摘要
由于深度学习在目标识别方面取得了显著的成绩,为提高合成孔径雷达(synthetic aperture radar,SAR)图像目标识别的精度与速度提供了新的思路。本文将区域全卷积网络(region-based fully convolutional networks,R-FCN)结构应用于SAR图像目标识别中,取得了良好的效果。对于数据集较小和数据相似度较高的问题,提出了基于迁移学习的R-FCN模型用于SAR图像目标识别。对更快的区域卷积神经网络(faster region convolutional neural networks,Faster R-CNN)和R-FCN进行模型训练及优化,并与所提出的基于迁移学习的改进R-FCN模型实验结果进行对比。结果表明,所提方法对SAR图像具有更好的识别效果和更快的识别速度。
With remarkable achievements in target recognition,deep learning provides new ideas for improving the accuracy and speed of target recognition in synthetic aperture radar(SAR)images.In this paper,region-based fully convolutional networks(R-FCN)are applied to SAR image target recognition,and good results have been achieved.In order to solve the problem of small data set and high data similarity,the R-FCN model based on transfer learning is proposed for target recognition in SAR images.The faster region convolutional neural networks(Faster R-CNN)and the R-FCN models are trained and optimized,and the experimental results are compared with the improved R-FCN model based on transfer learning proposed in this paper.The results show that the proposed method has better recognition effects and faster recognition speed for SAR images.
作者
周晓玲
张朝霞
鲁雅
王倩
王琨琨
ZHOU Xiaoling;ZHANG Zhaoxia;LU Ya;WANG Qian;WANG Kunkun(College of Physics and Optoelectronic, Taiyuan University of Technology, Taiyuan 030024, China)
出处
《系统工程与电子技术》
EI
CSCD
北大核心
2022年第4期1202-1209,共8页
Systems Engineering and Electronics
基金
2020年度山西省高等学校科技成果转化培育项目
山西省重点研发计划项目(高新技术领域)(201803D121057)资助课题。
关键词
机器视觉
目标识别
合成孔径雷达
全卷积网络
迁移学习
machine vision
target recognition
synthetic aperture radar(SAR)
fully convolutional network(FCN)
migration study
