基于深度卷积神经网络的弹道目标微动分类

针对传统弹道目标微动分类缺乏智能性及噪声条件下分类性能差的问题，利用深度学习的高维特征泛化学习能力，提出一种将深度卷积神经网络用于弹道目标微动分类的方法。首先，在建立弹道目标微动模型的基础上，分析3种微动形式下的微多普勒表示，并生成雷达回波信号的时频图,作为训练、验证及测试的数据集；然后，运用深度卷积神经网络中的迁移学习对AlexNet和GoogLeNet进行再训练；最后，利用训练后的网络实现3种微动形式下的目标分类，并研究信噪比对分类性能的影响。仿真结果表明，与传统的微动目标分类方法相比，该方法不仅智能化程度高，而且在低信噪比条件下分类准确性更强。

Micro Motion Classification of Ballistic Targets Based on Deep Convolutional Neural Network

Aimed at the problems that the traditional ballistic targets micro motion classification is lack of intelligence and the classification performance is poor under noise conditions , by using the high dimensional feature generalization learning ability of deep learning, a method of using deep convolution neural network for ballistic target micro motion classification is proposed. Firstly, based on the establishment of the ballistic target micro motion model, the micro Doppler representations of the three micro motion forms are analyzed, and the time frequency map of the radar echo signals is generated as the data set for training, verification and testing; The transfer learning in deep convolution neural network is used to retrain AlexNet and GoogLeNet. Finally, the target network classification in three micro motion forms is realized by using the trained network, and the influence of signal to noise ratio on classification performance is studied. The simulation results show that compared with the traditional micro motion target classification method, the method is not only high in intelligence, but also is good in classification accuracy under low SNR conditions, and is guidable in the classification of ballistic targets.