Cross-Modal Attention-Modulated Feature Enhancement Network for Visible-Infrared Ship Detection

Ship detection is crucial for tasks such as harbor dynamic surveillance and maritime traffic management, aiming to identify and locate ships. Multimodal object detection based on visible-infrared data is well suited for ship detection because it enables all-weather object detection. However, it lacks paired visible-infrared ship datasets, and there still exist problems such as insufficient consistency and complementarity of cross-modal semantic information, as well as a large amount of redundant information when fusing different modalities. To address these problems, two pseudo-infrared datasets are established, and a Cross-modal Attention-modulated Feature Enhancement Network (CAMFEN) approach for ship detection is proposed. CAMFEN mainly utilizes the Cross-modal Collaborative-Differential Enhancement Feature Fusion module ($ \rm{C}^{2}\rm{DEFF}$) to effectively fuse the information of different modalities, which consists of Collaborative Attention Modulation Block (CAMB), Differential Attention Modulation Block (DAMB), and Global Feature Guided Fusion Block (GFGFB). The CAMB achieves cross-modal semantic alignment through channel attention modulation, eliminates geometric offset with spatial adaptive calibration, strengthens common semantics representation from the perspective of collaborative integrity, and equalizes modal contribution. The DAMB selects discriminant difference features through channel attention screening, uses spatial attention to focus on target regions, and mines complementary or contradictory information from the perspective of difference specificity, enhancing modal specificity and suppressing redundant noise. Finally, the enhanced single modality features obtained by CAMB and DAMB are fed into GFGFB, which guides the optimized fusion of enhanced single modality features from a global perspective. The proposed method has been validated on HRSC2016 and DOTAv1.0 datasets, and the experimental results show that CAMFEN outperforms existing ship detection methods.