FusionAttNet: A Water Surface Object Detection Model with Dynamic Fusion and Cross-Modal Attention

Water surface object detection is critical to the intelligent advancement of the shipping industry, with extensive military and civilian applications. However, existing methods face challenges in detecting small, densely distributed objects in complex environments and generalizing to real-world scenarios. To address these issues, we propose FusionAttNet, a water surface object detection model that integrates dynamic fusion and cross-modal attention, enabling object detection under text guidance. Specifically, a convolution and self-attention dynamic fusion module is incorporated to simultaneously capture local details and global context information, ensuring precise detection of multi-scale objects. An aggregation based on shared attention matching module is introduced to adaptively select key features while suppressing background noise, enhancing the perception of small and dense objects. Additionally, a dual-way cross-modal attention module facilitates deep integration of visual and linguistic features, significantly improving generalization to unseen categories. Experimental results show that FusionAttNet achieves superior performance in detecting small and dense objects while exhibiting strong generalization capabilities, validating its effectiveness.