Predicting the quality of ryegrass using hyperspectral imaging

Table 2 Model performance of HSI models for nitrogen (PS + BL) (%) (ryegrass blades (BL) + pseudostem (PS))

HSI, Nitrogen (BL + PS)	R² calibration (N = 127)	RMSE calibration (%)	R² validation (N = 64)	RMSE validation (%)
PLSR (AW)	0.86	0.25	0.70	0.35
PLSR (AW0.99)	0.78	0.31	0.71	0.34
PLSR (CARS)	0.36	0.53	0.62	0.39
PLSR (VIP)	0.63	0.40	0.63	0.38
GPR	0.45	0.49	0.30	0.53
SVM	0.69	0.37	0.60	0.40
RF	0.67	0.38	0.14	0.58
MLR	0.77	0.32	0.67	0.36
SMLR	0.76	0.32	0.62	0.39
LASSO	0.73	0.34	0.65	0.37
RMLR	0.70	0.37	0.51	0.44

Calibration based on 66% data, with validation performed on the remaining 33% of data. Partial Least Squares Regression (PLSR) with latent variable selection based on the Adjusted Wold’s R criterion with thresholds on unity (AW) and 0.99 (AW0.99), Partial Least Squares Regression (PLSR) with wavelength selection according to Competitive Reweighted Adaptive Sampling (CARS) and Variable Importance Projections (VIP), Gaussian Process Regression (GPR), Support Vector Machine (SVM), Random Forest Regression (RF), Multiple Linear Regression (MLR), Stepwise Multiple Regression (SMLR), lasso regularization for linear regression (LASSO) and Robust Multiple Regression (RMLR)

ISSN: 1746-4811