Computerized image analysis applied to inspection of vetch seeds for varietal identification

Vicia sativa L. is an annual cleaning crop mainly used in rotation with durum wheat, both for seed production and for forage, either as a pure crop or as an undercrop, and is well adapted to the hot-dry climate and poor soils of Southern of Italy. Generally, cultivar identification is done on the basis of distinctive traits, such as shape, size, colour of the testa and ornamentations, identified during the variety registration, following official protocols. For some vetch varieties seed identification by specialized technicians is possible without difficulty, while for other varieties it is very difficult and sometimes impossible to achieve undoubted identification. Therefore, it is important, from both technical and economical points of view, to have a quick, reliable, repeatable, and non destructive method to be able to identify and classify seeds objectively. In order to promote computerised image analysis as a tool to aid visual inspection and replace human judgement in the discrimination of different seeds, this technique was applied to analyse and identify seeds of common vetch varieties. Seed size, shape and colour of nine Italian cultivars of common vetch (Vicia sativa L.) and one of hairy vetch (Vicia villosa Roth.) were measured using digital images acquired by a flatbed scanner and, on the basis of a Linear Discriminant Analysis algorithm, a statistical classifier able to identify the ten cultivars was implemented