Skip to main content

Classification of marine phytoplankton (dinoflagellates) using neural networks

Exploitable results

The team has developed a novel image analysis system, that can be trained to categorise images of natural objects. The work has focused on developing robust visual categorisation methods for use with natural objects. Its operation is based loosely on current theories of mammalian vision, but its computational load is not excessive. When developed further, the system should be capable of operating with images of any viewpoint of an object, and on multiple objects within the field of view. Experiments have shown that the system can be trained to catetgorise a variety of marine biological specimens and man-made objects without any modification to the software. The operational range of targets being defined by the training data. This system is of a type being developed in the light of new results arising from research into artificial neural networks and mammalian visual perception. It is, however, the first application of its type to be applied to marine biological specimens. Neural network based expert categorisers of this type will, in time, revolutionise modelling of marine ecology systems.