Multi-Scale Inshore Ship Detection Based on Feature Re-Focusing Network

Aiming at the problems of multi-scale inshore ship detection in surveillance videos, this paper proposes a ship target detection algorithm based on feature re-focusing network, and designs a feature re-focusing strategy, which consists of a multi-scale feature aggregation module (MFAM) and attention feature re-assignment module (AFRM). Specifically, MFAM fuses the semantic information of different levels of features of multi-scale ships by constructing a feature aggregation block based on the input feature pyramid. AFRM is composed of multi-branch dilated convolutions as well as channel and spatial attention mechanisms, which can improve the network's representation of target non-local information and suppressing interference of background, and a feature re-focusing pyramid is established for target detection. The experimental results on the Seaships7000 ship public data set show that compared with other algorithms, the algorithm has a better detection effect on multi-scale inshore ships in surveillance videos.