摘要
在基于深度学习的SAR图像目标检测中,为了降低SAR图像中噪声等特性对特征学习的干扰,提高检测方法中识别和定位任务的交互性,采用了联合任务检测方法。该方法利用联合任务网络,使识别和定位任务在尽可能共用特征的同时保留各自的特殊性,从而提升两个子任务对特征学习的监督能力。此外,该方法还利用联合任务学习方法,在锚框选择和损失函数计算中,同时考虑识别和定位任务的可靠程度,从而提升训练效果。公开数据集上的实验结果证明了该方法的有效性。
In SAR image target detection based on deep learning,in order to reduce the interference of noise and other characteristics in SAR images on feature learning and improve the interactivity of identifying and locating tasks in detection methods,a joint task detection method is adopted.The method makes use of the joint task network,so that the identification and location tasks share features as much as possible while retaining their own particularities,thus improving the supervision ability of the two types of tasks on feature learning.In addition,the method also uses the joint task learning method,and considers the reliability of identifying and locating tasks in the selection of anchor frame and the calculation of loss function,thus improving the training effect.The experimental results on public dataset prove the effectiveness of the method.
作者
艾淑芳
田壮壮
王坤
李琳
AI Shufang;TIAN Zhuangzhuang;WANG Kun;LI Lin(Science and Technology on Electro-Optic Control Laboratory,Luoyang 471000,China;State Key Laboratory of Complex Electromagnetic Environment Effects on Electronics and Information System,Luoyang 471000,China;Luoyang Institute of Science and Technology,Luoyang 471000,China)
出处
《电光与控制》
CSCD
北大核心
2023年第5期39-43,83,共6页
Electronics Optics & Control
