A Vision Transformer with Dynamic Masking and Cross-Modal Semantic Learning for Remote Sensing Scene Classification

Ni, Feng; Liu, Yi; Dai, Shibo; Chen, Lei; Feng, Changlei; Zhang, Fan; Wu, Xiang; Bo, Yuming

doi:10.3390/rs18162710

Open AccessArticle

A Vision Transformer with Dynamic Masking and Cross-Modal Semantic Learning for Remote Sensing Scene Classification

by

Feng Ni

^1,2,

Yi Liu

³,

Shibo Dai

¹,

Lei Chen

²,

Changlei Feng

²,

Fan Zhang

²,

Xiang Wu

^1,*

and

Yuming Bo

¹

School of Automation, Nanjing University of Science and Technology, Nanjing 210094, China

²

Beijing Microelectronics Technology Institute, Beijing 100029, China

³

National Key Laboratory of Information Systems Engineering, Nanjing Research Institute of Electronic Engineering, Nanjing 210007, China

^*

Author to whom correspondence should be addressed.

Remote Sens. 2026, 18(16), 2710; https://doi.org/10.3390/rs18162710

Submission received: 29 June 2026 / Revised: 23 July 2026 / Accepted: 31 July 2026 / Published: 12 August 2026

Download

Browse Figures

Versions Notes

Abstract

Remote sensing scene classification plays a vital role in various Earth observation applications. Although supervised learning remains the dominant paradigm, vast quantities of unlabeled imagery remain significantly underutilized. To leverage these unlabeled resources and enhance categorization accuracy, we propose a novel framework based on the Vision Transformer (ViT) that integrates a dynamic masking strategy with a cross-modal semantic learning mechanism. Specifically, a dynamic masking strategy guided by a smooth reconstruction loss is designed to learn robust feature representations from unlabeled samples prior to downstream fine-tuning. Furthermore, we incorporate cross-modal learning to enrich semantic information, thereby addressing the inherent supervisory limitations of conventional one-hot labels. Comprehensive experiments demonstrate that the proposed method significantly improves classification accuracy while maintaining high pre-training efficiency and strong generalization capabilities.

Keywords: remote sensing image; scene classification; transformer; masked autoencoder; cross-modal semantic learning

Share and Cite

MDPI and ACS Style

Ni, F.; Liu, Y.; Dai, S.; Chen, L.; Feng, C.; Zhang, F.; Wu, X.; Bo, Y. A Vision Transformer with Dynamic Masking and Cross-Modal Semantic Learning for Remote Sensing Scene Classification. Remote Sens. 2026, 18, 2710. https://doi.org/10.3390/rs18162710

AMA Style

Ni F, Liu Y, Dai S, Chen L, Feng C, Zhang F, Wu X, Bo Y. A Vision Transformer with Dynamic Masking and Cross-Modal Semantic Learning for Remote Sensing Scene Classification. Remote Sensing. 2026; 18(16):2710. https://doi.org/10.3390/rs18162710

Chicago/Turabian Style

Ni, Feng, Yi Liu, Shibo Dai, Lei Chen, Changlei Feng, Fan Zhang, Xiang Wu, and Yuming Bo. 2026. "A Vision Transformer with Dynamic Masking and Cross-Modal Semantic Learning for Remote Sensing Scene Classification" Remote Sensing 18, no. 16: 2710. https://doi.org/10.3390/rs18162710

APA Style

Ni, F., Liu, Y., Dai, S., Chen, L., Feng, C., Zhang, F., Wu, X., & Bo, Y. (2026). A Vision Transformer with Dynamic Masking and Cross-Modal Semantic Learning for Remote Sensing Scene Classification. Remote Sensing, 18(16), 2710. https://doi.org/10.3390/rs18162710

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Menu

A Vision Transformer with Dynamic Masking and Cross-Modal Semantic Learning for Remote Sensing Scene Classification

Abstract

Share and Cite

Article Metrics

Article Access Statistics

Further Information

Guidelines

MDPI Initiatives

Follow MDPI