傳統應用變異向量分析法(CVA：Change Vector Analysis)分析影像之變遷，基本上變遷向量門檻值的指定往往必須使用試誤法或經驗法則決定，導致變遷偵測的結果不夠客觀，因此本研究採用改良式變異向量分析法，自動化地指定差異影像之門檻值，以改善傳統變異向量分析法指定差異影像門檻值不客觀之問題。改良式變異向量分析法進行影像變遷分析主要有兩個步驟:(1)在偵測影像變遷像元位置方面，利用變遷先驗樣區輔助指定變遷向量之最佳門檻，且採用逐步搜尋法以及三次式極值法自動化地找出變遷向量之最佳門檻， (2)在判斷變遷像元型態方面，於前、後時期影像上圈選地物訓練區，以得到變異向量方向餘弦樣本，然後利用變異向量方向餘弦樣本決定每個變遷像元之變遷型態。由一組模擬影像以及兩組真實衛星影像的測試結果，可以得知:使用改良式變異向量分析法進行衛星影像變遷偵測的精確度在偵測變遷像元之精確度可達85%以上，判斷變遷像元型態之精確度也在82%以上，而且操作的過程也相當自動化，因此在土地變遷的應用上，具有很高的使用價值。 Traditionally, when Change-vector analysis (CVA) was applied to detect land cover change, the threshold of change magnitude was always determined according to empirical strategies, or from manual trial-and-error procedure. It made the result of change detection subjective. In this study, using improve Change-vector analysis to determine the threshold of the difference image automatically. The method consist of two stages, (1) prior training set, which aims at helping to determine the threshold of change magnitude, and using Pace Search method and third power polynomial fitting method find the threshold of change magnitude quickly, (2) getting the training areas from the first image and the second image severally to build up the look-up table of direction cosine of change vectors samples, then using the look-up table help to assign “from-to” type of change pixels. In this study , the improve Change-vector analysis was applied to the detection of one simulation image and two SPOT4 satellite images, overall accuracy of “change/no-change” detection was about 85%, and overall accuracy of “from-to” types of change detection is was 82%.The experimental results indicate that the improve CVA has good potential in land cover change detection.
