Automatic calibration and classification of seeds using digital image analysis

In this study, an automatic image analysis system is presented to measure a nd classify seeds. Seed samples are applied on a rotating glass-plate, phot ographed with a CCD-camera and the images are analyzed with a computer. In a first step, the external contour of the seeds is extracted and calculated using shape-parameters and Fourier descriptors. The seed classification is based on a data-set of representative seeds of all species analyzed. Unkno wn seeds are identified as the most similar species in the data-set. After classification, seeds are removed from the glass-plate by suction. Approxim ately 60 seeds can be classified within one minute. Samples of each 100 seeds of the crops winter-wheat (Triticum aestivum L.), winter-barley (Hordeum vulgare L.), winter-rye (Secale cereale L.) and oat s (Avena sativa L.) and the weed species Black-grass (Alopecurus myosuroide s Huds.), Corn chamomile (Anthemis arvensis L.), Field bindweed (Convolvulu s arvensis L.), Cleavers (Calium aparine L.), Black bindweed (Polygonum con volvulus L.), Charlock (Sinapis arvensis L.) and Field penny-cress (Thlaspi arvense L.) were classified using two different algorithms. With the minim um distance algorithm between 72 % and 97 % of the seeds were correctly ide ntified. The average rate of correct identification was 89 %. Based on a fu zzy logic algorithm between 66 % and 100 % of all seeds were classified to the right species. The average rate of correct identification was 86 %. The system presented in this study can be used to analyze the degree of pur ity or the sort of seed samples.