摘要
利用神经网络方法进行雷达信号识别存在两个问题,一是难以选择最优的网络结构;二是用传统的BP学习算法,常常收敛到局部解。本文提出一种GANN方法,即首先利用遗传算法优化两层前馈神经网络结构以确定中间隐层的节点数,然后用遗传算法进行学习。通过与BP算法相比较,遗传算法不仅速度快,而且能找到最优解。实验表明,将GANN应用于雷达信号识别,识别率更高。
There are two problems of using neuron network method to recognize radar signal :one is difficult to affirm a best net structure ,the other is that the network always converge toward some local minimum in error by using BP method. This paper proposes a GANN method of radar signal recognition, and namely makes use of genetic algorithm to optimize the two-layer feed forward network to determine the hidden layer's node, then use genetic algorithm to train the network.Compared with BP method,the GA method is not only faster but also can find the optimal solution. The correct rate of recognizing radar signals by GANN is higher through a number of experiments.