Estimation of Chemical Composition of Grass in Meadows using Hyperspectral Imaging

Abstract

The chemical composition of grass is useful field information for grassland management. The objective of this study was to develop a field-scale system for estimating the chemical compositions of grass in meadows by using hyperspectral imaging. A hyperspectral imaging sensor was mounted on the roof of a vehicle, and hyperspectral images of a whole meadow field were acquired as the vehicle was driven. Models for estimating seven chemical compositions of grass (crude protein, acid detergent fiber, neutral detergent fiber, calcium, phosphorus, magnesium and potassium) were developed using multiple linear regression analysis (MLR), multi-layered neural network (MLNN) and partial least squares regression analysis (PLSR), and these estimation models were compared and discussed. An EI test to confirm the practical accuracy was conducted, and as a result, EI values ranged from 15.6 to 33.9, and the EI ranks were B or C, except for the MLNN models for Ca and P. Therefore, most of the estimation models were effective in estimating of chemical compositions. In conclusion, this study showed the possibility of field-scale estimation of the chemical compositions of grass by using the hyperspectral imaging system.