Ship target detection based on adverse meteorological conditions

Abstract

ship target detection is significant for the marine economy and safe driving of marine autonomous systems. However, recent ship target detection is hardly suitable to adverse meteorological conditions. Different weather conditions will disturb the clarity of taken pictures and hurt the performance of the ship detection. Consequently, accurately classifying and locating the ship target is a big challenge. To address the problems, we propose a ship detection system that contains an adaptive weather classification algorithm, adaptive enhancement algorithm, and RetinaNet. Specifically, we construct a classification network for distinguishing fog, low illumination, and clear pictures. Then, the adaptive enhancement algorithm is proposed to eliminate meteorological interference and restore clear pictures. Finally, the RetinaNet network is used to detect the ship targets with fine processed pictures. Extensive experiments show that our ship system consistently exceeds the strong baseline and improves the performance for ship detection based on adverse meteorological conditions.