Domain adapted probabilistic inspection using deep probabilistic segmentation

This paper introduces the concept of domain-adapted probabilistic segmentation for marine vessel classification. The evolution corrosion is continuous and it is, therefore, impossible to acquire inspection datasets representative entire active fleet. Additionally, human surveyors introduce high levels subjectiveness in classification process, resulting potentially multiple equally valid but ambiguous results. Consequently, deterministic flawed. goal this address these challenges by using a approach while performing domain adaptation align feature space across different stages age degradation. We test Probabilistic U-Net on both simulated images from real vessels compare against two novel models. have evaluated models quantitative — energy distance as distribution similarity qualitative reduction visualization approaches. Our results indicate that combination adaption could impact surveys future.