Ship detection and identification is the key part of the maritime monitoring and safety. Ship monitoring methods based on coastal video surveillance, satellite imagery, and synthetic aperture radar have been well developed. As the emerging remote sensing technology, distributed acoustic sensing (DAS) technology which continuously detects vibrations along underwater optical fiber cables facilitates all-weather, all-day, and real-time ship detection capabilities, possessing the potential for detecting dark ships. However, the reliance on expert knowledge for analyzing ship passage signals hinders the development of an automated framework for ship detection, limiting the application of DAS technology in the ship detection. Additionally, the scarcity of datasets for ship passage events in the DAS field hampers the adoption of deep learning technologies for enhancing ship detection. To address these challenges, an automatic annotation method is proposed, utilizing 18625 cleaned ship records based on the automatic identification system to annotate ship passages adaptively from 5-month DAS data. Thus a large-scale, high-quality annotated dataset named DAShip is established, containing 55875 ship passage samples. Furthermore, an online ship detection and identification framework is proposed to achieve real-time ship detection from the massive DAS data flow and further identify coarse-grained ship features, such as ship speed, heading, angle, and ship type. In this proposed framework, YOLO models, primarily trained on DAShip, are used as ship detectors and ship feature classifiers, achieving accurate dark ship detection combined with automatic identification system message and demonstrating competitive performance in ship feature classification.