雷达探测领域,各种虚假目标的主动干扰对单体雷达目标探测带来了极大挑战。为了提高单体雷达的抗干扰性能,文中基于宽带雷达目标回波数据,计算了多角度一维距离像,并作为标签训练数据,引进成熟的卷积神经映射网络,通过大量虚假目标与真实目标回波的训练调参,成功构建雷达目标识别网络,可以从雷达接收机大量各类型中频回波数据中提取真实雷达目标。文中的雷达特征数据是通过CST 软件进行三维目标建模仿真计算得到的,为了提高网络识别的鲁棒性,在仿真雷达目标回波样本的基础上增加了微小的扰动,雷达目标提取网络可以依据训练数据的增大而使得雷达目标探测更加准确。

In domain of radar detection, great challenges are brought by all kinds of active jamming for single radar detection. In order to improve the anti-interference performance of single radar, according to the data of radar target echo, one-dimensional range profiles are calculated and used as label training data. The mature convolutional neural mapping network is introduced to construct the radar target recognition network through training by large numbers of false targets and real targets, and real radar target can be recognized easily from the large amount of IF data. At present, radar characteristic data is obtained through 3D modeling and simulation of CST software, small perturbation is added on the basis of simulating radar target echo samples for improving the robustness of network. Based on the increase of training data, the optimized radar target extraction network can make the radar more accurate.

TN974

电子信息系统复杂电磁环境效应国家重点实验室基金资助项目(CEMEE2023Z0201)