Use of statistical image analysis to discriminate carrot cultivars

Abstract Appearance of a plant is an important element in the certification of crop cultivars — there is usually a requirement that it be consistent and distinct from other cultivars. It can, therefore, also be used in the identification of samples of unknown cultivar. Appearance has traditionally been recorded using a mixture of simple measurements (for features such as length and breadth) and subjective scoring (for features such as colour and curvature). This paper describes the results of a project to study the use of automatic, objective image analysis in the discrimination of carrot cultivars. Within-year matching achieves a success rate of 64% with 15 cultivars, which improves to 85% if groups of cultivars included because of their similarity, are amalgamated. Across-year matching to a set of 47 cultivars results in 29% exact matches and 60% in the top five.