关键词: 天然牧草, 无人机, 高光谱, 营养成分, 机器学习

Abstract: Nitrogen (N),phosphorus (P) and potassium (Kalium,K) contents are important indicators of the nutritional value of pasture grasses,and their accurate detection is of great significance for the evaluation of grassland degradation and the development of animal husbandry. In this study,temperate desert steppe and temperate steppe in Inner Mongolia were used as research objects,and hyperspectral unmanned aerial vehicle (UAV) data and ground-based measured data were combined to screen the sensitive bands by using the least absolute value contraction and selection operator (Lasso) regression. Then,Partial least squares regression (PLSR) and Random forest (RF) algorithms were used to construct the estimation models of N,P,and K contents of natural pasture grasses,respectively. The results showed that the first derivative (FD) and the reciprocal of logarithm (Log(1/R)) treatments could help to improve the correlation between sensitive bands and N,P,and K contents;and the Lasso regression method greatly reduced the number of bands. The best inversion models constructed for N and K contents under both grass types were RF models,with R² between 0.76 and 0.90. The P content was optimized by the PLSR model under the temperate desert steppe with an R² of 0.72 and the RF model under the temperate steppe (R²=0.75). The results of this study have important reference value for grassland classification and grading,grazing management and grassland agriculture.

Key words: Natural forage, Unmanned aerial vehicle (UAV), Hyperspectral, Nutritional composition, Machine learning