Description
The objective of this trial was to transfer calibration from a laboratory instrument, to be used with two different handheld spectrometer. One hundred twenty samples of dried ground (1mm) corn plant or corn ear were scanned with two handheld and one laboratory spectrometers :1) a diode array (DA) with 256 diodes, 950-1650 every 2nm (AuroraNir, GraiNit s.r.l.); 2) a digital light processing (DLP) also scanning between 950 and 1650 every 2 nm (Texas Instrument, USA); 3) Foss NIRSystem 5000 (LAB) scanning between 1100-1498 every 2nm (Foss, DK). Each instrument had a different sample presentation and scanning mode. For the DA samples were scanned at two spots, 5 spots for the DLP and 24 scans in a rotating ring cup for the LAB. All repeated scan were averaged by sample before any computation. Samples were analysed, crude protein (CP), neutral detergent fiber (NDF) and starch. The data set was divided in 20 samples for validation and 100 for calibration. Within the calibration set a subset of 20 samples were selected to evaluate instrument to instrument differences and calculate transfer functions (Standardization). All the calculations were performed using UCAL 4.0 (Unity Scientific, USA). Standardiztion algorithm tested were, spectral bias, Shenk and Westerhaus (1991) method and Piece Wise Direct Standardization (Bouveresse and Massart, 1996). Predicting the DA validation samples, the SEPc for CP was 0.52 vs. 0.72, for NDF was 2.09 vs. 2.19 and for Starch was 2.74 vs. 2.47, for DA and LAB transferred, respectively. The LAB calibration transferred to DLP had very large biases of predictions for NDF (3.83 %DM) and Starch (-4.25 %DM), but a small bias for CP (-0.136 %DM). Compared to the prediction performances with the original handheld, the error of the LAB transferred prediction were similar for the DA, but were much higher for the DLP.
| Consider for full paper in JNIRS | No, thank you |
|---|
