Visual Attribute based Prompt Learning Method for Fine-grained Ship Recognition

ObjectiveShip image recognition technology plays a significant role in marine domain. A Visual Attribute Prompt Learning (VAPT) mechanism is proposed in this study to address the challenges of strong interference, data scarcity, and inadequate semantic feature modeling encountered in deep learning methods for ship image recognition tasks. MethodsThe framework establishes a large-scale pre-trained visual attribute codebook and incorporates a Multi-head Cross-Attention Mechanism (MCA) to achieve attribute matching and selection procedures, enabling effective alignment with deep visual models to enhance their capability in identifying critical ship features. ResultsExperimental results on a self-constructed dataset containing approximately 50,000 ship images demonstrates that the proposed method achieves approximately 4% accuracy improvement compared to baseline models. ConclusionThe research outcomes provide a novel low-cost technical pathway for feature decoupling and knowledge transfer in target recognition tasks under complex marine environments, offering significant implications for intelligent maritime monitoring systems.