改进条件对抗域适应网络的弱磁信号分类研究  

Research on the classification of weak magnetic signals in improved conditional adversarial domain adaptation networks

作　　者：林玲 付世沫 王耀力 常青 LIN Ling;FU Shimo;WANG Yaoli;CHANG Qing(School of Information and Computer,Taiyuan University of Technology,Taiyuan 030024,China;Taiyuan Water Supply Design and Research Institute Co.,Ltd.,Taiyuan 030024,China)

机构地区：[1]太原理工大学信息与计算机学院,山西太原030024 [2]太原供水设计研究院有限公司,山西太原030024

出　　处：《电子设计工程》2024年第13期190-195,共6页Electronic Design Engineering

基　　金：山西省重点研发项目(201903D321003);太原供水设计研究院有限公司项目(RH2000005391)。

摘　　要：针对小目标铁磁体弱磁场序列图像数据获取困难的小样本问题,采用基于稠密光流法与关联特征对齐的条件对抗域适应网络,解决缺乏标签数据的弱磁信号分类检测任务。算法引入稠密光流法学习磁场序列图像之间的动态特征,并将其与原始图像得到的静态特征相结合,提高分类器对纹理相似图像的鉴别能力;同时引入关联特征对齐模块,减少源域与目标域的特征分布差异。网络模型基于近场磁场分布图像与远场磁场分布图像进行验证评估,与原始算法相比,在有无添加熵条件两种情况下准确率分别提升了5.50%和4.50%。Aiming at the problem that it is difficult to obtain the weak magnetic field sequence image data of small target ferromagnets,the conditional adversarial domain adaptation network based on the dense optical flow method aligned with the associated features is used to solve the weak magnetic signal classification detection task lacking label data.The algorithm introduces the dense optical flow method to learn the dynamic features between magnetic field sequence images and combines them with the static features obtained from the original images to improve the discriminative ability of the classifier for texture similar images;at the same time,the correlated feature alignment module is introduced to reduce the difference of feature distribution between the source domain and the target domain.The network model is validated and evaluated based on the near-field magnetic field distribution images and the far-field magnetic field distribution images,and the accuracy is improved by 5.50%and 4.50%,respectively,compared with the original algorithm in both cases with and without the added entropy condition.

关 键 词：小目标铁磁体 序列图像 对抗域适应 稠密光流法 关联特征对齐 

分 类 号：TN91[电子电信—通信与信息系统]