基于双通道特征增强集成注意力网络的红外弱小目标检测方法  

Infrared Dim and Small Target Detection Based on DualChannel Feature-Enhancement Integrated Attention Network

作　　者：蔡云泽[1,2,3,4,5] 张彦军 CAI Yunze;ZHANG Yanjun(Department of Automation,Shanghai Jiao Tong University,Shanghai,200240;Key Laboratory of System Control and Information Processing,Ministry of Education of China,Shanghai,200240;Shanghai Engineering Research Center of Intelligent Control and Management,Shanghai 200240;Key Laboratory of Marine Intelligent Equipment and System of Ministry of Education,Shanghai Jiao Tong University,Shanghai,200240;Institute of Marine Equipment,Shanghai Jiao Tong University,Shanghai,200240)

机构地区：[1]上海交通大学自动化系,上海200240 [2]系统控制与信息处理教育部重点实验室,上海200240 [3]上海工业智能管控工程技术研究中心,上海200240 [4]海洋智能装备与系统集成技术教育部实验室,上海交通大学,上海200240 [5]上海交通大学海洋装备研究院,上海200240

出　　处：《空天防御》2021年第4期14-22,共9页Air & Space Defense

基　　金：国家自然科学基金重大科研仪器研制项目(No.61627810);国家科技重大专项(No.2018YEB1305003);国防科技卓越青年科学基金(No.2017-JCJQ-ZQ-031)。

摘　　要：针对目前远距离红外目标探测中存在的弱小目标特征信息少、环境复杂、噪音干扰多,且传统目标检测算法漏检率和虚警率高等问题,提出了基于双通道特征增强集成注意力网络的红外弱小目标检测算法。整体网络结构主要包括双通道特征提取模块、特征增强模块和集成上下注意力模块三部分。与单通道特征提取相比,双通道特征提取可以获得更多的特征信息,并通过特征增强模块进一步丰富目标特征,再结合集成上下注意力模块自适应地增强目标特征和弱化背景噪音,从而提升弱小目标的检测效果。最后,通过对比传统算法、其他深度学习算法以及本文提出算法的测试结果,验证了本文提出的算法具有较好的检测效果,漏检率和误检率更低。Aiming at the problems existing in long-distance infrared target detection, such as less feature information,complex environment, more noise interference, and high missed detection rate and false alarm rate of traditional target detection algorithms, an infrared small target detection algorithm based on dual-channel feature enhancement attention network is proposed in this paper. The overall network structure mainly includes three parts: dual channel feature extraction module, feature enhancement module and integrated top-bottom attention module. Compared with single channel feature extraction, dual channel feature extraction can obtain more feature information. Feature enhancement module can enrich target features further. Moreover, the integrated top bottom attention module can adaptively enhance target features and weaken background noise. And then the algorithm improves the detection effect of dim and small targets in the infrared images. Finally, it is verified that the algorithm proposed in this paper has a better detection effect, and has a lower rate of missed detection and false detection.

关 键 词：红外图像 弱小目标检测 深度学习 特征增强 注意力机制 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]