Diode Array Near Infrared Spectrometer Calibrations for Composition Analysis of Single Plant Canola (Brassica napus) Seed

摘要： A canola breeder needs an accurate, rapid, non-destructive method for analyzing seeds from a single plant to select the most promising samples for further breeding trials. Near Infrared Spectroscopy (NIRS) is widely used for quantitative analysis of oilseeds in a non-destructive manner. This research was aimed at developing NIRS calibration models for single plant canola seed using a diode array NIRS (950-1650 nm wavelength range), multivariate prediction models, and a mirrored sample cup. Eighteen different NIRS calibration models were developed using 100 samples for each constituent with different pre-processing techniques (mean center, derivatives, variates) and models (PLS, PCR). The relative performance of different calibration models for each constituent was compared using R2, SEP, and ratio performance deviation (RPD) values obtained from the validation set of 30 samples. NIRS models developed using the PLS regression algorithm for moisture content (R2 = 0.97, SEP = 0.32, RPD = 6.13) and oil content (R2 = 0.84, SEP = 0.61, RPD = 4.16) were successful. However, acceptable NIRS models were not obtained for fatty acid and glucosinolates content likely due to limited variability and low levels of the constituent and a narrow wavelength range of the DA-NIR instrument.