Extended target invariant recognition based on multiple BP neural networks

A new method of extended target invariant recognition by multiple BP neural networks is proposed in this paper. The boundary moment invariants of image target are extracted as the neural network inputs and the reasonable extraction approach is given. Through analyzing the criticisms of the traditional back propagation neural network (BPNN), the multiple neural networks based on improved sub-BPNNs are proposed to promote convergence performance. When new target class is added, this recognition system can be retrained very fast. The experimental results indicate that extended targets which have rotation, translation and scale variance can be recognized with high efficiency and correct ratio. © 2008 by Binary Information Press.