A Comparative Image-Based Analysis of Yolo Models for the Detection and Counting of Marine Vehicles

Abstract

Monitoring maritime traffic plays a critical role in applications such as security and surveillance operations, maritime situational awareness, and autonomous marine systems. However, image-based detection of vessels on the sea surface presents a more challenging problem than object detection in terrestrial environments. The continuously dynamic nature of the water surface, sunlight reflections, wave textures, atmospheric variability, and particularly the small apparent size of vessels in long-range imagery can significantly degrade object detection performance. These challenges increase the need for effective image-based approaches capable of reliably and accurately detecting and counting maritime vessels. In this study, a controlled and comparative experimental investigation was conducted to detect and count small-scale maritime vessels on the sea surface. For this purpose, recent versions of the YOLO (You Only Look Once) family of deep learning-based object detection models, namely YOLOv8, YOLOv9, YOLOv10, and YOLOv11, were evaluated. Furthermore, the proposed approach has the potential to contribute indirectly to the reduction of unnecessary navigation activities, fuel consumption, and operational resource utilization by enabling more effective and systematic monitoring of maritime traffic. In addition, the automated monitoring infrastructure designed to complement Automatic Identification System (AIS) systems facilitates the continuous assessment of the environmental impacts of maritime operations and supports the development of sustainable maritime transportation practices. The experimental results demonstrated a general improvement in object detection performance with successive YOLO model generations. While the YOLOv8 model achieved approximately 97% precision and 93% recall, the YOLOv11 model increased these values to 98.4% and 95.2%, respectively. Similarly, the mAP@0.5 score improved from approximately 98% to nearly 99%, whereas the mAP@0.5:0.95 metric increased from 93.5% to 96.5%. Moreover, the observed reductions in Box Loss, Classification Loss, and Distribution Focal Loss (DFL) indicate improvements in both bounding box localization accuracy and training stability. Based on the comparative evaluation, YOLOv11 exhibited the highest overall performance across all considered metrics and achieved successful results in the vessel counting task.

Keywords

• Deep Learning
• Object Detection
• Marine Vessels
• Object Counting
• YOLO

Deniz Aracı Tespiti ve Sayımı için YOLO Modellerinin Görüntü Tabanlı Karşılaştırmalı Analizi

Abstract

Deniz trafiğinin izlenmesi, güvenlik ve gözetim faaliyetleri ile otonom deniz sistemleri gibi uygulamalarda görüntü tabanlı nesne tespiti kritik bir rol oynamaktadır. Bununla birlikte, deniz yüzeyindeki araçların görüntü tabanlı olarak tespit edilmesi, karasal ortamlardaki nesne tespitine kıyasla daha zorlu bir problemdir. Su yüzeyinin sürekli hareketli yapısı, güneş ışığı yansımaları, dalga dokusu, atmosferik koşullardaki değişkenlik ve özellikle uzak mesafelerden elde edilen görüntülerde nesnelerin küçük boyutlarda görünmesi, nesne tespit performansını olumsuz etkileyebilmektedir. Bu koşullar, deniz araçlarının güvenilir ve hassas bir şekilde tespit edilmesi ve sayılabilmesi için etkili görüntü tabanlı yöntemlere duyulan ihtiyacı artırmaktadır. Bu çalışmada, deniz yüzeyinde bulunan küçük boyutlu deniz araçlarının tespiti ve sayımı amacıyla kontrollü ve karşılaştırmalı bir deneysel çalışma gerçekleştirilmiştir. Bu kapsamda, derin öğrenme tabanlı nesne tespit yöntemlerinden biri olan YOLO (You Only Look Once) ailesinin güncel sürümleri YOLOv8, YOLOv9, YOLOv10 ve YOLOv11 ele alınmıştır. Geliştirilen yaklaşım, deniz trafiğinin daha etkin ve sistematik bir biçimde izlenmesine olanak sağlayarak gereksiz seyir faaliyetlerinin, yakıt tüketiminin ve operasyonel kaynak kullanımının azaltılmasına dolaylı katkı sunma potansiyeline sahiptir. Bunun yanı sıra, Otomatik Tanımlama Sistemini (AIS) tamamlayıcı nitelikte tasarlanan otomatik izleme altyapısı, denizcilik faaliyetlerinin çevresel etkilerinin sürekli olarak değerlendirilmesine imkân tanımakta ve sürdürülebilir deniz taşımacılığı uygulamalarının geliştirilmesini desteklemektedir. Deneysel sonuçlar, YOLO model sürümleri ilerledikçe nesne tespit performansında genel olarak artış eğilimi olduğunu göstermiştir. YOLOv8 modelinde yaklaşık %97 kesinlik ve %93 duyarlılık elde edilirken YOLOv11 modelinde bu değerler sırasıyla %98,4 ve %95,2 seviyesine ulaşmıştır. Benzer şekilde, mAP@0.5 değeri yaklaşık %98 seviyesinden %99 seviyesine yaklaşırken, mAP@0.5:0.95 değeri %93,5’ten %96,5’e yükselmiştir. Bununla birlikte, kutu kaybı (Box Loss), sınıflandırma kaybı (Classification Loss) ve DFL (Distribution Focal Loss) değerlerinde gözlenen azalma, modelin sınır kutusu konumlandırma hassasiyetinin ve eğitim sürecindeki kararlılığının geliştiğine işaret etmektedir. Karşılaştırmalı değerlendirme sonucunda YOLOv11, ele alınan performans metrikleri açısından en yüksek genel performansı sergileyen model olarak belirlenmiş ve nesne sayımı uygulamasında başarılı sonuçlar vermiştir.

Keywords

• Deniz Araçları
• Nesne Tespiti
• Nesne Sayımı
• Derin Öğrenme
• YOLO

References

1. Casas, E., Ramos, L., Bendek, E., & Rivas-Echeverria, F. (2024). YOLOv5 vs. YOLOv8: Performance benchmarking in wildfire and smoke detection scenarios. Journal of Image and Graphics, 12(2), 127-136. https://doi.org/10.18178/joig.12.2.127-136
2. Chen, X., Wei, C., Xin, Z., Zhao, J., & Xian, J. (2023). Ship detection under low-visibility weather interference via an ensemble generative adversarial network. Journal of Marine Science and Engineering, 11(11), 2065. https://doi.org/10.3390/jmse11112065
3. Dhivya, K., Venkatesan, M., Prabhavathy, P. (2025). Enhance YOLOV5 model to detect the ocean ships using real-time hyperspectral satellite image. In A. Kanhe, S. Balanethiram, P.-A. Hsiung, & D. N. K. Jayakody (Eds.), Advances in VLSI, signal processing and wireless communication Lecture Notes in Electrical Engineering, (pp. 121–137). Springer. https://doi.org/10.1007/978-981-96-1587-2_10
4. Fefilatyev, S., Goldgof, D., Shreve, M., & Lembke, C. (2012). Detection and tracking of ships in open sea with rapidly moving buoy-mounted camera system. Ocean Engineering, 54, 1–12. https://doi.org/10.1016/j.oceaneng.2012.06.028
5. Han, W., Dong, X., Wang, G., Ding, Y., & Yang, A. (2025). Application and improvement of YOLO11 for brain tumor detection in medical images. Frontiers in Oncology, 15, 1643208. https://doi.org/10.3389/fonc.2025.1643208
6. He, L., Zhou, Y., Liu, L., Cao, W., & Ma, J. (2025). Research on object detection and recognition in remote sensing images based on YOLOv11. Scientific Reports, 15, 14032. https://doi.org/10.1038/s41598-025-96314-x
7. Hu, Z., Wu, W., Yang, Z., Zhao, Y., Xu, L., Kong, L., Chen, Y., Chen, L., & Liu, G. (2025). A cost-sensitive small vessel detection method for maritime remote sensing imagery. Remote Sensing, 17(14), 2471. https://doi.org/10.3390/rs17142471
8. Jocher, G., Chaurasia, A., Qiu, J. (2023). Ultralytics YOLO (Version 8.0.0) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.7347926

9. Jocher, G., Qiu, J. (2024). Ultralytics YOLO11. (Version 11.0.0) [Computer software]. Ultralytics. https://docs.ultralytics.com/models/yolo11
10. Khanam, R., & Hussain, M. (2024). YOLOv11: An overview of the key architectural enhancements. arXiv. https://doi.org/10.48550/arXiv.2410.17725
11. Li, J., Zhang, Y., Sun, Y., & Zhang, C. (2026). Robust ship detection algorithm under complex occlusion conditions. Journal of Marine Science and Engineering, 14(15), 1423. https://doi.org/10.3390/jmse14151423
12. Liu, Q., Li, Z., Yuan, S., Zhu, Y., & Li, X. (2021). Review on vehicle detection technology for unmanned ground vehicles. Sensors, 21(4), 1354. https://doi.org/10.3390/s21041354
13. Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S., Fu, C. Y., Berg, A. C. (2016). Ssd: Single shot multibox detector. In European conference on computer vision (pp. 21-37). Springer. https://doi.org/10.1007/978-3-319-46448-0_2
14. Magalhães, R., Falcão, A. P., & Barbosa, A. (2025). Vessel detection leveraging satellite imagery and YOLO in maritime surveillance. Remote Sensing Applications: Society and Environment, 40, 101730. https://doi.org/10.1016/j.rsase.2025.101730
15. Pixabay. (2026). Pixabay. Retrieved May 10, 2026, from https://pixabay.com/tr/
16. Redmon, J., Divvala, S., Girshick, R., & Farhadi, A. (2016). You only look once: Unified, real-time object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 779-788). https://doi.org/10.1109/CVPR.2016.91
17. Redmon, J., & Farhadi, A. (2018). YOLOv3: An incremental improvement. arXiv preprint arXiv:1804.02767. https://doi.org/10.48550/arXiv.1804.02767
18. Ren, S., He, K., Girshick, R., & Sun, J. (2015). Faster R-CNN: Towards real-time object detection with region proposal networks. In C. Cortes, N. D. Lawrence, D. D. Lee, M. Sugiyama, & R. Garnett (Eds.), Advances in neural information processing systems (Vol. 28). Curran Associates, Inc. https://doi.org/10.48550/arXiv.1506.01497
19. Roboflow. (2026). Roboflow. Retrieved May 20, 2026, https://roboflow.com/
20. Sharma, D. K., Pal, M. K., & Singh, A. K. (2025). Explainable AI for lung cancer detection using YOLOv11 and LLMs. Life Cycle Reliability and Safety Engineering, 15(3), 887–897. https://doi.org/10.1007/s41872-025-00372-z
21. Shutterstock. (2026) Shutterstock. Retrieved May 10, 2026, https://shutterstock.com/tr/video
22. Teng, H., Wang, Y., Li, W., Chen, T., & Liu, Q. (2025). Advancing rice disease detection in farmland with an enhanced YOLOv11 algorithm. Sensors, 25(10), 3056. https://doi.org/10.3390/s25103056
23. Ultralytics. (2023). YOLOv8 Documentation. https://docs.ultralytics.com/models/yolov8/
24. Wang, C. Y., Yeh, I. H., & Mark Liao, H. Y. (2024a). YOLOv9: Learning what you want to learn using programmable gradient information. In European conference on computer vision- ECCV 2024 (pp. 1-21). Springer Nature Switzerland. https://doi.org/10.1007/978-3-031-72751-1_1
25. Wang, A., Chen, H., Liu, L., Chen, K., Lin, Z., Han, J., & Ding, G. (2024b). YOLOv10: Real-time end-to-end object detection. In Advances in neural information processing systems (Vol. 37, pp. 107984–108011). Curran Associates, Inc. https://doi.org/10.52202/079017-3429
26. Wang, N., Wang, Y., Feng, Y., & Wei, Y. (2024c). MDD-ShipNet: Math-data integrated defogging for fog-occlusion ship detection. IEEE Transactions on Intelligent Transportation Systems, 25(10), 15040–15052. https://doi.org/10.1109/TITS.2024.3394573
27. Wang, N., Wang, Y., Feng, Y., Wei, Y., (2024d). AodeMar: Attention-aware occlusion detection of vessels for maritime autonomous surface ships. IEEE Transactions on Intelligent Transportation Systems, 25(10), 13584–13597.
28. Wang, Y., Wang, M., Shi, J., Ni, Z., & Li, G. (2026). eAodeMar: A lightweight and real-time occluded marine vessel detection network for embedded marine platforms. Frontiers in Marine Science, 13, 1778827. https://doi.org/10.3389/fmars.2026.1778827
29. Yang, S., Luo, H., Xu, Z., Yang, J., & Zhang, W. (2026). OShipNet: Occlusion ship detection based on multidomain fusion and multiscale refinement. Journal of Marine Science and Engineering, 14(9), 804. https://doi.org/10.3390/jmse14090804
30. Zeng, Y., Wang, X., Zou, J., & Wu, H. (2025). YOLO-Ssboat: Super-small ship detection network for large-scale aerial and remote sensing scenes. Remote Sensing, 17(11), 1948. https://doi.org/10.3390/rs17111948
31. Zhang, Q., Wang, L., Meng, H., Zhang, Z., & Yang, C. (2024). Ship detection in maritime scenes under adverse weather conditions. Remote Sensing, 16(9), 1567. https://doi.org/10.3390/rs16091567
32. Zou, Z., Shi, Z., Guo, Y., & Ye, J. (2019). Object detection in 20 years: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 43(6), 1720–1745.