 The paper proposes a framework called rotation dense feature pyramid networks, RDFPN, to effectively detect ships in different scenes including ocean and port, which addresses the limitations of traditional ship detection methods such as complexity of application scenarios, difficulty of intensive object detection, and redundancy of the detection region. The proposed method uses a dense feature pyramid network, DFPN, that builds high-level semantic feature maps for all scales by means of dense connections to enhance feature propagation and reuse, and also includes a rotation anchor strategy to predict the minimum circumscribed rectangle of the object and multi-scale region of interest, ROI, aligned to maintain completeness of semantic and spatial information. Experiments show that the detection method based on RDFPN representation has state-of-the-art performance. This article was authored by Xue Yang, Hao Sun, Ku and Fu, and others.