YOLO-Argus: An instance segmentation model based on scale-aware downsampling and parallel multi-scale edge fusion for SAR ship images

Qiming Li; Hao Yin; Daozheng Chen; Qiming Li; Hao Yin; Daozheng Chen

doi:10.3934/math.2026923

AIMS Mathematics

2026, Volume 11, Issue 7: 22920-22942. doi: 10.3934/math.2026923

Previous Article Next Article

Research article

YOLO-Argus: An instance segmentation model based on scale-aware downsampling and parallel multi-scale edge fusion for SAR ship images

1.
College of Information Engineering, Shanghai Maritime University, Shanghai 201306, China
2.
Informatization Office, Shanghai Maritime University, Shanghai 201306, China

Received: 29 April 2026 Revised: 01 July 2026 Accepted: 17 July 2026 Published: 29 July 2026
MSC : 68T07, 68U10

With the rapid growth of global maritime transportation, accurate ship instance segmentation from SAR images has become a core technology for maritime surveillance and management. However, speckle noise, significant scale variations, and the presence of dense and structurally ambiguous targets in complex nearshore SAR scenes pose substantial challenges to feature extraction, multi-scale representation, and precise boundary delineation in instance segmentation. Therefore, we have proposed YOLO-Argus, an enhanced model for SAR ship instance segmentation. First, a spatial parallel context perception (SPCP) module is introduced to construct a dual-path feature extraction framework using multi-dilated convolutions, thereby improving multi-scale perception, particularly for small targets in complex backgrounds. Second, a multi-scale edge fusion (MSEF) module integrates edge enhancement with contextual semantic information to improve discrimination of blurred boundaries and contour structures. Furthermore, an adaptive tanh unit (ATU) embeds adaptive normalization mechanisms into attention blocks, effectively suppressing speckle noise while preserving fine-grained details. Finally, a multi-scale channel decoupling (MSCD) module leverages grouped and multi-scale convolutions to jointly optimize feature representations, enhancing robustness under noise interference. Experimental results on the public HRSID and PSeg-SSDD datasets quantitatively confirmed the effectiveness of YOLO-Argus. Compared with the baseline model, YOLO-Argus improves $ AP_{50} $ and $ AP_{75} $ by 4.3 and 4.7 percentage points on HRSID, and by 3.5 and 9.2 percentage points on PSeg-SSDD, respectively. The proposed method also achieved superior segmentation accuracy compared with existing state-of-the-art methods, including both general-purpose instance segmentation models and SAR ship-specific approaches. These results indicate that the proposed approach effectively mitigates the interference of noise, scale variation, and background complexity in SAR images, providing a more reliable solution for ship instance segmentation.
- instance segmentation,
- ship targets,
- synthetic aperture radar images,
- YOLOv11,
- multi-scale feature fusion,
- edge enhancement
Citation: Qiming Li, Hao Yin, Daozheng Chen. YOLO-Argus: An instance segmentation model based on scale-aware downsampling and parallel multi-scale edge fusion for SAR ship images[J]. AIMS Mathematics, 2026, 11(7): 22920-22942. doi: 10.3934/math.2026923

Related Papers:

Abstract

With the rapid growth of global maritime transportation, accurate ship instance segmentation from SAR images has become a core technology for maritime surveillance and management. However, speckle noise, significant scale variations, and the presence of dense and structurally ambiguous targets in complex nearshore SAR scenes pose substantial challenges to feature extraction, multi-scale representation, and precise boundary delineation in instance segmentation. Therefore, we have proposed YOLO-Argus, an enhanced model for SAR ship instance segmentation. First, a spatial parallel context perception (SPCP) module is introduced to construct a dual-path feature extraction framework using multi-dilated convolutions, thereby improving multi-scale perception, particularly for small targets in complex backgrounds. Second, a multi-scale edge fusion (MSEF) module integrates edge enhancement with contextual semantic information to improve discrimination of blurred boundaries and contour structures. Furthermore, an adaptive tanh unit (ATU) embeds adaptive normalization mechanisms into attention blocks, effectively suppressing speckle noise while preserving fine-grained details. Finally, a multi-scale channel decoupling (MSCD) module leverages grouped and multi-scale convolutions to jointly optimize feature representations, enhancing robustness under noise interference. Experimental results on the public HRSID and PSeg-SSDD datasets quantitatively confirmed the effectiveness of YOLO-Argus. Compared with the baseline model, YOLO-Argus improves $ AP_{50} $ and $ AP_{75} $ by 4.3 and 4.7 percentage points on HRSID, and by 3.5 and 9.2 percentage points on PSeg-SSDD, respectively. The proposed method also achieved superior segmentation accuracy compared with existing state-of-the-art methods, including both general-purpose instance segmentation models and SAR ship-specific approaches. These results indicate that the proposed approach effectively mitigates the interference of noise, scale variation, and background complexity in SAR images, providing a more reliable solution for ship instance segmentation.

References

[1]	J. Zhang, H. Li, Z. Tang, Q. Lu, X. Zheng, J. Zhou, An improved quantum-inspired genetic algorithm for image multilevel thresholding segmentation, Math. Probl. Eng., 2014 (2014), 295402. https://doi.org/10.1155/2014/295402 doi: 10.1155/2014/295402
[2]	R. Storn, K. Price, Differential evolution – a simple and efficient heuristic for global optimization over continuous spaces, J. Global Optim., 11 (1997), 341–359. https://doi.org/10.1023/A:1008202821328 doi: 10.1023/A:1008202821328
[3]	W. Cao, J. Li, J. Liu, P. Zhang, Two improved segmentation algorithms for whole cardiac CT sequence images, 9th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI), 2016,346–351. https://doi.org/10.1109/cisp-bmei.2016.7852734
[4]	L. Vincent, P. Soille, Watersheds in digital spaces: an efficient algorithm based on immersion simulations, IEEE Trans. Pattern Anal. Mach. Intell., 13 (1991), 583–598. https://doi.org/10.1109/34.87344 doi: 10.1109/34.87344
[5]	M. Kang, C. M. Ting, F. F. Ting, R. C. W. Phan, ASF-YOLO: A novel YOLO model with attentional scale sequence fusion for cell instance segmentation, Image Vis. Comput., 147 (2024), 105057. https://doi.org/10.1016/j.imavis.2024.105057 doi: 10.1016/j.imavis.2024.105057
[6]	Z. Gu, H. Chen, Z. Xu, DiffusionInst: diffusion model for instance segmentation, 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2024, 2730–2734. https://doi.org/10.1109/icassp48485.2024.10447191
[7]	J. Zhao, Y. Wang, Y. Zhou, W. L. Du, R. Yao, A. El Saddik, GLFRNet: global-local feature refusion network for remote sensing image instance segmentation, IEEE Trans. Geosci. Remote Sens., 63 (2025), 1–12. https://doi.org/10.1109/tgrs.2025.3530515 doi: 10.1109/tgrs.2025.3530515
[8]	Y. Liu, H. Li, C. Hu, S. Luo, Y. Luo, C. W. Chen, Learning to aggregate multi-scale context for instance segmentation in remote sensing images, IEEE Trans. Neural Netw. Learn. Syst., 36 (2024), 595–609. https://doi.org/10.1109/tnnls.2023.3336563 doi: 10.1109/tnnls.2023.3336563
[9]	J. Xie, W. Li, X. Li, Z. Liu, Y. S. Ong, C. C. Loy, MosaicFusion: diffusion models as data augmenters for large vocabulary instance segmentation, Int. J. Comput. Vis., 133 (2025), 1456–1475. https://doi.org/10.1007/s11263-024-02223-3 doi: 10.1007/s11263-024-02223-3
[10]	P. Guo, H. Huang, P. He, X. Liu, T. Xiao, W. Zhang, OpenVIS: open-vocabulary video instance segmentation, Proceedings of the AAAI Conference on Artificial Intelligence, 39 (2025), 3275–3283. https://doi.org/10.1609/aaai.v39i3.32338 doi: 10.1609/aaai.v39i3.32338
[11]	J. Zhu, X. Chen, K. He, Y. LeCun, Z. Liu, Transformers without normalization, 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025, 14901–14911. https://doi.org/10.1109/cvpr52734.2025.01388
[12]	T. Wu, S. Tang, R. Zhang, J. Cao, Y. Zhang, CGNet: a light-weight context guided network for semantic segmentation, IEEE Trans. Image Process., 30 (2021), 1169–1179. https://doi.org/10.1109/tip.2020.3042065 doi: 10.1109/tip.2020.3042065
[13]	M. Chen, T. Wang, C. Xu, J. Chen, E. Chen, Z. Pan, Gradient prior guidance and image adaptation enhancement for semi-supervised SAR ship instance segmentation, IEEE Sensors J., 24 (2024), 36216–36229. https://doi.org/10.1109/jsen.2024.3467030 doi: 10.1109/jsen.2024.3467030
[14]	M. C. El Rai, J. H. Giraldo, M. Al-Saad, M. Darweech, T. Bouwmans, SemiSegSAR: a semi-supervised segmentation algorithm for ship SAR images, IEEE Geosci. Remote Sens. Lett., 19 (2022), 1–5. https://doi.org/10.1109/lgrs.2022.3185306 doi: 10.1109/lgrs.2022.3185306
[15]	F. Gao, Y. Huo, J. Wang, A. Hussain, H. Zhou, Anchor-free SAR ship instance segmentation with centroid-distance based loss, IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., 14 (2021), 11352–11371. https://doi.org/10.1109/jstars.2021.3123784 doi: 10.1109/jstars.2021.3123784
[16]	Z. Sun, C. Meng, J. Cheng, Z. Zhang, S. Chang, A multi-scale feature pyramid network for detection and instance segmentation of marine ships in SAR images, Remote Sens., 14 (2022), 6312. https://doi.org/10.3390/rs14246312 doi: 10.3390/rs14246312
[17]	X. Ke, X. Zhang, T. Zhang, GCBANet: a global context boundary-aware network for SAR ship instance segmentation, Remote Sens., 14 (2022), 2165. https://doi.org/10.3390/rs14092165 doi: 10.3390/rs14092165
[18]	S. Wei, X. Zeng, H. Zhang, Z. Zhou, J. Shi, X. Zhang, LFG-Net: low-level feature guided network for precise ship instance segmentation in SAR images, IEEE Trans. Geosci. Remote Sens., 60 (2022), 1–17. https://doi.org/10.1109/tgrs.2022.3188677 doi: 10.1109/tgrs.2022.3188677
[19]	X. Yang, Q. Zhang, Q. Dong, Z. Han, X. Luo, D. Wei, Ship instance segmentation based on rotated bounding boxes for SAR images, Remote Sens., 15 (2023), 1324. https://doi.org/10.3390/rs15051324 doi: 10.3390/rs15051324
[20]	M. Rahimi, S. Sharifian, SAMSAR: a modified SAM architecture for oceanic ship segmentation of satellite SAR images using CNN-based Cross-Fused Attention, Expert Syst. Appl., 284 (2025), 127852. https://doi.org/10.1016/j.eswa.2025.127852 doi: 10.1016/j.eswa.2025.127852
[21]	K. Chen, C. Liu, H. Chen, H. Zhang, W. Li, Z. Zou, et al., RSPrompter: learning to prompt for remote sensing instance segmentation based on visual foundation model, IEEE Trans. Geosci. Remote Sens., 62 (2024), 1–17. https://doi.org/10.1109/tgrs.2024.3356074 doi: 10.1109/tgrs.2024.3356074
[22]	X. Xu, X. Zhang, S. Wei, J. Shi, W. Zhang, T. Zhang, et al., DiffSARShipInst: diffusion model for ship instance segmentation from synthetic aperture radar imagery, ISPRS J. Photogramm. Remote Sens., 223 (2025), 440–455. https://doi.org/10.1016/j.isprsjprs.2025.02.030 doi: 10.1016/j.isprsjprs.2025.02.030
[23]	T. Zhang, X. Zhang, A full-level context squeeze-and-excitation ROI extractor for SAR ship instance segmentation, IEEE Geosci. Remote Sens. Lett., 19 (2022), 1–5. https://doi.org/10.1109/lgrs.2022.3166387 doi: 10.1109/lgrs.2022.3166387
[24]	T. Zhang, X. Zhang, A mask attention interaction and scale enhancement network for SAR ship instance segmentation, IEEE Geosci. Remote Sens. Lett., 19 (2022), 1–5. https://doi.org/10.1109/lgrs.2022.3189961 doi: 10.1109/lgrs.2022.3189961
[25]	S. Wei, X. Zeng, Q. Qu, M. Wang, H. Su, J. Shi, HRSID: a high-resolution SAR images dataset for ship detection and instance segmentation, IEEE Access, 8 (2020), 120234–120254. https://doi.org/10.1109/access.2020.3005861 doi: 10.1109/access.2020.3005861
[26]	T. Zhang, X. Zhang, J. Li, X. Xu, B. Wang, X. Zhan, et al., SAR ship detection dataset (SSDD): Official release and comprehensive data analysis, Remote Sens., 13 (2021), 3690. https://doi.org/10.3390/rs13183690 doi: 10.3390/rs13183690
[27]	J. Li, C. Qu, J. Shao, Ship detection in SAR images based on an improved faster R-CNN, 2017 SAR in Big Data Era: Models, Methods and Applications (BIGSARDATA), 2017, 1–6. https://doi.org/10.1109/bigsardata.2017.8124934
[28]	A. Chattopadhay, A. Sarkar, P. Howlader, V. N. Balasubramanian, Grad-CAM++: generalized gradient-based visual explanations for deep convolutional networks, 2018 IEEE Winter Conference on Applications of Computer Vision (WACV), 2018,839–847. https://doi.org/10.1109/wacv.2018.00097
[29]	D. Bolya, C. Zhou, F. Xiao, Y. J. Lee, YOLACT: real-time instance segmentation, 2019 IEEE/CVF International Conference on Computer Vision (ICCV), 2019, 9157–9165. https://doi.org/10.1109/iccv.2019.00925
[30]	S. Liu, L. Qi, H. Qin, J. Shi, J. Jia, Path aggregation network for instance segmentation, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018, 8759–8768. https://doi.org/10.1109/cvpr.2018.00913
[31]	Z. Wang, Y. Li, S. Wang, Noisy boundaries: lemon or lemonade for semi-supervised instance segmentation, 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, 16805–16814. https://doi.org/10.1109/cvpr52688.2022.01632
[32]	C. Lyu, W. Zhang, H. Huang, Y. Zhou, Y. Wang, Y. Liu, et al., RTMDet: an empirical study of designing real-time object detectors, arXiv preprint, 2022.
[33]	M. Yuan, H. Meng, J. Wu, S. Cai, Global recurrent Mask R-CNN: marine ship instance segmentation, Comput. Graph., 126 (2025), 104112. https://doi.org/10.1016/j.cag.2024.104112 doi: 10.1016/j.cag.2024.104112
[34]	Z. Shao, X. Zhang, S. Wei, J. Shi, X. Ke, X. Xu, et al., Scale in scale for SAR ship instance segmentation, Remote Sens., 15 (2023), 629. https://doi.org/10.3390/rs15030629 doi: 10.3390/rs15030629
[35]	T. Zhang, X. Zhang, HTC+ for SAR ship instance segmentation, Remote Sens., 14 (2022), 2395. https://doi.org/10.3390/rs14102395 doi: 10.3390/rs14102395

Reader Comments

Your name:*

Email:*
© 2026 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)