BP网在多灰度运动目标识别中的应用

Moving object Recognition by BP Neural Network

介绍了利用ＢＰ网进行多灰度运动目标识别的方法。针对传统的ＢＰ网训练时间长等缺点以及多灰度缓目标识别的特点，对传统ＢＰ算法作了适当的改进，提出了训练样本组织和选择的一些方法和技巧，并将以上的改进算法及组织和选择样本的原则应用到多灰度坦克目标识别的ＢＰ同训练的实验中．实验结果表明改进的算法及所采用的原则是合理的，它可以有效地缩短ＢＰ网的学习时间、防止网络的振荡，并且能在一定程度改善ＢＰ网的性能．

This paper introduced the pattern characters of the general BP neural network and proposed a improved learning method and some approaches of organizing the training samples'set, which can avoid oscillation of the network while training, reduce the learning time and enhance the recognition rate. At last, we apply these approaches in the recognition of tank in real world, and get an ideal result.