Enhanced-RSMamba: State space model for semantic segmentation of remote sensing images

Kangning Wang; Xiao Wang; Yongfei Xian; Zhiguo Jiang; Haopeng Zhang

doi:10.1117/12.3069876

Enhanced-RSMamba: State space model for semantic segmentation of remote sensing images

Kangning Wang
, Xiao Wang
, Yongfei Xian
, Zhiguo Jiang
, Haopeng Zhang^*

^*Corresponding author for this work

School of Astronautics

Beihang University
Key Laboratory of Precision Opto-Mechatronics Technology (Ministry of Education)

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

Semantic segmentation of remote sensing images is vital for land use analysis and environmental monitoring but faces challenges in capturing long-range dependencies and multiscale features in high-resolution scenes. We propose an optimized MambaVision architecture that integrates 2D selective scanning with adaptive dilated convolutions to effectively capture local and global features. Our novel Multi-Frequency Multi-Scale (MFMS) decoder fuses frequency-domain features, derived from Discrete Cosine Transform, with multiscale spatial information and orientation-aware attention, improving texture and boundary detection. As the first application of MambaVision to remote sensing segmentation, our model achieves an mIoU of 84.24% on the ISPRS Potsdam dataset, a 6.57% improvement over baselines. Ablation studies confirm the effectiveness of 2D scanning and orientation attention, making our approach efficient for resource-constrained applications.

Original language	English
Title of host publication	Artificial Intelligence and Image and Signal Processing for Remote Sensing XXXI
Editors	Lorenzo Bruzzone, Francesca Bovolo, Fabio Bovenga
Publisher	SPIE
ISBN (Electronic)	9781510692794
DOIs	https://doi.org/10.1117/12.3069876
State	Published - 29 Oct 2025
Event	31st Artificial Intelligence and Image and Signal Processing for Remote Sensing - Madrid, Spain Duration: 15 Sep 2025 → 17 Sep 2025

Publication series

Name	Proceedings of SPIE - The International Society for Optical Engineering
Volume	13670
ISSN (Print)	0277-786X
ISSN (Electronic)	1996-756X

Conference

Conference	31st Artificial Intelligence and Image and Signal Processing for Remote Sensing
Country/Territory	Spain
City	Madrid
Period	15/09/25 → 17/09/25

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 15 Life on Land

Keywords

MambaVision
Multi-Frequency Multi-Scale Decoder
Orientation Attention
Remote Sensing Image Semantic Segmentation
State Space Model

Access to Document

10.1117/12.3069876

Cite this

Wang, K., Wang, X., Xian, Y., Jiang, Z., & Zhang, H. (2025). Enhanced-RSMamba: State space model for semantic segmentation of remote sensing images. In L. Bruzzone, F. Bovolo, & F. Bovenga (Eds.), Artificial Intelligence and Image and Signal Processing for Remote Sensing XXXI Article 136700B (Proceedings of SPIE - The International Society for Optical Engineering; Vol. 13670). SPIE. https://doi.org/10.1117/12.3069876

Wang, Kangning ; Wang, Xiao ; Xian, Yongfei et al. / Enhanced-RSMamba : State space model for semantic segmentation of remote sensing images. Artificial Intelligence and Image and Signal Processing for Remote Sensing XXXI. editor / Lorenzo Bruzzone ; Francesca Bovolo ; Fabio Bovenga. SPIE, 2025. (Proceedings of SPIE - The International Society for Optical Engineering).

@inproceedings{7b739e7abd5e405d9acab7ebadf34db8,

title = "Enhanced-RSMamba: State space model for semantic segmentation of remote sensing images",

abstract = "Semantic segmentation of remote sensing images is vital for land use analysis and environmental monitoring but faces challenges in capturing long-range dependencies and multiscale features in high-resolution scenes. We propose an optimized MambaVision architecture that integrates 2D selective scanning with adaptive dilated convolutions to effectively capture local and global features. Our novel Multi-Frequency Multi-Scale (MFMS) decoder fuses frequency-domain features, derived from Discrete Cosine Transform, with multiscale spatial information and orientation-aware attention, improving texture and boundary detection. As the first application of MambaVision to remote sensing segmentation, our model achieves an mIoU of 84.24\% on the ISPRS Potsdam dataset, a 6.57\% improvement over baselines. Ablation studies confirm the effectiveness of 2D scanning and orientation attention, making our approach efficient for resource-constrained applications.",

keywords = "MambaVision, Multi-Frequency Multi-Scale Decoder, Orientation Attention, Remote Sensing Image Semantic Segmentation, State Space Model",

author = "Kangning Wang and Xiao Wang and Yongfei Xian and Zhiguo Jiang and Haopeng Zhang",

note = "Publisher Copyright: {\textcopyright} COPYRIGHT SPIE. Downloading of the abstract is permitted for personal use only.; 31st Artificial Intelligence and Image and Signal Processing for Remote Sensing ; Conference date: 15-09-2025 Through 17-09-2025",

year = "2025",

month = oct,

day = "29",

doi = "10.1117/12.3069876",

language = "英语",

series = "Proceedings of SPIE - The International Society for Optical Engineering",

publisher = "SPIE",

editor = "Lorenzo Bruzzone and Francesca Bovolo and Fabio Bovenga",

booktitle = "Artificial Intelligence and Image and Signal Processing for Remote Sensing XXXI",

address = "美国",

}

Wang, K, Wang, X, Xian, Y, Jiang, Z & Zhang, H 2025, Enhanced-RSMamba: State space model for semantic segmentation of remote sensing images. in L Bruzzone, F Bovolo & F Bovenga (eds), Artificial Intelligence and Image and Signal Processing for Remote Sensing XXXI., 136700B, Proceedings of SPIE - The International Society for Optical Engineering, vol. 13670, SPIE, 31st Artificial Intelligence and Image and Signal Processing for Remote Sensing, Madrid, Spain, 15/09/25. https://doi.org/10.1117/12.3069876

Enhanced-RSMamba: State space model for semantic segmentation of remote sensing images. / Wang, Kangning; Wang, Xiao; Xian, Yongfei et al.
Artificial Intelligence and Image and Signal Processing for Remote Sensing XXXI. ed. / Lorenzo Bruzzone; Francesca Bovolo; Fabio Bovenga. SPIE, 2025. 136700B (Proceedings of SPIE - The International Society for Optical Engineering; Vol. 13670).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Enhanced-RSMamba

T2 - 31st Artificial Intelligence and Image and Signal Processing for Remote Sensing

AU - Wang, Kangning

AU - Wang, Xiao

AU - Xian, Yongfei

AU - Jiang, Zhiguo

AU - Zhang, Haopeng

PY - 2025/10/29

Y1 - 2025/10/29

N2 - Semantic segmentation of remote sensing images is vital for land use analysis and environmental monitoring but faces challenges in capturing long-range dependencies and multiscale features in high-resolution scenes. We propose an optimized MambaVision architecture that integrates 2D selective scanning with adaptive dilated convolutions to effectively capture local and global features. Our novel Multi-Frequency Multi-Scale (MFMS) decoder fuses frequency-domain features, derived from Discrete Cosine Transform, with multiscale spatial information and orientation-aware attention, improving texture and boundary detection. As the first application of MambaVision to remote sensing segmentation, our model achieves an mIoU of 84.24% on the ISPRS Potsdam dataset, a 6.57% improvement over baselines. Ablation studies confirm the effectiveness of 2D scanning and orientation attention, making our approach efficient for resource-constrained applications.

AB - Semantic segmentation of remote sensing images is vital for land use analysis and environmental monitoring but faces challenges in capturing long-range dependencies and multiscale features in high-resolution scenes. We propose an optimized MambaVision architecture that integrates 2D selective scanning with adaptive dilated convolutions to effectively capture local and global features. Our novel Multi-Frequency Multi-Scale (MFMS) decoder fuses frequency-domain features, derived from Discrete Cosine Transform, with multiscale spatial information and orientation-aware attention, improving texture and boundary detection. As the first application of MambaVision to remote sensing segmentation, our model achieves an mIoU of 84.24% on the ISPRS Potsdam dataset, a 6.57% improvement over baselines. Ablation studies confirm the effectiveness of 2D scanning and orientation attention, making our approach efficient for resource-constrained applications.

KW - MambaVision

KW - Multi-Frequency Multi-Scale Decoder

KW - Orientation Attention

KW - Remote Sensing Image Semantic Segmentation

KW - State Space Model

UR - https://www.scopus.com/pages/publications/105024871092

U2 - 10.1117/12.3069876

DO - 10.1117/12.3069876

M3 - 会议稿件

AN - SCOPUS:105024871092

T3 - Proceedings of SPIE - The International Society for Optical Engineering

BT - Artificial Intelligence and Image and Signal Processing for Remote Sensing XXXI

A2 - Bruzzone, Lorenzo

A2 - Bovolo, Francesca

A2 - Bovenga, Fabio

PB - SPIE

Y2 - 15 September 2025 through 17 September 2025

ER -

Wang K, Wang X, Xian Y, Jiang Z , Zhang H. Enhanced-RSMamba: State space model for semantic segmentation of remote sensing images. In Bruzzone L, Bovolo F, Bovenga F, editors, Artificial Intelligence and Image and Signal Processing for Remote Sensing XXXI. SPIE. 2025. 136700B. (Proceedings of SPIE - The International Society for Optical Engineering). doi: 10.1117/12.3069876

Enhanced-RSMamba: State space model for semantic segmentation of remote sensing images

Abstract

Publication series

Conference

UN SDGs

Keywords

Access to Document

Other files and links

Fingerprint

Cite this