基于CNN的SAR图像目标和场景分类算法

陈立福; 武鸿; 崔先亮; 郭正华; 贾智伟

引用本文:	陈立福，武鸿，崔先亮，郭正华，贾智伟. 基于CNN的SAR图像目标和场景分类算法[J]. 雷达科学与技术, 2018, 16(6): 627-632.[点击复制]
	CHEN Lifu, WU Hong, CUI Xianliang, GUO Zhenghua, JIA Zhiwei. SAR Image Target and Scene Classification Algorithm Based on CNN[J]. Radar Science and Technology, 2018, 16(6): 627-632.[点击复制]

【打印本页】【下载PDF全文】【查看/发表评论】【下载PDF阅读器】【关闭】

←前一篇|后一篇→

过刊浏览高级检索

本文已被：浏览 5174次下载 822次	码上扫一扫！
分享到：微信更多字体:加大+\|默认\|缩小-
基于CNN的SAR图像目标和场景分类算法
陈立福，武鸿，崔先亮，郭正华，贾智伟
长沙理工大学电气与信息工程学院，湖南长沙410114

摘要:

随着合成孔径雷达（Synthetic Aperture Radar， SAR）成像技术的发展和SAR图像数据的急剧增加，SAR图像解译技术成为了当前的研究热点。针对SAR图像的目标和场景分类问题，提出了一种改进的基于卷积神经网络的图像分类算法。为克服卷积神经网络训练过程中因数据量不足而出现的过拟合问题，采用数据增强人工增加训练样本的大小；针对高层卷积层参数过多的问题，采用一种多尺度卷积模块替代高层的卷积层；在输出层采用卷积和全局均值池化的组合替代传统的全连接层，大幅度减少了网络参数。网络训练阶段，通过误差反向传播来更新网络参数。针对MSTAR数据集和高分辨率的机载SAR图像分别进行目标及场景分类，实验结果表明该算法实现了较好的分类性能。

关键词: 卷积神经网络深度学习合成孔径雷达误差反向传播

DOI：10.3969/j.issn.1672-2337.2018.06.008

分类号:TN958;TP183;TP751

基金项目:国家自然科学基金（No.41201468,81401490）；湖南省教育厅项目（No.16B004）

SAR Image Target and Scene Classification Algorithm Based on CNN

CHEN Lifu, WU Hong, CUI Xianliang, GUO Zhenghua, JIA Zhiwei

College of Electrical and Information Engineering, Changsha University of Science and Technology, Changsha 410114, China

Abstract:

With the development of synthetic aperture radar (SAR) imaging technology and the sharp increase of SAR image data, SAR image interpretation technology has become the current research hotspot. An improved convolution neural network algorithm is proposed for the target and scene classification of SAR images. In order to overcome the problem of over-fitting due to insufficient data in the training process of convolutional neural network, data enhancement is adopted to increase the size of training samples. A multi-scale convolution module is used to replace the high-level convolutional layer. In the output layer, the combination of convolution and global mean pooling is used to replace the traditional full connection layer, which greatly reduces network parameters. During the network training phase, the network parameters are updated by the error back propagation. According to the target and scene classification of the MSTAR dataset and the high resolution airborne SAR images, the experimental results show that the algorithm achieves better classification performance.

Key words: convolutional neural network deep learning synthetic aperture radar error back propagation