Dual-Stream Spatially Aware Transformer for Remote Sensing Image Captioning

Remote sensing image captioning (RSIC) aims to generate semantically rich and syntactically accurate descriptions for remote sensing images (RSIs). However, due to the complex spatial layouts, occlusions, and overlapping objects in such images, caption generation is often challenged by semantic ambi...

Full description

Saved in:
Bibliographic Details
Main Authors: Haifeng Sima, Xiangtao Ding, JianLong Wang, Mingliang Xu
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Subjects:
Online Access:https://ieeexplore.ieee.org/document/11104798/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849404453632868352
author Haifeng Sima
Xiangtao Ding
JianLong Wang
Mingliang Xu
author_facet Haifeng Sima
Xiangtao Ding
JianLong Wang
Mingliang Xu
author_sort Haifeng Sima
collection DOAJ
description Remote sensing image captioning (RSIC) aims to generate semantically rich and syntactically accurate descriptions for remote sensing images (RSIs). However, due to the complex spatial layouts, occlusions, and overlapping objects in such images, caption generation is often challenged by semantic ambiguity. To address these issues, we propose a novel <italic>dual-stream spatially aware transformer</italic> (DSAT), which explicitly models both global and local spatial relationships to enhance spatial understanding. Specifically, DSAT introduces a <italic>dual-stream feature interaction</italic> module that extracts grid-level global features and region-level object features, and further enhances their respective spatial dependencies through multibranch convolution and a graph attention network. In addition, we design a spatially aware attention mechanism that encodes relative spatial relationships into the Transformer, allowing the model to better capture object distribution patterns and geometric relationships. Extensive experiments conducted on three benchmark datasets, namely Sydney-Captions, UCM-Captions, and remote sensing image description, highlight the superior performance of DSAT. The proposed method achieves impressive CIDEr scores of 338.59%, 450.93%, and 275.36% on these datasets, respectively, demonstrating its effectiveness in generating high-quality captions for RSIs.
format Article
id doaj-art-b8361c0fa2b54cd69471e3176c0f74da
institution Kabale University
issn 1939-1404
2151-1535
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
spelling doaj-art-b8361c0fa2b54cd69471e3176c0f74da2025-08-20T03:36:58ZengIEEEIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing1939-14042151-15352025-01-0118195461956210.1109/JSTARS.2025.359388711104798Dual-Stream Spatially Aware Transformer for Remote Sensing Image CaptioningHaifeng Sima0https://orcid.org/0000-0002-2049-3637Xiangtao Ding1https://orcid.org/0009-0001-6680-5608JianLong Wang2https://orcid.org/0000-0001-8117-0631Mingliang Xu3https://orcid.org/0000-0002-6885-3451School of Computer Science and Technology &amp; School of Software, Henan Polytechnic University, Jiaozuo, ChinaSchool of Computer Science and Technology &amp; School of Software, Henan Polytechnic University, Jiaozuo, ChinaSchool of Computer Science and Technology &amp; School of Software, Henan Polytechnic University, Jiaozuo, ChinaSchool of Computer and Artificial Intelligence, Zhengzhou University, Zhengzhou, ChinaRemote sensing image captioning (RSIC) aims to generate semantically rich and syntactically accurate descriptions for remote sensing images (RSIs). However, due to the complex spatial layouts, occlusions, and overlapping objects in such images, caption generation is often challenged by semantic ambiguity. To address these issues, we propose a novel <italic>dual-stream spatially aware transformer</italic> (DSAT), which explicitly models both global and local spatial relationships to enhance spatial understanding. Specifically, DSAT introduces a <italic>dual-stream feature interaction</italic> module that extracts grid-level global features and region-level object features, and further enhances their respective spatial dependencies through multibranch convolution and a graph attention network. In addition, we design a spatially aware attention mechanism that encodes relative spatial relationships into the Transformer, allowing the model to better capture object distribution patterns and geometric relationships. Extensive experiments conducted on three benchmark datasets, namely Sydney-Captions, UCM-Captions, and remote sensing image description, highlight the superior performance of DSAT. The proposed method achieves impressive CIDEr scores of 338.59%, 450.93%, and 275.36% on these datasets, respectively, demonstrating its effectiveness in generating high-quality captions for RSIs.https://ieeexplore.ieee.org/document/11104798/Image captioningremote sensing image captioning (RSIC)spatial-aware information (SAI)transformer
spellingShingle Haifeng Sima
Xiangtao Ding
JianLong Wang
Mingliang Xu
Dual-Stream Spatially Aware Transformer for Remote Sensing Image Captioning
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Image captioning
remote sensing image captioning (RSIC)
spatial-aware information (SAI)
transformer
title Dual-Stream Spatially Aware Transformer for Remote Sensing Image Captioning
title_full Dual-Stream Spatially Aware Transformer for Remote Sensing Image Captioning
title_fullStr Dual-Stream Spatially Aware Transformer for Remote Sensing Image Captioning
title_full_unstemmed Dual-Stream Spatially Aware Transformer for Remote Sensing Image Captioning
title_short Dual-Stream Spatially Aware Transformer for Remote Sensing Image Captioning
title_sort dual stream spatially aware transformer for remote sensing image captioning
topic Image captioning
remote sensing image captioning (RSIC)
spatial-aware information (SAI)
transformer
url https://ieeexplore.ieee.org/document/11104798/
work_keys_str_mv AT haifengsima dualstreamspatiallyawaretransformerforremotesensingimagecaptioning
AT xiangtaoding dualstreamspatiallyawaretransformerforremotesensingimagecaptioning
AT jianlongwang dualstreamspatiallyawaretransformerforremotesensingimagecaptioning
AT mingliangxu dualstreamspatiallyawaretransformerforremotesensingimagecaptioning