Nondestructive and rapid determination of lignocellulose components of biofuel pellet using online hyperspectral imaging system

Table 2 Prediction results of the pre-processing models constructed by partial least-squares regression (PLSR) and principal component multiple linear regression (PC-MLR) for lignocellulose components of biomass pellets

Indices	Model type	Pre-processing	Par^a	Calibration set		Prediction set
Indices	Model type	Pre-processing	Par^a	\(R_{\text{p}}^{2}\)	RMSE_C (%)	\(R_{\text{p}}^{2}\)	RMSE_P (%)
Cellulose	PC-MLR	Raw	29	0.93	2.21	0.91	2.49
		SNV	21	0.86	3.19	0.87	2.95
		2nd	45	0.93	2.26	0.79	3.82
		MSC	21	0.84	3.39	0.85	3.27
	PLSR	Raw	10	0.91	2.64	0.91	2.51
		SNV	11	0.84	3.36	0.83	3.39
		2nd	2	0.61	5.23	0.61	5.23
		MSC	10	0.83	3.56	0.81	3.63
Hemicellulose	PC-MLR	Raw	21	0.83	1.54	0.79	1.68
		SNV	16	0.81	1.61	0.83	1.53
		2nd	45	0.87	1.36	0.78	1.74
		MSC	18	0.80	1.66	0.78	1.73
	PLSR	Raw	12	0.82	1.54	0.80	1.86
		SNV	10	0.81	1.60	0.82	1.58
		2nd	9	0.83	1.56	0.76	1.83
		MSC	7	0.72	1.95	0.59	2.67
Lignin	PC-MLR	Raw	38	0.88	1.19	0.75	1.75
		SNV	21	0.71	1.90	0.60	2.23
		2nd	32	0.85	1.30	0.74	1.78
		MSC	13	0.61	2.23	0.52	2.44
	PLSR	Raw	13	0.86	1.31	0.74	1.79
		SNV	8	0.61	2.20	0.46	2.58
		2nd	16	0.82	1.47	0.71	1.87
		MSC	6	0.58	2.28	0.45	2.61

^a Model parameters indicate the optimal number of latent variables for establishing the PLSR calibration model and optimal number of principal components for PC-MLR; \(R_{\text{c}}^{2}\) and \(R_{\text{p}}^{2}\), coefficients of determination for calibration and prediction sets, respectively; RMSE_C and RMSE_P, root mean square errors of calibration and prediction sets, respectively; SNV, standard normal variate; 2nd, second derivative; MSC, multiplicative scatter correction

ISSN: 2731-3654