摘要
针对红外小目标占用像素较少、背景相似性强、网络容易受到背景杂波信息干扰的问题,提出了一种基于注意力机制的红外小目标检测方法。利用注意力机制模块抑制背景杂波,增强小目标特征,并使用红外小目标检测模块实现检测任务;为了增强网络鲁棒性,通过高斯噪声与原图通道堆叠输入的数据增强方式提升网络抗杂波干扰的能力。实验表明,提出的方法在MDvsFA数据集中的性能超过了目前最新的对比算法。
Aiming at the problems of small infrared targets occupying less pixels,strong background similarity and easy to be interfered by background clutter information,this paper proposes a small infrared target detection method based on attention mechanism.In this method,the attention mechanism module is used to suppress background clutter and enhance small target features,and the infrared small target detection module is used to achieve the detection task.At the same time,in order to enhance the robustness of the network,this paper adopts the data enhancement method of gaussian noise and original image channel stacking input to improve the anti-clutter interference ability of the network.Through experimental verification,the performance of the proposed method in MDVSFA data set exceeds that of the latest methods.
作者
董亚盼
高陈强
谌放
刘芳岑
DONG Yapan;GAO Chenqiang;CHEN Fang;LIU Fangcen(School of Communication and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,P.R.China;Chongqing Key Laboratory of Signal and Information Processing,Chongqing 400065,P.R.China)
出处
《重庆邮电大学学报（自然科学版）》
CSCD
北大核心
2023年第2期219-226,共8页
Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)
基金
国家自然科学基金(62176035,61906025)
重庆市自然科学基金(cstc2020jcyj-msxmX0835,cstc2021jcyj-bsh0155)。
关键词
红外小目标
深度学习
注意力机制
高斯噪声
infrared small target
deep learning
attentional mechanism
gaussian noise
