Mamba-enhanced RT-DETR for ship detection in optical remote sensing images
Abstract
Ship detection from optical remote sensing images remains challenging due to cluttered sea–land backgrounds, large scale variations, low-contrast small targets, and dense ship occlusion. Existing detectors struggle to balance global modeling capability, detection precision, and computational efficiency. This paper proposes a Mamba-enhanced Real-Time Detection Transformer (RT-DETR) framework for high-precision ship detection and identification. First, a Mamba-based Deformable Spatial Attention (Mamba-DSA) backbone is proposed to integrate linear-time state-space feature modeling and deformable spatial attention, which strengthens ship-specific feature representation and suppresses background interference. Second, a Context-Guided Spatial Feature Reconstruction Feature Pyramid Network (CGRFPN) is designed to enhance multi-scale context awareness and rectangular region calibration, effectively reducing missed detections in dense ship scenes. Third, a Spatial-to-Depth Depthwise Convolution (SPD-DWConv) module is developed to enhance small-target features without introducing excessive computation overhead. Experiments on HRSC2016, NWPU VHR-10, and DOTA-v1.5 datasets show that the proposed method outperforms the baseline RT-DETR by 9.0% in precision, 16.5% in recall, 19.0% in mAP@50, and 14.1% in mAP@50:95. The proposed method achieves strong robustness and efficiency, showing high practical value for maritime monitoring, port security, and emergency response.
Related
- DEMANet: A dehazing enhanced multi-branch attention network for remote sensing images
- iSL-YOLOv11: global–local attention enhanced small lesion detection in mammography images
- WTe2 Weyl semimetal conical microarray enabled quantum-enhanced terahertz photocurrent sensing for organophosphorus toxin detection in simulated body fluid
- Photon Absorption Remote Sensing Virtual Histopathology: A Preliminary Exploration of Diagnostic Equivalence to Gold-Standard H&E Staining in Skin Cancer Excisional Biopsies
- Spatially-adaptive dynamic dictionary learning for fine-grained remote sensing image segmentation
BibTeX
@article{zhao2026mambaenhanced,
title = {{Mamba-enhanced RT-DETR for ship detection in optical remote sensing images}},
author = {Qichang Zhao and Dandan Zhi and Yiquan Wu},
journal = {Scientific Reports},
year = {2026},
month = oct,
doi = {10.1038/s41598-026-64122-6},
url = {https://www.nature.com/articles/s41598-026-64122-6},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - JOUR
TI - Mamba-enhanced RT-DETR for ship detection in optical remote sensing images
AU - Qichang Zhao
AU - Dandan Zhi
AU - Yiquan Wu
PY - 2026
DA - 2026/10/21
JO - Scientific Reports
DO - 10.1038/s41598-026-64122-6
UR - https://www.nature.com/articles/s41598-026-64122-6
AB - Ship detection from optical remote sensing images remains challenging due to cluttered sea–land backgrounds, large scale variations, low-contrast small targets, and dense ship occlusion. Existing detectors struggle to balance global modeling capability, detection precision, and computational efficiency. This paper proposes a Mamba-enhanced Real-Time Detection Transformer (RT-DETR) framework for high-precision ship detection and identification. First, a Mamba-based Deformable Spatial Attention (Mamba-DSA) backbone is proposed to integrate linear-time state-space feature modeling and deformable spatial attention, which strengthens ship-specific feature representation and suppresses background interference. Second, a Context-Guided Spatial Feature Reconstruction Feature Pyramid Network (CGRFPN) is designed to enhance multi-scale context awareness and rectangular region calibration, effectively reducing missed detections in dense ship scenes. Third, a Spatial-to-Depth Depthwise Convolution (SPD-DWConv) module is developed to enhance small-target features without introducing excessive computation overhead. Experiments on HRSC2016, NWPU VHR-10, and DOTA-v1.5 datasets show that the proposed method outperforms the baseline RT-DETR by 9.0% in precision, 16.5% in recall, 19.0% in mAP@50, and 14.1% in mAP@50:95. The proposed method achieves strong robustness and efficiency, showing high practical value for maritime monitoring, port security, and emergency response.
SN - 2045-2322
ER -