利用高光譜資料評估芻料燕麥營養價值

Abstract

燕麥 (Avena sativa. L) 是世界上重要的芻料作物之一，也是臺灣大宗的進口乾草。牧草化學組成反映牧草品質，並仰賴化學分析，其分析時間長且成本高。高光譜影像為近年來用於近似牧草化學成分的工具，須建立符合待測樣本之光譜與化學成分之關係以利成分預測。為以高光譜資料近似國產燕麥乾草化學成分，本研究拍攝燕麥植株的高光譜影像，搭配傳統化學分析，使用偏最小平方回歸法配適燕麥植株之反射光譜與三種產量性狀及八種營養成分之關聯性以建立各性狀相對應之檢量線並評估其應用於芻料燕麥牧草品質預測之可行性。本試驗為期兩年，第一年度選用四個燕麥品系，於不同生育期間收穫十次；第二年度增加八個品系，於各品系達抽穗時收穫。本研究比較不同波段範圍之性狀預測性能，包含全波段 (400 – 1000 nm)、可見光 (400 – 700 nm)、近紅外光 (700 – 1000 nm) 和近似市售多光譜儀器之18個光譜波段。結果顯示，參試品系於抽穗期收穫可獲得相對穩定的生物量及乾草品質。另，不同波段範圍與不同性狀的相關性不一，如近紅外光於可溶性碳水化合物相對重要；且僅使用18個光譜波段於乾草品質的預測相關係數可達0.63 – 0.75。本研究之結果可提供未來芻料燕麥育種即時且精確之性狀評估與試驗所需設備之參考。

Oat (Avena sativa L.) is one of the important forage crops globally and represents a significant portion of imported hay in Taiwan. The chemical composition of forage reflects its quality. It is determined through chemical analysis, which is time-consuming and costly. Recently, hyperspectral imaging has been used to approximate the chemical composition of forage. However, it is necessary to establish a relationship between the spectral reflectance and the chemical composition of the samples to facilitate the prediction of chemical composition. The aim of this study was to use the hyperspectral data to approximate the chemical composition of locally produced oat hay. The hyperspectral data, together with three yield-related traits and eight chemical components of forage oats, were modeled using partial least squares regression to establish the respective calibration curves. The field experiment of this study was conducted during two years. In the first year, four oat lines were grown and harvested at ten dates corresponding to different growth stages. In the second year, eight lines were added and harvested at heading for each line. We compared the predictive performance of different wavelength ranges, including full wavelength (400 – 1000 nm), visible (400 – 700 nm), near-infrared (700 – 1000 nm), and commercially available multispectral wavelength range. The results showed that all the tested lines could achieve relatively stable biomass and hay quality when harvested at the heading stage. Furthermore, different wavelength ranges showed different predictive performance for different traits. For example, near-infrared was relatively important for water soluble carbohydrates, and the predictive ability of using only 18 wavebands could reach 0.63 – 0.75. The results of this study provide guidance for future forage oat phenotyping research and breeding.