Ship detection with deep learning: a survey

Ship detection plays a pivotal role in efficient marine monitoring, port management, and safe navigation. However, the development of ship detection techniques is vastly behind other detection techniques, such as face detection, pedestrian detection, traffic sign/light detection, text detection, etc. In this paper, we explore the status quo and identify the following reasons for the slow development: (1) the existing methodologies are weakly systematic; (2) there are no unified evaluation criteria; (3) there are no widely accepted datasets which vastly hinder its development in deep learning era. In this context, we conduct a critical review of the state-of-the-art ship detection techniques based on deep learning. The main contributions of this work are: (1) existing works on object detection are comprehensively reviewed; (2) popular/benchmark datasets are extensively collected and analysed; (3) evaluation criteria for ship detection are ultimately unified; and (4) challenges and optimization methods are discussed and future directions projected.