BP网络进化及其在雷达目标识别中的应用

Evolutionary Algorithm for BP Neural Network and Its Application to Radar Target Recognition

针对常规ＢＰ神经网络的ＢＰ算法只能训练固定结构的神经网络，存在诸如易落入局部极值、没有引入提高泛化能力的训练机制等固有不足之处，以及一些神经网络进化算法的进化机制中存在的缺陷，本文提出一种ＢＰ神经网络进化算法，并用于高分辨雷达目标一维距离像的识别问题。实验结果表明，经所述方法优化后的神经网络结构简单、泛化能力优于ＢＰ算法和一些进化算法训练的网络。

Presents a new evolutionary algorithm for neural network and the algorithm is used for the classification of high range resolution radar targets. The algorithm overcomes inherent shortcomings of classical BP algorithms, such as fixed architecture weights training, local extrema problem and lack of generalization of training mechanism, and improves sme existing evolutionary algorithms for neural network. Simulations show that the trained neural network with the algorithm has simple architecture and better generalization ability over classical BP neural network and some other existing evolutionary algorithms for neural network.