基于PSO-TDNN的空间目标识别

Discrimination of Space Target Based on PSO-TDNN

针对空间目标的RCS特征识别的问题,提出了基于粒子群算法(PSO)训练的时延神经网络(TDNN)识别方法。首先研究了时延神经网络的结构模型和梯度下降训练法,由于梯度下降训练法存在收敛速度缓慢、容易陷入局部极小值等缺点,提出了基于粒子群算法的训练方法,将时延神经网络的训练过程转化为群体随机优化问题。最后,提取两类空间目标的RCS实测数据小波特征,利用各类神经网络进行识别比较发现:基于粒子群算法的时延神经网络(PSO-TDNN)具有分类能力强,收敛速度快等优点。

In order to identify the characteristic of the exo-atmospheric space target＇s RCS,the time-delay neural network（TDNN）with particle swarm optimization（PSO）training method is proposed.First,the TDNN model and the grads descend method learning algorithm are studied.particle swarm optimiza-tion training method is then proposed,because the grads descend method（GDM）learning algorithm has the defects of slow-paced constringency,slumping into part minimum easily and so on.This method chan-ges the neural network learning process to collective random optimum problem.Last,by picking up the wavelet characteristic of space target＇s RCS from the data of radar,the performances of neural network classifiers results show that PSO-TDNN has a good discrimination ability and faster-paced constringency.