| 注册
首页|期刊导航|现代雷达|基于显著性检测与地球移动距离的SAR目标小样本识别方法

基于显著性检测与地球移动距离的SAR目标小样本识别方法

杜钊旭 李响 殷君君 杨健

现代雷达2025,Vol.47Issue(11):38-45,8.
现代雷达2025,Vol.47Issue(11):38-45,8.DOI:10.16592/j.cnki.1004-7859.2025070101

基于显著性检测与地球移动距离的SAR目标小样本识别方法

A SAR Few-shot Target Recognition Method Based on Saliency Detection and Earth Mover's Distance

杜钊旭 1李响 2殷君君 1杨健2

作者信息

  • 1. 北京科技大学 计算机与通信工程学院,北京 100083
  • 2. 清华大学 电子工程系,北京 100084
  • 折叠

摘要

Abstract

Synthetic aperture radar(SAR),as an active microwave remote sensing imaging system,offers all-day,all-weather ob-servation capabilities and strong adaptability to complex weather conditions.With the development of deep learning techniques,SAR target recognition based on deep models has become a key research focus in the radar field.However,SAR data are expensive to acquire and require expert knowledge for annotation,resulting in limited labeled samples.This makes few-shot learning(FSL)an important approach to improving SAR application performance.To address the insufficient target-region awareness in feature matching of existing FSL methods,a SAR few-shot target recognition method that integrates saliency detection with the earth mover's distance(EMD)is proposed in this paper.Specifically,the adaptive coordinate cooperation network is introduced to generate sali-ency maps for the images and accordingly construct their spatial weight distributions.These spatial weights are then incorporated as a prior knowledge to guide the image weighting process within EMD computation,thereby enhancing the focus on critical regions during feature transfer in FSL.Experiments on the moving and stationary target acquisition and recognition and OpenSARShip data-sets demonstrate that the proposed method significantly outperforms existing mainstream few-shot recognition approaches under vari-ous experimental settings,verifying the effectiveness and robustness of incorporating saliency priors in feature matching.

关键词

合成孔径雷达/目标识别/小样本学习/显著性检测/度量学习

Key words

synthetic aperture radar(SAR)/target recognition/few-shot learning(FSL)/saliency detection/metric learning

分类

信息技术与安全科学

引用本文复制引用

杜钊旭,李响,殷君君,杨健..基于显著性检测与地球移动距离的SAR目标小样本识别方法[J].现代雷达,2025,47(11):38-45,8.

基金项目

国家重点研发计划资助项目(2024YFB3909800) (2024YFB3909800)

国家自然科学基金资助项目(62222102,62171023) (62222102,62171023)

中央高校基本科研业务费资助项目(FRF-TP-22-005C1) (FRF-TP-22-005C1)

现代雷达

OA北大核心

1004-7859

访问量0
|
下载量0
段落导航相关论文