Edge Prior Guidance and Hybrid Receptive Field Perception for Water Surface Object Detection in Adverse Navigation Environments

Objectives Robust water surface object detection is essential for the safe autonomous navigation of intelligent ships. However, image degradation caused by adverse navigation environments, progressive attenuation of edge information in deep neural networks, and significant variations in the scale and geometric morphology of water surface objects severely limit the performance of existing detection algorithms. Methods To address the above three coupled challenges, this paper proposes a water surface object detection framework named EP²H-DETR (Edge-Prior and Hybrid-Receptive-Field DETR), which integrates edge-prior guidance and hybrid receptive field perception. At the input stage, a Difference-of-Gaussian based edge prior is explicitly embedded, and a dedicated edge pathway is constructed to enable cross-stage feature propagation, thereby effectively alleviating edge degradation caused by adverse imaging conditions and edge information loss in deep networks. Furthermore, a hybrid receptive field perception module based on the Cross Stage Partial architecture is designed to adaptively fuse three types of receptive fields, including square, horizontal strip, and vertical strip receptive fields, enabling flexible adaptation to the diverse scales and geometric shapes of water surface objects. In addition, a dataset for inland waterway object detection under adverse conditions (ACIWD) is constructed, containing 7 270 images and 38 421 annotated instances, with adverse conditions accounting for 88.6% of the dataset. Results On the ACIWD dataset, EP²H-DETR achieves a mean Average Precision (mAP) of 73.7%, improving by 4.4% over the baseline while reducing the number of parameters by 24.1%. Under foggy and nighttime conditions, the proposed method improves mAP by 10.0% and 7.7%, respectively. Moreover, EP²H-DETR demonstrates strong generalization ability on three public datasets, including WSODD, SeaShips, and FloW. Conclusions The proposed method significantly improves the robustness of water surface object perception for intelligent ships operating in complex adverse navigation environments, providing effective technical support for inland waterway safety assurance and intelligent navigation perception systems.