Output details
12 - Aeronautical, Mechanical, Chemical and Manufacturing Engineering
University of Birmingham : A - Mechanical Engineering
Evolutionary feature selection applied to artificial neural networks for wood-veneer classification
FeaSANNT, an evolutionary algorithm for optimisation of neural networks developed by Dr Castellani, was used to train a multi-layer neural network to identify wood veneer defects. Given a fixed network structure, FeaSANNT concurrently evolved the input feature vector and the network weights. The novelty of the method lay in the implementation of the embedded approach in an evolutionary feature selection paradigm. Tests showed that FeaSANNT produced high-performing solutions with robust learning results. Compared with two standard wrapper and filter methods, FeaSANNT gave solutions with equal or higher accuracy using fewer features. This will facilitate industrial use of pattern classification.