Content area
Full Text
Author for correspondence: Birte Boelt, Email: [email protected]
Introduction
Seed quality is a multiple component characterization of seeds, including varietal and analytical purity, germination capacity, vigour, seed health and uniformity. Currently, testing for seed quality relies on physical and chemical as well as visual inspections, which are costly and time consuming. Furthermore, visual inspections are subjective and difficult to reproduce.
Visible features like colour, surface structure, morphology and seed size are determined by machine vision, primarily based on RGB (red, green and blue) cameras. Another optical sensing technology is near-infrared spectroscopy (NIRS) for the determination of chemical composition, both on the seed surface and internally. Hyperspectral imaging combines the two technologies and provides information on both spatial and spectral aspects (Huang et al., 2015). Rahman and Cho (2016) have reviewed a number of non-destructive measurement techniques including machine vision, spectroscopy and hyperspectral imaging for the assessment of quality parameters of agricultural seeds. In particular, they found hyperspectral imaging to be a promising tool for seed quality aspects, as it combines machine vision and spectroscopy into one instrument. Applications of hyperspectral imaging in seed quality assessment are variety identification and classification, quality grading, detection of insect damage and fungi infection, and prediction of chemical composition (Huang et al., 2015). The spectral regions employed cover a broad range of spectra in the visible (380–780 nm) and NIR (780–2500 nm) region (Huang et al., 2015; Rahman and Cho, 2016). However, adoption of hyperspectral imaging by the seed industry requires a decrease of costs without compromising the accuracy of analysis (Dell'Aquila, 2007; Huang et al., 2015; Rahman and Cho, 2016).
Selecting only those spectra providing specific information about the quality aspect in question is one way of reducing costs, yet still maintaining the high quality assessment. Multispectral imaging (MSI) systems include light sources providing a restricted number of wavelengths, which will provide the specific spectra. Examples are reported from the food industry where the spectra from four wavelength bands provide 100% correct classification of damaged mushrooms (Esquerre et al., 2012), and similarly Xing et al. (2010) were able to distinguish sprouted from non-sprouted wheat kernels by the use of four specific wavelength bands.
The Organisation for Economic Co-operation and Development (OECD) defines the standards for...