Using back propagation algorithm and genetic algorithm to train and refine neural networks for object detection

Abstract

We introduce a two stage approach to the use of pixel based neural networks for finding relatively small objects in large pictures. In the first stage the network is trained by using a back propagation algo­rithm and a genetic algorithm on sample objects which have been cut out from the large pictures. In the second stage the weights of the two trained networks are refined on the full training images using a second genetic algorithm. Four methods are formed by the two training algo­rithms in stage one and combining them with the genetic algorithm in stage two. We have tested these methods on three object detection prob­lems of increasing difficulty. In all cases the methods with the refined genetic algorithm resulted in improved detection performance over those without the refinement. The method with the best performance is the use of the back propagation algorithm for the network training and the genetic algorithm for the network refinement.