摘要
针对基于雷达无人机目标识别难度高、精确度低、适应性差等问题,本文提出了利用深度学习的方法,对雷达回波进行无人机检测。首先,利用相参累积的方法生成雷达回波的距离-多普勒图像,增强目标特征并提高信噪比;其次,采用生成对抗模型对距离-多普勒图像进行数据扩充,以获得充足的图像数据减小网络的过拟合并提高网络鲁棒性;最后,使用基于位置感知的卷积神经网络增强特征,通过构建基于距离-多普勒图像的感知模块,实现对目标距离和运动速度的检测。通过在雷达回波序列中弱小飞机目标检测跟踪数据集上验证的结果表明:最终检测结果在召回率89%的情况下达到了91%的准确率。相比于基准方法,本文提出的方案具有更高的检测精度和更好的网络运行效率。
In view of the problems of unmanned aerial vehicle(UAV) detection based on radar, e. g., high difficulty, low accuracy, and poor adaptability, a deep learning method is proposed to detect the UAV targets from radar echoes. Firstly, the Range-Doppler images of the radar echoes are generated by the coherent accumulation method, which enhances the target features and improves the signal-to-noise ratio(SNR). Secondly, a generative adversarial model is used to augment the Range-Doppler image data so as to obtained sufficient image data, by which the overfitting of the network is reduced and the network robustness is improved. Finally, the convolutional neural network based on position perception is used to enhance the features, and the detection of the distance and speed of the target is realized by constructing a perception module based on the Range-Doppler images. The effectiveness of the proposed method is verified on the weak small aircraft target detection and tracking data set in the radar echo sequence.The final detection result achieves the accuracy of 91% with a recall rate of 89%. Compared with the baseline method, the proposed scheme has higher detection accuracy and better network inference efficiency.
作者
赵宏宇
张志文
公茂果
武越
叶舟
吕宇宙
张杨
ZHAO Hongyu;ZHANG Zhiwen;GONG Maoguo;WU Yue;YE Zhou;LYU Yuzhou;ZHANG Yang(School of Electronic Engineering,Xidian University,Xi’an 710071,Shaanxi,China;School of Computer Science and Technology,Xidian University,Xi’an 710071,Shaanxi,China;Shanghai Institute of Aerospace Electronic and Communication Equipment,Shanghai 201108,China)
出处
《上海航天（中英文）》
CSCD
2023年第1期61-69,共9页
Aerospace Shanghai（Chinese&English）
基金
广东省重点领域研发计划资助项目(2020B090921001)。
关键词
卷积神经网络
目标检测
多普勒雷达
距离-多普勒图像
空洞卷积
convolutional neural network
object detection
Doppler radar
Range-Doppler image
dilated convolution
