Automatic target recognition using a constructive approach

This paper presents the use of constructive neural networks for pattern recognition. The work presented here is part of the SAPRI project (System for Acquisition, Processing and Image Recognition) which is being developed for the Brazilian Navy. The objective of this paper is to investigate the behavior of constructive neural networks in image recognition tasks, specially with images from the SAPRI project. Constructive neural networks differ from standard networks due to their ability to change their own number of elements, adding or removing units and connections. Four constructive algorithms are presented in this paper: cascade correlation, tower, pyramid and upstart.