Weather-specific detector fusion for ship detection in adverse-weather airborne imagery
Maritime Search and Rescue (SAR) operations require reliable ship detection under adverse weather. This study proposes a weather-robust framework combining synthetic weather augmentation and prediction-level model ensembling. Fog, rain and snow effects at three intensity levels are applied to airborne maritime imagery. Three strategies are evaluated: integrated mixed-weather models, weather-specific models and an ensemble combining predictions without prior weather information. Adaptive Consensus Rescue Fusion (ACR-Fusion), which uses inter-model detection agreement and auxiliary support, is proposed and compared with five existing fusion methods. ACR-Fusion achieved the highest or joint-highest AP50 among the six fusion methods across nine synthetic adverse-weather conditions. On real-world adverse-weather aerial imagery and dense-fog drone imagery, it achieved AP50 values of 0.666 and 0.923, respectively, and an AR50 of 0.967 on drone imagery, tied for highest with WBF, Calibration + WBF and MoCaE. These results highlight the potential of prediction-level fusion for robust ship detection in degraded maritime environments, particularly when detection completeness is important.
Authors
- Jeonghyo Oh (ORCID: https://orcid.org/0009-0003-4083-101X)
- Youngon Oh (ORCID: https://orcid.org/0000-0001-8300-5381)
- Bui Ngoc An (ORCID: https://orcid.org/0000-0001-6988-6979)
- Impyeong Lee (ORCID: https://orcid.org/0000-0003-0706-1060)
Institutions
- University of Seoul (KR)
Publication Details
- Journal
- Geocarto International
- Published
- 2026-11-21
- DOI
- https://doi.org/10.1080/10106049.2026.2765050
- Primary Topic
- Advanced Neural Network Applications
- Type
- article
- Field-Weighted Citation Impact
- 0.00