This website requires JavaScript.

Weakly-Supervised Semantic Segmentation of Ships Using Thermal Imagery

Rushil JoshiEthan AdamsMatthew ZiemannChristopher A. Metzler
Dec 2022
The United States coastline spans 95,471 miles; a distance that cannot beeffectively patrolled or secured by manual human effort alone. Unmanned AerialVehicles (UAVs) equipped with infrared cameras and deep-learning basedalgorithms represent a more efficient alternative for identifying andsegmenting objects of interest - namely, ships. However, standard approaches totraining these algorithms require large-scale datasets of densely labeledinfrared maritime images. Such datasets are not publicly available and manuallyannotating every pixel in a large-scale dataset would have an extreme laborcost. In this work we demonstrate that, in the context of segmenting ships ininfrared imagery, weakly-supervising an algorithm with sparsely labeled datacan drastically reduce data labeling costs with minimal impact on systemperformance. We apply weakly-supervised learning to an unlabeled dataset of7055 infrared images sourced from the Naval Air Warfare Center AircraftDivision (NAWCAD). We find that by sparsely labeling only 32 points per image,weakly-supervised segmentation models can still effectively detect and segmentships, with a Jaccard score of up to 0.756.