基于特征重聚焦网络的多尺度近岸舰船检测  

Multi-Scale Inshore Ship Detection Based on Feature Re-Focusing Network

作　　者：刘荻 张焱[1] 赵琰 石志广[1] 张景华 张宇[1] Liu Di;Zhang Yan;Zhao Yan;Shi Zhiguang;Zhang Jinghua;Zhang Yu(National Key Laboratory of Science and Technology on Automatic Target Recognition,College of Electronic Science and Technology,National University of Defense Technology,Changsha,Hunan 410073,China;State Key Laboratory of Complex Electromagnetic Environment Effects on Electronics and Information System,College of Electronic Science and Technology,National University of Defense Technology,Changsha,Hunan 410073,China)

机构地区：[1]国防科技大学电子科学学院ATR重点实验室,湖南长沙410073 [2]国防科技大学电子科学学院电子信息系统与复杂电磁环境效应国家重点实验室,湖南长沙410073

出　　处：《光学学报》2021年第22期129-141,共13页Acta Optica Sinica

基　　金：国家自然科学基金(62075239,61302145)。

摘　　要：针对监控视频中的多尺度近岸舰船检测问题,提出了一种基于特征重聚焦网络的舰船目标检测算法,设计了由多维特征聚合模块(MFAM)与注意力特征重构模块(AFRM)组成的特征重聚焦策略。其中,MFAM基于输入的特征金字塔构建特征聚合块,进一步融合多尺度舰船不同层次特征的语义信息。AFRM基于多分支空洞卷积以及通道与空间注意力机制提升网络对目标非局部信息的表征和对背景干扰的抑制,并构建了用于目标检测的特征重聚焦金字塔。在Seaships7000舰船公开数据集上的实验结果表明,相比其他算法,本算法对监控视频中多尺度近岸舰船的检测效果更好。Aiming at the problems of multi-scale inshore ship detection in surveillance videos,this paper proposes a ship target detection algorithm based on feature re-focusing network,and designs a feature re-focusing strategy,which consists of a multi-scale feature aggregation module(MFAM)and attention feature re-assignment module(AFRM).Specifically,MFAM fuses the semantic information of different levels of features of multi-scale ships by constructing a feature aggregation block based on the input feature pyramid.AFRM is composed of multi-branch dilated convolutions as well as channel and spatial attention mechanisms,which can improve the network′s representation of target non-local information and suppressing interference of background,and a feature re-focusing pyramid is established for target detection.The experimental results on the Seaships7000 ship public data set show that compared with other algorithms,the algorithm has a better detection effect on multi-scale inshore ships in surveillance videos.

关 键 词：机器视觉 卷积神经网络 舰船目标检测 特征重聚焦网络 注意力机制 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]