- Invariant properties can provide important clues for intelligent perception of visual objects, but have not been effectively exploited in marine scenarios. This paper looks into an invariant property of ships, ship structure. Specifically, we define for the first time the visual structure of a ship as ship skeleton produced based on semantic and structural characteristics of deep features. With such a definition, we build the first ship structural dataset named Maritime Ships with Visual Structure (MSVS), where each ship is annotated with ship skeleton along with its common properties such as category, bounding box, IMO number, etc. By performing extensive experiments on the newly created dataset, we find that the keypoints that form ship skeleton play an important role in robust and effective training of ship recognition models. Furthermore, we show that ship skeleton conveying both 2D and 3D visual information contributes significantly to the perception of a ship. + Ship identification is a prerequisite for the intelligent management of maritime transportation, yet existing research is limited to the overall detection and categorization of ships, and there is a lack of data and related research on ship identification. + Inspired by the research on vehicle license plates, we make the first attempt to propose the concept of ship license plate and construct the first ship license plate detection and recognition (SLPDR) dataset, which contains 88, 862 images and 1,420 instances. + In addition, this paper presents a state space attention based YOLO model, which effectively integrate global + information and local details for fast and accurate ship license plate detection. + Furthermore, we explore the application of ship license plate in intelligent maritime transportation, including ship identification, connecting images and ships and ship compliance assessment, which provides new ideas for intelligent management of maritime vessels.