台灣的森林由於海拔以及氣候的關係分布較廣,如果要利用人工管理必須要耗費極大的人力,因此遙測影像被廣泛地應用在此一用途,然而人工影像判讀不僅僅需要大量人力也依賴判讀者的經驗。 近幾年來多層次形態學動態輪廓演算法(MMAC)被提出來用以自動偵測樹木與輪廓描繪,此一演算法可以有效地解決高山地區的樹木輪廓辨識率,不過卻也面臨運算量龐大的問題,使其難以運用於大面積的遙測影像。 近年來平行運算架構逐漸趨於成熟,被視為用以解決龐大運算量的有效方案,然而目前的平行運算架構,依然受限於資料與運算相依性的問題。而多層次形態學動態輪廓演算法由於具有高度的運算相依性,因此難以利用平行運算架構來實踐,並解決運算量龐大的問題。 本論文將修改多層次形態學動態輪廓演算法,透過影像編碼的概念,減少其運算過程中的相依性,使一演算法得以被實踐於平行運算的平台上,並採用統一計算架構為範例,進一步驗證本論文的概念與平行運算的效能。
Forests in Taiwan distribute vertically along the central region and can be categorized into broadleaved, mixed, and conifer forests. Terrain features make manual inspection of forests nearly impossible. By utilizing remote sensing data, the amount of field sampling could be significantly reduced. However, the visual interpretation is labor-intensive and heavily dependent on the interpreter’s experience. An automatic algorithm, called multi-level morphological active contour algorithm (MMAC) has been proposed to address these issues in 2011. The MMAC could effectively increase recognition rate of individual tree in mountainous areas, which is the common case in Taiwan. However, the design of algorithm comes with huge computational complexity for delineation of tree crowns, which prevents it from being implemented practically in medium- or large-scale remote sensing data. The infrastructure of parallel computing provides a solution for many algorithms with huge computational complexity. Unfortunately, its implementation was normally restricted by dependency of data and operations. Due to high operational dependency of MMAC, it is very difficult to be implemented in parallel processing to reduce its computational complexity. This thesis will reduce the operational dependency of the MMAC algorithm by image coding method. As a result, it would realize the parallel processing of the MMAC algorithm. CUDA will be exploited as an example of parallel processing platform to demonstrate the proposed algorithm.
為了持續優化網站功能與使用者體驗,本網站將Cookies分析技術用於網站營運、分析和個人化服務之目的。
若您繼續瀏覽本網站,即表示您同意本網站使用Cookies。