시뮬레이션 기반 합성 데이터셋을 활용한 선박 탐지 모델의 성능 개선

김태준¹^{, *} ; 한민성² ; 이화섭³

1어드바이저로렌 AI연구실 연구원
2어드바이저로렌 대표이사
3해군소령/해군 작전사령부 해양작전본부 전투발전참모처 무인체계작전혁신과

Improving Ship Detection Performance Using Simulation-Based Synthetic Datasets

Tae-Jun Kim¹^{, *} ; Min-Sung Han² ; Hwa-Seob Lee³

1Research engineer, AI Research & Data Div., AdvisorLauren
2CEO, AdvisorLauren
3LCDR, ROK Navy/Combat Development Department, Naval Operation Command, ROK Navy

Correspondence to: *Tae-Jun Kim 1161 Gyebaek-ro, Seo-gu, Daejeon, Republic of Korea Tel: +82-2-2088-4050 E-mail: taejun.kim@advl.ai

초록

군사·해상 감시 분야는 보안, 운용 제약으로 실선박 데이터 확보가 어렵고, 확보된 데이터도 기상, 조도 변동성을 충분히 담지 못해 탐지 성능 저하로 이어진다. 본 연구는 시뮬레이션에서 domain randomization으로 생성한 합성 선박 이미지로 YOLOv8n과 Faster R-CNN을 사전학습해 더 나은 초기 가중치를 확보하고, 이를 선박 탐지 데이터셋에서 파인튜닝해 성능 개선 효과를 검증하였다. 두 모델 모두 mAP50:95에서 일관된 개선을 보였으며, 특히 Faster R-CNN은 고IoU 구간 위치 정확도 향상과 오탐 감소가 두드러져, 합성데이터 사전학습이 탐지 정밀도 개선에 기여함을 확인하였다.

Abstract

In military and maritime surveillance applications, security and operational constraints limit the acquisition of sufficient real-world ship imagery, while existing datasets often fail to adequately represent diverse weather and lighting conditions that can degrade detection performance. In this study, YOLOv8n and Faster R-CNN are pretrained on synthetic ship images generated via domain randomization in simulation to obtain improved initial weights, followed by fine-tuning on a ship detection dataset to evaluate the resulting performance gains. Both models demonstrate consistent improvements in mAP50:95, with Faster R-CNN showing particularly notable gains in localization accuracy at high IoU thresholds and a reduction in false positives. These results indicate that synthetic-data pretraining can contribute to improved ship detection precision.

Keywords:

Synthetic Data, Domain Randomization, Object Detection, Pretraining, Marine Target Recognition

키워드:

합성 데이터, 도메인 랜덤화, 객체 탐지, 사전학습, 해상 표적 인식

Acknowledgments

본 연구는 창업진흥원과 대구창조경제혁신센터의 “2026년 「민관협력 오픈이노베이션 지원」 방산 스타트업 챌린지” 지원을 받아 수행된 연구결과이다(과제번호: 20406831, 과제명: Non-GPS 대응 함상 자율복귀·자동 이착함 AI 드론 운용체계). 본 논문의 투고 및 심사에 소요되는 비용은 2026년 해군사관학교 학술연구과제(과제명 : 소형표적의 효과적 대응을 위한 양자센싱 기반 탐지체계 개념연구)의 지원을 받았다.

References

Josh Tobin, Rachel Fong, Alex Ray, Jonas Schneider, Wojciech Zaremba, & Pieter Abbeel, ‘Domain Randomization for Transferring Deep Neural Networks from Simulation to the Real World,’ in proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Vancouver, Canada, 2017, pp. 23-30. [https://doi.org/10.1109/IROS.2017.8202133]
Xiaomeng Zhu, Jacob Henningsson, Duruo Li, Pär Mårtensson, Lars Hanson, Mårten Björkman, & Atsuto Maki, ‘Domain Randomization for Object Detection in Manufacturing Applications using Synthetic Data: A Comprehensive Study,’ arXiv:2506.07539, , 2025.
Jie Fan, Yongqiang Xie, & Zhongbo Li, ‘Domain Randomization for Object Detection: A Survey,’ in proceedings of the 8th International Conference on Advanced Electronic Materials, Computers and Software Engineering (AEMCSE), Nanjing, China, 2025, pp. 201-214. [https://doi.org/10.1109/AEMCSE65292.2025.11042418]
Luisa Torquato Niño & Hamza A. A. Gardi, ‘Synthetic-to-Real Object Detection Using YOLOv11 and Domain Randomization Strategies,’ arXiv:2509.15045, , 2025.
NVIDIA, Isaac Sim - Robotics Simulation and Synthetic Data Generation, NVIDIA Developer. https://developer.nvidia.com/isaac/sim, (Accessed 2026.08.25.)
Sicong Gao, Maurice Pagnucco, Tomasz Bednarz, & Yang Song, ‘NVIDIA Isaac Sim: Enabling Scalable, GPU-Accelerated Simulation for Robotics,’ arXiv:2606.03551, , 2026.
Martina Hutter-Mironovova, ‘Sim-to-Real Fruit Detection Using Synthetic Data: Quantitative Evaluation and Embedded Deployment with Isaac Sim,’ arXiv:2603.28670, , 2026.
Jingyu Song, Haoyu Ma, Onur Bagoren, Advaith V. Sethuraman, Yiting Zhang, & Katherine A. Skinner, ‘OceanSim: A GPU-Accelerated Underwater Robot Perception Simulation Framework,’ arXiv:2503.01074, , 2025. [https://doi.org/10.1109/IROS60139.2025.11246878]
Min-Seok Han & Jaeryong Hwang, ‘Dynamic Pruning and Integrated Lightweight Optimization of YOLOv11 for Real-Time Object Detection in Variable Maritime Environments,’ Journal of the KNST, VOL. 8, NO. 4, 2025, pp. 916-921. [https://doi.org/10.31818/JKNST.2025.12.8.4.916]
Changju Bak, ‘Edge-Enhanced CNN Training for Fine-Grained Ship Classification in Remote Sensing,’ Journal of the KNST, VOL. 8, NO. 4, 2025, pp. 710-719. [https://doi.org/10.31818/JKNST.2025.12.8.4.710]
Yun-Gyo Im, Youjeong Youn, Jonggu Kang, Seoyeon Kim, Ye-min Jeong, Soyeon Choi, Youngmin Seo, & Yang-Won Lee, ‘Ship Detection from SAR Images Using YOLO: Model Constructions and Accuracy Characteristics According to Polarization,’ Korean Journal of Remote Sensing, VOL. 39, NO. 5, 2023, pp. 997-1008.
Junghwan Kim & Hoseok Moon, ‘A Model Development for Object Detection and Classification at Sea Using Modified YOLO v4,’ Journal of the Military Operations Research Society of Korea, VOL. 47, NO. 2, 2021, pp. 13-25.
Zhisheng Li, Zhihui Deng, Kun Hao, Xiaofang Zhao & Zhigang Jin, ‘A Ship Detection Model Based on Dynamic Convolution and an Adaptive Fusion Network for Complex Maritime Conditions,’ Sensors, VOL. 24, NO. 3, 2024, Article 859. [https://doi.org/10.3390/s24030859]
Yanfeng Gong, Zihao Chen, Jiawan Tan, Chaozhong Yin, & Wen Deng, ‘Two-Stage Ship Detection at Long Distances Based on Deep Learning and Slicing Technique,’ PLoS ONE, VOL. 19, NO. 11, 2024, Article e0313145. [https://doi.org/10.1371/journal.pone.0313145]
Bogdan Iancu, Valentin Soloviev, Luca Zelioli & Johan Lilius, ‘ABOships — An Inshore and Offshore Maritime Vessel Detection Dataset with Precise Annotations,’ Remote Sensing, VOL. 13, NO. 5, 2021, Article 988. [https://doi.org/10.3390/rs13050988]
Junxiong Liang, Mengwei Bao, Tianxiang Wang, Xinggang Wang, An-An Liu, & Ryan Wen Liu, ‘WUTDet: A 100K-Scale Ship Detection Dataset and Benchmarks with Dense Small Objects,’ arXiv:2604.07759, , 2026.
Zhenfeng Shao, Wenjing Wu, Zhongyuan Wang, Wan Du, & Chengyuan Li, ‘SeaShips: A Large-Scale Precisely Annotated Dataset for Ship Detection,’ IEEE Transactions on Multimedia, VOL. 20, NO. 10, 2018, pp. 2593-2604. [https://doi.org/10.1109/TMM.2018.2865686]
Erhan Gundogdu, Berkan Solmaz, Veysel Yücesoy, & Aykut Koç, ‘MARVEL: A Large-Scale Image Dataset for Maritime Vessels,’ in proceedings of the 13th Asian Conference on Computer Vision (ACCV), Taipei, Taiwan, 2016, pp. 165-180. [https://doi.org/10.1007/978-3-319-54193-8_11]
Abhilasha Nanda, Sung Won Cho, Hyeopwoo Lee, & Jin Hyoung Park, ‘KOLOMVERSE: Korea Open Large-Scale Image Dataset for Object Detection in the Maritime Universe,’ IEEE Transactions on Intelligent Transportation Systems, VOL. 25, NO. 12, 2024, pp. 20832-20840. [https://doi.org/10.1109/TITS.2024.3449122]
Zikun Liu, Liu Yuan, Lubin Weng, & Yiping Yang, ‘A High Resolution Optical Satellite Image Dataset for Ship Recognition and Some New Baselines,’ in proceedings of the 6th International Conference Pattern Recognition Applications and Methods (ICPRAM), Porto, Portugal, 2017, pp. 324-331. [https://doi.org/10.5220/0006120603240331]
Zhengning Zhang, Lin Zhang, Yue Wang, Pengming Feng, & Ran He, ‘ShipRSImageNet: A Large-Scale Fine-Grained Dataset for Ship Detection in High-Resolution Optical Remote Sensing Images,’ IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, VOL. 14, 2021, pp. 8458-8472. [https://doi.org/10.1109/JSTARS.2021.3104230]
Sketchfab, The Leading Platform for 3D & AR on the Web. https://sketchfab.com, (accessed 2026.08.25.)
Glenn Jocher, Ayush Chaurasia, & Jing Qiu, ‘Ultralytics YOLOv8,’ version 8.0.0, 2023. https://github.com/ultralytics/ultralytics, (accessed 2026.08.25.)
Shaoqing Ren, Kaiming He, Ross Girshick, & Jian Sun, ‘Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks,’ Advances in Neural Information Processing Systems (NIPS), 2015, pp. 91-99.