We present a method for multispectral seed phenotyping as a fast and robust tool for managing genebank accessions. A multispectral vision system was used to take images of the seeds of 20 diverse varieties of rice (approximately 30 seeds for each variety). This was followed by extraction of feature information from the images. Multivariate analysis of the feature data was used to classify seed phenotypes according to accession. The proportion of correctly classified rice seeds was 93%. We conclude that the multispectral image analysis could play a role in comparing incoming seeds against existing accessions, identifying different seed types within a sample of seeds and/or in checking whether regenerated seeds match the original seeds.
