Journal
JOURNAL OF NEAR INFRARED SPECTROSCOPY
Volume 20, Issue 5, Pages 529-535Publisher
N I R PUBLICATIONS
DOI: 10.1255/jnirs.1018
Keywords
classification; maize hardness; reflectance imaging; partial least squares discriminant analysis; principal components analysis
Categories
Funding
- National Research Foundation (NRF), South Africa [71390]
Ask authors/readers for more resources
Maize is an internationally important food crop that is usually milled before use. Milling yield is strongly influenced by maize hardness which, in turn, is controlled by the relative proportions of vitreous and floury endosperm within a single kernel. Current conventional near infrared (NIR) spectroscopic methods for determining maize hardness require reference data, which is obtained by destroying multiple kernels. In contrast, NIR hyperspectral imaging (NIR-HSI) has shown promise for determining the hardness of individual maize kernels without sample destruction or the need for reference data. This is possible due to the spatial dimension, in addition to the spectral dimension, offered by NIR-HSI. To illustrate this, NIR-HSI was used to characterise regions of germ, vitreous endosperm and floury endosperm from both the endosperm-rich (germ-down; GD) and endosperm-poor (germ-up; GU) surfaces of 155 single maize kernels. The proportions (expressed as % of whole image) of germ, vitreous and floury endosperm were determined from these images after principal component analysis was applied. Subsequent manual kernel dissection confirmed that the ratio of vitreous to floury endosperm was higher in kernels determined to be harder by NIR-HSI. The correlation coefficients between the manually obtained proportions (i.e. dissected) and proportions determined from NIR hyperspectral images (i.e. GU and GD images averaged) were 0.61, 0.59 and 0.11 for vitreous endosperm, floury endosperm and germ, respectively. The incongruence between the two types of determination reflects the surface-bias of reflectance spectroscopy. Irrespectively, NIR-HSI reflectance models could be developed without a reference method and applied to rapidly determine very hard from very soft kernels.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available