本研究依據林木樹冠的空間特徵觀念,設計與發展出一套偵測與描繪立木樹冠的新技術,名為SMICID演算法。它結合高解像力衛星影像的空間資訊與多光譜資訊,濾除非樹木的草類與人工鋪面等地類的影響,可以正確且自動化地繪製出影像內立木樹冠的位置圖,其製圖正確率約為80%。SMICID演算法的主要流程包括決定影像中所有可能樹冠中心位置的像元、決定局部最小亮度像元與粹取樹冠可能邊界的谷線網、以及單株立木樹冠像元之描繪與再評估等,利用NDVI光譜指標資料的變異係數與全色態影像資料的半變異圖可以非常有效地確認立木樹冠種子像元;在此一技術規範下,可以選擇較小的移動視窗決定可能的樹冠種子,再由試驗決定的決策係數與樹冠種子像元資料值描繪出立木樹冠邊界。研究結果顯示,SMICID演算法比VFA及TIDA演算法更適合應用於開闊地立木樹冠之自動化偵測與描繪。
This paper developed a new algorithm SMICID that was designed according to the geospatial characters of tree crowns to automatically identify and delineate the individual tree crowns (ITCs) in the satellite image. SMICID algorithm integrates multispectral and textural information of high resolution images to filter out the influences of the spectra of non-vegetative land covers and grass and can map the ITCs efficiently in an open area. An 80% accuracy of the mapped ITCs could be achieved by the SMICID technique using the spectral information of NDVI image and semivariogram texture of panchromatic image. Procedures for automatic mapping the ITCs map include locating the possible tree crown seeds, find out local minima and valley network extraction, and region growing for ITC delineation and refinement. Using SMICID technique, a small window filtering can locate all possible pixels representing tree seeds, and then a suitable value of decision factor could be derived by interactively experiments to delineate the boundaries of each tree crown. Results demonstrated that the SMICID algorithm is much suitable and efficient in automatically identifying and delineating tree individuals in an open area using remote sensing images then VFA and TIDA algorithms.