| 摘要: |
| 深度卷积神经网络（DCNN）可自动学习目标层次化特征，在合成孔径雷达（SAR）自动目标识别（SAR-ATR）领域具有广泛应用前景。首先，介绍了DCNN的基本原理以及DCNN在光学图像上的应用与发展；然后，介绍了SAR-ATR的基本概念，综述了DCNN在SAR图像语义特征提取、片段级SAR图像分类、基于数据增强技术的SAR自动目标识别、异质图像变化检测等领域中的前沿应用研究及代表性网络架构；最后，总结并讨论了DCNN在SAR-ATR应用中存在的参数设置经验化、算法泛化能力较弱等不足，并对未来研究方向进行了展望。 |
| 关键词: 合成孔径雷达 自动目标识别 深度卷积神经网络 应用综述 |
| DOI： |
|
| 基金项目:国家自然科学基金资助项目（61302153）；航科学空基金资助项目（20160196003） |
|
| Applications of deep convolutional neural network in SAR automatic target recognition:a summarization |
| XU Qiang,LI Wei,PIERRE Loumbi |
| () |
| Abstract: |
| Deep Convolutional Neural Network(DCNN) can automatically learn the target’s hierarchical features,and it has wide application prospect in SAR-Automatic Target Recognition(SAR-ATR). Firstly,the basic principle of DCNN is introduced,and the application and development of DCNN in optical image are studied. Then,the basic concept of SAR-ATR is introduced,and the frontier application research and representative network architecture of DCNN in SAR image semantic feature extraction,fragment-level SAR image classification,SAR automatic target recognition based on data enhancement technology,heterogeneous image change detection are reviewed. Finally,the lack of parameter setting and the weak generalization ability of DCNN in SAR-ATR applications are summarized and discussed,and the future research direction is presented. |
| Key words: synthetic aperture radar(SAR) automatic target recognition(ATR) deep convolutional neural network(DCNN) application summarization |
