Few Shot Object Detection for SAR Images via Feature Enhancement and Dynamic Relationship Modeling

被引:14
作者
Chen, Shiqi [1 ]
Zhang, Jun [1 ]
Zhan, Ronghui [1 ]
Zhu, Rongqiang [2 ]
Wang, Wei [1 ]
机构
[1] Natl Univ Def Technol, NationalKey Lab Sci & Technol Automat Target Reco, Changsha 410073, Peoples R China
[2] Southwest Elect & Telecommun Technol Res Inst, Chengdu 610000, Peoples R China
基金
中国国家自然科学基金;
关键词
Synthetic Aperture Radar (SAR); few-shot object detection; attention; support-guided module; graph convolutional network;
D O I
10.3390/rs14153669
中图分类号
X [环境科学、安全科学];
学科分类号
08 ; 0830 ;
摘要
Current Synthetic Aperture Radar (SAR) image object detection methods require huge amounts of annotated data and can only detect the categories that appears in the training set. Due to the lack of training samples in the real applications, the performance decreases sharply on rare categories, which largely inhibits the detection model from reaching robustness. To tackle this problem, a novel few-shot SAR object detection framework is proposed, which is built upon the meta-learning architecture and aims at detecting objects of unseen classes given only a few annotated examples. Observing the quality of support features determines the performance of the few-shot object detection task, we propose an attention mechanism to highlight class-specific features while softening the irrelevant background information. Considering the variation between different support images, we also employ a support-guided module to enhance query features, thus generating high-qualified proposals more relevant to support images. To further exploit the relevance between support and query images, which is ignored in single class representation, a dynamic relationship learning paradigm is designed via constructing a graph convolutional network and imposing orthogonality constraint in hidden feature space, which both make features from the same category more closer and those from different classes more separable. Comprehensive experiments have been completed on the self-constructed SAR multi-class object detection dataset, which demonstrate the effectiveness of our few-shot object detection framework in learning more generalized features to both enhance the performance on novel classes and maintain the performance on base classes.
引用
收藏
页数:21
相关论文
共 43 条
[1]   Detection and Discrimination of Ship Targets in Complex Background From Spaceborne ALOS-2 SAR Images [J].
Ao, Wei ;
Xu, Feng ;
Li, Yongchen ;
Wang, Haipeng .
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, 2018, 11 (02) :536-550
[2]  
Bochkovskiy A., 2020, PREPRINT
[3]   Learning Slimming SAR Ship Object Detector Through Network Pruning and Knowledge Distillation [J].
Chen, Shiqi ;
Zhan, Ronghui ;
Wang, Wei ;
Zhang, Jun .
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, 2021, 14 :1267-1282
[4]   Dynamic Convolution: Attention over Convolution Kernels [J].
Chen, Yinpeng ;
Dai, Xiyang ;
Liu, Mengchen ;
Chen, Dongdong ;
Yuan, Lu ;
Liu, Zicheng .
2020 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2020), 2020, :11027-11036
[5]   Prototype-CNN for Few-Shot Object Detection in Remote Sensing Images [J].
Cheng, Gong ;
Yan, Bowei ;
Shi, Peizhen ;
Li, Ke ;
Yao, Xiwen ;
Guo, Lei ;
Han, Junwei .
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, 2022, 60
[6]   Ship Detection in Large-Scale SAR Images Via Spatial Shuffle-Group Enhance Attention [J].
Cui, Zongyong ;
Wang, Xiaoya ;
Liu, Nengyuan ;
Cao, Zongjie ;
Yang, Jianyu .
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, 2021, 59 (01) :379-391
[7]   Few-Shot Object Detection with Attention-RPN and Multi-Relation Detector [J].
Fan, Qi ;
Zhuo, Wei ;
Tang, Chi-Keung ;
Tai, Yu-Wing .
2020 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), 2020, :4012-4021
[8]   Few-Shot SAR Target Classification via Metalearning [J].
Fu, Kun ;
Zhang, Tengfei ;
Zhang, Yue ;
Wang, Zhirui ;
Sun, Xian .
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, 2022, 60
[9]   FUSAR-Ship: building a high-resolution SAR-AIS matchup dataset of Gaofen-3 for ship detection and recognition [J].
Hou, Xiyue ;
Ao, Wei ;
Song, Qian ;
Lai, Jian ;
Wang, Haipeng ;
Xu, Feng .
SCIENCE CHINA-INFORMATION SCIENCES, 2020, 63 (04)
[10]  
Hu J, 2018, PROC CVPR IEEE, P7132, DOI [10.1109/CVPR.2018.00745, 10.1109/TPAMI.2019.2913372]