台灣位在地震活躍區，東臨環太平洋火山帶，西臨菲律賓海板塊，其以每年平均82mm朝西北碰撞歐亞板塊，因此地震頻繁，且台灣每年夏秋兩季會有數個颱風經過，然而伴隨颱風所帶來的大雨以及地震都可能會造成嚴重的土石流或是山崩等自然災害，因此，土石流的變遷偵測對於災後重建或是災害評估就顯得非常有用。 變遷偵測是遙測科技中一項應用相當廣泛的技術，除了應用在變遷偵測之外，都市發展或是農業管理也常使用，通常我們可以比較兩張不同時期相同位置的多光譜影像就能得知前後的差異，然而不同時期拍攝的影像，他們的大氣條件並不完全相同，大氣條件可能包括太陽照度、大氣輻射和大氣灰霾等等，而這些差異可能會使相同位置影像中相同的物質卻有不同的光譜，造成後續影像處理增加誤差，因此在進行變遷偵測之前須要對影像作輻射校正，本文中我們提出利用多維的直方圖均值法對影像作前處理，目的在於修正多光譜影像中不同的大氣條件，讓不同時期相同位置的多光譜影像中相同的物質能有近似的光譜。 研究中採用SPOT的多光譜影像進行前處理的步驟，接著再利用非監督式分類法對影像分類，比較修正前後的影像是否可以成功降低分類的錯誤率，得到更精確的變遷結果。 Taiwan is located at Circum-Pacific seismic zone therefore there are a lot of earthquakes in this region. Besides, in this subtropical region, there are usually several typhoons pass through each year. These two natural phenomena may cause serious landslides in the mountainous regions. For landslides hazard assessment, change detection with remote sensing images is an efficient and effective approach. Change detection is one of the most important applications of remote sensing technique, and it provides useful information for various applications, including disaster monitoring, urban development and agriculture management. Compare two images collected at different time from same located, the ground surface change can be detected. However, the difference in spectrum may not solely result from the changes on the ground. The spectrum of the same material in two remote sensing images may not be the same due to the different condition of solar illumination and atmosphere condition while the images were obtained. Therefore, radiometric calibration is required before applying the change detection algorithm and comparing the spectrum. In this study, we propose a multi-dimensional histogram equalization algorithm as a pre-process step for relative calibration. It modifies multispectral images collected under different atmospheric conditions to have similar spectrum for the same land cover. A set of SPOT images is adopted for experiments and results show the proposed method can reduce the misclassification rate.
