A decision tree of neural networks for classifying images of wood veneer

Artificial neural networks for the classification of wood veneer by an automatic visual inspection system are presented. Initially, a single large neural network is implemented with eleven image features as inputs and thirteen outputs — one for each class of veneer. In order to improve on the classification accuracy of this single network, a decision tree of smaller and more specialised modular neural networks is introduced to achieve a classification by successive refinement. This results in a substantial improvement in classification accuracy. A key process in the design of a modular neural network is the use of “normalised inter-class variation” in the selection of the most appropriate image features to be used for its particular specialised classification task.