Predicting the quality of ryegrass using hyperspectral imaging

Table 3 Model performance over different wavelength ranges

Variable	HSI data	Wavelength range	# LV PLSR	R² calibration	RMSE calibration	R² validation	RMSE validation	Validation intercept	Validation slope	Validation prediction bias
Sugars (mg/g)	BL + PS	550–1700 nm	12	0.70	28.9	0.58	34.4	10.1 ± 9.6 mg/g	0.93 ± 0.10	− 3.8 mg/g
Sugars (mg/g)	BL + PS	550–900 nm	5	0.18	47.6	0.20	47.7	− 15.3 ± 27.6 mg/g	1.20 ± 0.30	− 2.54 mg/g
Sugars (mg/g)	BL + PS	900–1700 nm	8	0.53	36.0	0.43	40.1	18.8 ± 11.6 mg/g	0.85 ± 0.12	− 4.9 mg/g
Nitrogen (%)	BL + PS	550–1700 nm	20	0.86	0.25	0.70	0.35	0.69 ± 0.22%	0.80 ± 0.067	− 0.05%
Nitrogen (%)	BL + PS	550–900 nm	13	0.71	0.36	0.41	0.48	1.19 ± 0.31%	0.64 ± 0.10	− 0.06%
Nitrogen (%)	BL + PS	900–1700 nm	7	0.62	0.41	0.44	0.47	0.64 ± 0.37%	0.79 ± 0.11	0.04%

HSI data, mean reflectance spectra analysed by PLSR model (Blades (BL) + pseudostems (PS) using Standard Normal Variate (SNV) preprocessing. Models are based on 550–1700 nm (original model based on full spectral range) and restricted wavelength ranges of 550–900 nm and 900–1700 nm

ISSN: 1746-4811