Classification of euceratium gran. in neural networks

Abstract

Two species of Euceratium, Ceratium arcticum (Ehrb.) and C. longipes (Bail.), were selected in an attempt to automatically distinguish one species from another. A backward error-propagation neural network model was given data obtained from processing images of specimen Ceratium, trained to respond correctly to that data, and was subsequently found to have a significant ability to classify new (unseen) data. This ability to learn a set of training data shows that the network was able to formulate an internal representation that distinguished between the two species. Furthermore, that the network was able to correctly classify new data indicates that the representation so formed is sufficient to distinguish the species in general. It is suggested that the application of neural networks to this domain will enable automatic classification of species from images.<>