基于深度卷积神经网络的红外小目标检测  

Small target detection in infrared images using deep convolutional neural networks

作　　者：吴双忱 左峥嵘[1] WU Shuang-Chen;ZUO Zheng-Rong(National Key Laboratory of Science&Technology on Multi-spectral Information Processing,School of Artificial Intelligence and Automation,Huazhong University of Science and Technology,Wuhan 430074,China)

机构地区：[1]华中科技大学人工智能与自动化学院多谱信息处理技术国家级重点实验室

出　　处：《红外与毫米波学报》2019年第3期371-380,共10页Journal of Infrared and Millimeter Waves

基　　金：国家自然科学基金(61773389)~~

摘　　要：提出了一种新的解决红外图像小目标检测问题的深度卷积网络将对小目标的检测问题转化为对小目标位置分布的分类问题检测网络由全卷积网络和分类网络组成全卷积网络对红外小目标进行增强和初步筛选实现红外图像的背景抑制分类网络以原始图像和背景抑制后的图像为输入对目标点后续筛选网络中引入 SEnet(Squeeze and Excitation Networks)对特征图进行选择实验验证了整个检测网络相对于传统小目标检测算法的优势所提出的基于深度卷积神经网络的小目标检测方法对复杂背景下低信噪比且存在运动模糊的小目标具有很好的检测效果.A new deep convolutional network for detecting small targets in infrared images is proposed. The problem of small targets detection is transformed into the classification of small targets’location distribution. First,a Fully Convolutional Networks is used for enhancing and initially screening the small targets. After that,the original image and the background suppressed image are selected as the inputs for classification network which is used for the follow-up screening,and then the SEnet ( Squeeze-and-Excitation Networks) is used to select the feature maps. The experimental results show that the detection network is superior to multiple typical infrared small target detection methods and has an excellent result on different signal-to-noise ratio,different scenes and motion blur targets.

关 键 词：模式识别与智能系统 红外小目标检测 深度卷积网络 低信噪比 运动模糊 SEnet 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置]