一种基于卷积神经网络的HRRP序列空间目标识别方法

Spatial Target Recognition Method of HRRP Sequence Based on Convolutional Neural Network

本文利用高分辨距离像(HRRP)序列所含的丰富信息,提出了一种基于卷积神经网络(CNN)的空间目标识别方法。利用CNN自动地从序列图中学习稳定的特征,进而实现目标分类。该方法不仅考虑了目标的尺寸、结构等信息,同时也在一定程度上利用了目标的运动特性,提高了目标的区分度。在仿真实验中,从雷达部署点、飞行轨迹以及两者的混合影响等三个方面,与传统的目标识别算法进行了对比,结果表明,该方法可以实现对不同空间目标的有效识别,平均识别准确率超过95%,具有较强的鲁棒性。

In this paper,a method of spatial target recognition based on Convolution Neural Network(CNN)is proposed,which is based on the abundant information contained in the High-resolution Range Profile(HRRP)sequence.CNN is used to automatically learn the features from the sequence diagram,and then realizes the target classification.The method not only takes into the size and structure of target,but also makes use of kinetic characteristics of target to a certain extent and improves the discrimination of the target.In the simulation experiment,the traditional target recognition algorithm is compared from three aspects:radar deployment point,trajectory and the mixed influence of both them.The results show the validity and robustness of this method for identifying different spatial targets,and the average recognition accuracy is more than 95%.