基于局部连接型多层前馈网络的雷达目标动态回波识别

Dynamic Radar Target Waveform Recognition Based on Locally Connected Multilayer Perceptron

提出了一种适合于舰船雷达目标动态回波序列识别用的新的有效多层前馈网络──局部连接平行处理型网络研究了ＢＰ网络学习算法的综合改进方法，并相对于常规全连接型ＢＰ网络分析了局部连接型网络的动态特征抽取能力和目标识别性能。

A new type of feedforward neural network,which is a locally connected multilayer perceptron,is proposed for dynamic radar target waveform recognition。To counter the huge training set of problem domain,an improved systematical training method is explored to acceleratethe back-propagation algorithms learming speed。The practical tests demonstrate that the proposed locally connected multilayer network has much less time and space complexity, strong ability of dynamic feature extraction and representation,higher accuracy of pattern space separation and recognition than the conventional multilayered feedforward networks.