Ship Detection Framework Based on Deep Learning Network

Abstract

Ship detection and tracking have been recognized as a challenging task in the maritime administration. This paper focuses on the maritime traffic situation, inspects and produces the ship dataset. We improved the existing deep learning method through experiments. It is mainly reflected in the addition of feature reused dense blocks which are used in the feature extraction stage and the addition of contextual information in the low-scale feature map which is used in the multi-scale prediction stage. The improved network model can effectively identify and calibrate the ship image in the dataset, thus improving maritime surveillance efficiency.