基于卷积神经网络的ISAR图像目标识别  

ISAR Image Target Recognition Based on Convolutional Neural Network

作　　者：刘瑾 秦长海 占银玉 王凤杰 涂建 LIU Jin;QIN Changhai;ZHAN Yinyu;WANG Fengjie;TU Jian(Department of Computer Science and Engineering,Shanghai Jiao Tong University,Shanghai 200240,China;The 723 Institute of China State Shipbuliding Corporation Limited,Yangzhou 225000,Jiangsu,China;Shanghai Radio Equipment Research Institute,Shanghai 201109,China)

机构地区：[1]上海交通大学计算机科学与工程系,上海200240 [2]中国船舶集团有限公司第七二三研究所,江苏扬州225000 [3]上海无线电设备研究所,上海201109

出　　处：《制导与引信》2024年第3期27-34,53,共9页Guidance & Fuze

基　　金：上海航天科技创新基金(USCAST2021-2)。

摘　　要：利用电磁仿真软件构建了多角度、多分辨率、多目标的逆合成孔径雷达(inverse synthetic aperture radar,ISAR)图像数据集,采用AlexNet、VGGNet、ResNet、MobileNet等卷积神经网络对仿真ISAR图像进行了训练和多目标分类识别实验,并对实验结果进行了比较分析。实验结果表明,4种网络模型识别性能接近,目标识别时间在15 ms内,识别准确率接近100%,且具备轻量化识别条件。A multi-angle,multi-resolution and multi-target inverse synthetic aperture radar(ISAR)image database was constructed using electromagnetic simulation software.Convolutional neural networks,including AlexNet,VGGNet,ResNet and MobileNet were trained on simulated ISAR images for multi-target classification and recognition experiments.The experimental results were compared and analyzed.The results demonstrate that the four networks exhibit similar recognition performance,achieving target recognition within 15 ms with an accuracy rate close to 100%.Additionally,the four networks possess the potential for lightweight recognition.

关 键 词：深度学习 卷积神经网络 图像分类 目标识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]