摘要
本文研究了基于模块化神经网络的谐振区雷达目标识别方法,该方法先用BP网络进行波形预测,再用最大后验概率准则或修正最小平方误差准则进行分类。通过计算机模拟,证实了该方法有较好的识别性能,且对信号留数变化,即目标的姿态变化不敏感。另外,该方法还具有实现简单,结构扩展方便等优点。
A new method for radar target recognition based on modular neural networks is reported in this paper. In this method, the response from an unknown target is first sent to several waveform predicators that are some BP neural networks trained by responses from known targets respectively. Then the predicator errors are inputted to a classifier using the rule of maximum a posteriori or the rule of modified minimum squared errors. The simulation on PC computer shows that the new method has a good performance on radar target recognition. The method also has other advantages such as easy realization, clear structure and easy expansion.
