基于深度迁移学习的低分辨雷达目标识别方法  

Low-resolution radar target recognition method based on deep transfer learning

作　　者：施端阳 林强[1] 胡冰[1] 牛闯 SHI Duanyang;LIN Qiang;HU Bing;NIU Chuang(Air Force Early Warning Academy,Wuhan 430019,China;The No.95174 of PLA,Wuhan 430040,China)

机构地区：[1]空军预警学院,武汉430019 [2]95174部队,武汉430040

出　　处：《兵器装备工程学报》2023年第11期248-255,共8页Journal of Ordnance Equipment Engineering

基　　金：全军军事类研究生重点资助课题(JY2020B150)。

摘　　要：针对低分辨雷达人工目标识别效率较低的问题,提出了基于深度迁移学习的雷达自动目标识别方法。该方法利用雷达回波序列轮廓像构建空中目标数据集,使用深度卷积神经网络自动提取回波数据中的深层特征,并对雷达目标进行分类识别。为了解决深度学习对样本量的巨大需求,在分类模型训练时,引入迁移学习思想,将经ImageNet数据集预训练过的初始网络模型迁移到雷达目标识别任务中,再通过空中目标数据集对模型参数进行微调,实现小样本条件下对空中目标的粗分类。实测数据的结果表明:所提方法能够在小样本条件下较为准确地对空中目标的大小和架次进行分类识别,具有良好的识别性能。Aiming at the low efficiency of low-resolution radar artificial target recognition,a radar automatic target recognition method based on deep transfer learning is proposed.In this method,sequence outline images of radar echo are used to construct aerial target dataset,and deep convolutional neural network is used to automatically extract the deep features in the echo data,and the radar target is classified and recognized.In order to solve the huge demand for the sample size of deep learning,the idea of transfer learning is introduced during the training of the classification model.The initial network model pre-trained by the ImageNet data set is transferred to the radar target recognition task,and then the model parameters are fine-tuned through the aerial target data set to realize the rough classification of air targets under the condition of small samples.The experimental results of the measured data show that the proposed method can classify and recognize the size and sorties of aerial targets more accurately under the condition of small samples and has good recognition performance.

关 键 词：深度学习 迁移学习 低分辨雷达 小样本 目标识别 

分 类 号：TN957[电子电信—信号与信息处理]