A Center-Masked Transformer for Hyperspectral Image Classification

被引:16
作者
Jia, Sen [1 ,2 ]
Wang, Yifan [1 ,2 ]
Jiang, Shuguo [3 ]
He, Ruyan [1 ,2 ]
机构
[1] Shenzhen Univ, Coll Comp Sci & Software Engn, Guangdong Hong Kong Macau Joint Lab Smart Cities, Minist Nat Resources, Shenzhen 518060, Peoples R China
[2] Shenzhen Univ, Key Lab Geoenvironm Monitoring Coastal Zone, Minist Nat Resources, Shenzhen 518060, Peoples R China
[3] Wuhan Univ, Sch Comp Sci, Wuhan 430071, Peoples R China
来源
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING | 2024年 / 62卷
基金
中国国家自然科学基金;
关键词
Convolutional transformer; deep learning (DL); hyperspectral image (HSI) classification; mask autoencoder; SPATIAL CLASSIFICATION; NEURAL-NETWORKS; REPRESENTATION; ATTENTION;
D O I
10.1109/TGRS.2024.3369075
中图分类号
P3 [地球物理学]; P59 [地球化学];
学科分类号
0708 ; 070902 ;
摘要
Convolutional neural networks (CNNs) are widely used in hyperspectral image (HSI) classification. However, the fixed receptive field of CNN-based methods limits their capability to extract global features. In recent years, transformer has been introduced into networks to tackle this limitation, but it brings other challenges, including a significant increase in model size, the number of labeled training samples required, and the limited effectiveness of sample encoding-reconstruction pretraining methods for HSI classification. To address these issues, a center-masked transformer (CMT) approach is proposed to improve the HSI classification accuracy from two perspectives. On one hand, a local-to-global token embedding (L2GTE) framework coupled with a multiscale convolutional token embedding (MCTE) module is used, which is well-designed to obtain local and global embedding tokens. This effectively reduces the number of model parameters. On the other hand, a regularized center-masked pretraining (RCPT) task is proposed and first introduced into the transformer-based network, which enables the network to learn the dependencies between central ground objects and neighboring objects without labels during the pretraining process. The experimental results conducted on five public HSI datasets demonstrate that our CMT approach outperforms other state-of-the-art methods for HSI classification when training samples are insufficient.
引用
收藏
页码:1 / 16
页数:16
相关论文
共 69 条
[1]  
Bao H., 2021, arXiv
[2]  
Basha SHS, 2018, I C CONT AUTOMAT ROB, P1222, DOI 10.1109/ICARCV.2018.8581147
[3]   Classification of hyperspectral data from urban areas based on extended morphological profiles [J].
Benediktsson, JA ;
Palmason, JA ;
Sveinsson, JR .
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, 2005, 43 (03) :480-491
[4]  
Chen Ting., 2020, Advances in neural information processing systems, V33, P22243
[5]  
Chen Ting, 2019, 25 AMERICAS C INFO
[6]  
Chen XL, 2021, Arxiv, DOI arXiv:2104.02057
[7]  
Chen XL, 2020, Arxiv, DOI arXiv:2003.04297
[8]   Spectral-Spatial Classification of Hyperspectral Data Based on Deep Belief Network [J].
Chen, Yushi ;
Zhao, Xing ;
Jia, Xiuping .
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, 2015, 8 (06) :2381-2392
[9]   Deep Learning-Based Classification of Hyperspectral Data [J].
Chen, Yushi ;
Lin, Zhouhan ;
Zhao, Xing ;
Wang, Gang ;
Gu, Yanfeng .
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, 2014, 7 (06) :2094-2107
[10]   Classification of hyperspectral remote-sensing data with primal SVM for small-sized training dataset problem [J].
Chi, Mingmin ;
Feng, Rui ;
Bruzzone, Lorenzo .
ADVANCES IN SPACE RESEARCH, 2008, 41 (11) :1793-1799