應用高光譜影像探討水稻產量與生育性狀空間變異之研究

Abstract

高光譜相較於多光譜影像具有波長小且連續之特性，可將空間上的地表特徵光譜變化鉅細靡遺的展示，有利於地表分類與調查研究。本研究利用所拍攝農試所農場的高光譜影像，探討各波段與水稻產量和重要生育性狀之關係，篩選出建立模式的最佳鑑別波段，並據以研究田間水稻產量和生育性狀的空間分佈與變異原因，以供田間管理改良之參考。研究結果顯示，多光譜影像僅適用於進行產量分佈推估，然而利用高光譜影像不僅產量推估較應用多光譜影像更為準確，且能推估如株高、?重、穗數、穎花數、稔實率和千粒重等重要之生育性狀的空間分佈，有利於田區管理工作和改良措施之運用。 Hyperspectral imaging takes advantage of hundreds of contiguous spectral channels to uncover materials that usually cannot be resolved by multispectral sensors. A hyperspectral image of experimental farm of Agricultural Research Institute was used in this study. The goals were to identify the bands which can differentiate important growth characteristics, such as plant height, straw weight, and yield, and to study spatial variation of those growth characteristics within the fields accordingly. Results indicated that multispectral images were only suitable for yield estimation. However, hyperspectral images could provide not only better yield estimation than multispectral images, but also the estimation of many important growth characteristics, such as plant height, straw weight, effective tiller number, number of panicles per till, fertility rate, and grain weight. It can be of great assistance to farm management and improvement measures to take