images. In the vegetation detection, a region-based segmentation followed by the knowledge-based classification is employed to detect tree regions. In the next step, we perform the tree crown extraction in vegetation regions. We use watershed segmentation and local maximum search to extract tree crowns. In last step, we use tree height, and terrain information to build the model. The validation data include two test sites in Taiwan and one site in Finland. Taiwan data are in Hsin-Chu and Tai-Chung respecting an urban area and orchard place, respectively. The Finland data was released by the International Society for Photogrammetry and Remote sensing (ISPRS), and Euro Spatial Data Research Organisation (EuroSDR) as a sample test site. The experimental results in Taiwan indicate that the accuracy of extracted individual tree is better than 90%. The accuracy of determined tree heights is better than 0.62m. The experimental results in Finland indicate that the accuracy of extracted individual tree is 80%. The accuracy of tree heights reaches 1.12m.
