Hybrid Conv-Attention Networks for Synthetic Aperture Radar Imagery-Based Target Recognition

被引：1
作者：
Yoon, Jiseok [1 ]
Song, Jeongheon [2 ]
Hussain, Tanveer [3 ]
Khowaja, Sunder Ali [4 ]
Muhammad, Khan [5 ]
Lee, Ik Hyun [1 ,6 ]
机构：
[1] IKLAB Inc, Siheung Si 15073, South Korea
[2] Korea Aerosp Res Inst, Daejeon 34133, South Korea
[3] Edge Hill Univ, Dept Comp Sci, Ormskirk L39 4QP, England
[4] Univ Sindh, Dept Telecommun Engn, Jamshoro 76080, Pakistan
[5] Sungkyunkwan Univ, Dept Appl Artificial Intelligence, Seoul 03063, South Korea
[6] Tech Univ Korea, Dept Mechatron Engn, Siheung Si 15073, South Korea
来源：
IEEE ACCESS | 2024年 / 12卷
基金：
新加坡国家研究基金会;
关键词：
Radar polarimetry; Feature extraction; Target recognition; Data models; Adaptation models; Transformers; Synthetic aperture radar; Deep learning; Convolutional neural networks; Synthetic aperture radar (SAR); target recognition; deep learning (DL); transfer learning; convolutional neural networks (CNNs); transformers; SAR; CLASSIFICATION; DATASET;
D O I：
10.1109/ACCESS.2024.3387314
中图分类号：
TP [自动化技术、计算机技术];
学科分类号：
0812 ;
摘要：
In this study, we propose hybrid conv-attention networks that combine convolutional neural networks (CNNs) and transformers to recognize targets from synthetic aperture radar (SAR) images automatically. The proposed model is designed to obtain robust features from global and local patterns in the SAR image, utilizing the weights of a pre-trained backbone model with self-attention structures. Furthermore, we adopted pre-processing and training methods optimized for transfer learning to enhance performance. By comparing and analyzing the performance between the proposed model and conventional models using the OpenSARShip and MSTAR dataset, we found that our system significantly outperforms conventional approaches, with a performance improvement of 24.06%. This considerable enhancement is attributed to the ability of the model to leverage the 2D kernel-based approach of CNNs and the sequence vector-based approach of transformers, offering a comprehensive method for SAR image target recognition.
引用
收藏
页码：53045 / 53055
页数：11
相关论文
共 54 条
[1]  
Agarap A F., Deep Learning using Rectified Linear Units, DOI DOI 10.48550/ARXIV.1803.08375
[2]  
Al Mufti M, 2018, 2018 INTERNATIONAL CONFERENCE ON INTELLIGENT AUTONOMOUS SYSTEMS (ICOIAS), P1, DOI 10.1109/ICoIAS.2018.8494149
[3]  
Antoniou A, 2018, Arxiv, DOI [arXiv:1711.04340, 10.48550/arXiv.1711.04340]
[4]  
Ba J L., LAYER NORMALIZATION
[5]   Data Augmentation with Variational Autoencoders and Manifold Sampling [J].
Chadebec, Clement ;
Allassonniere, Stephanie .
DEEP GENERATIVE MODELS, AND DATA AUGMENTATION, LABELLING, AND IMPERFECTIONS, 2021, 13003 :184-192
[6]   Target Classification Using the Deep Convolutional Networks for SAR Images [J].
Chen, Sizhe ;
Wang, Haipeng ;
Xu, Feng ;
Jin, Ya-Qiu .
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, 2016, 54 (08) :4806-4817
[7]  
Dai Z, 2021, ADV NEUR IN, V34
[8]   Amplitude-Phase CNN-Based SAR Target Classification via Complex-Valued Sparse Image [J].
Deng, Jiarui ;
Bi, Hui ;
Zhang, Jingjing ;
Liu, Zehao ;
Yu, Lingjuan .
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, 2022, 15 :5214-5221
[9]  
Dosovitskiy A, 2021, Arxiv, DOI arXiv:2010.11929
[10]  
Dumoulin V, 2018, Arxiv, DOI [arXiv:1603.07285, 10.48550/arXiv.1603.07285, DOI 10.48550/ARXIV.1603.07285]
123456