本研究主要探討如何從高光譜影像中篩選出重要的光譜資訊,並以水稻田為主要判釋對象;水稻田的判斷是以實地探勘方式取得地真資料後,搭配監督式及非監督式學習的分類器進行判釋,分別為線性判別分析及細菌覓食演算法,並擬以上述兩種演算法先對於高光譜影像與多光譜影像中的原始波段之光譜資訊進行影像判釋,接著使用粒子群優化演算法篩選出高光譜中對於水稻田判釋最有利的的數十個波段,再進行上述兩種演算法進行計算,進而設計以下四種研究案例: (a)原始波段搭配線性判別分析 (b)粒子群優化演算法篩選出重要光譜特徵搭配線性判別分析 (c) 原始波段搭配細菌覓食演算法(d)粒子群優化演算法篩選出重要光譜特徵搭配細菌覓食演算法,最後使用誤差矩陣表以及主題圖呈現出分類後之成果比較。
This study focused on how to extract the important from hyperspectral imaging spectral information. The techniques of image classification on paddy fields with supervised and unsupervised learning approaches. In this study, Linear Discriminant Analysis and bacterial foraging optimization for hyperspectral imagery for image classification. The preprocessing of Particle Swarm Optimization is used to extract the important influenced factor, and then design into the following four different case studies: (a) the original band with Linear Discriminant Analysis (b) Particle Swarm Optimization filter out important information with Linear Discriminant Analysis (c) the original band with bacterial foraging optimization (d) Particle Swarm Optimization filter out important information with bacterial foraging optimization, finally using the error matrix and topic maps showing the results of the classification after comparison.