Application of neural network to identification of soybean varieties usingleaflet shape images

We suggested a model that substitutes of experts' visual evaluation on leaf shape. In previous studies, several researchers successfully created leaf s hape evaluation models based on shape features, but it is difficult to find the common leaf shape features to cover various species simultaneously We, therefore, contrived a new generalized model requiring no features using n eural network with direct image input. Tn this study, we applied this model to the varietal discrimination of soybean leaflet shape. The identificatio n rate of 38 varieties among 364 individual leaflets given by the model was 80%. Copyright (C) 1998 IFAC.