Research on Dense Object Detection Method Based on Symmetry

Abstract This article addresses the challenges of occlusion, overlap, and background noise in dense object detection, particularly in symmetric dense object scenarios such as ship detection in harbors. The proposed SSI YOLOv8-OBB model leverages symmetry-based methods to enhance detection accuracy. It integrates self-supervised symmetry into the OBB rotation box to optimize positioning and introduces the SPD-Conv module and SE Module to improve detection efficiency and accuracy. The model’s loss function is optimized using Inner-IoU, which effectively handles bounding box regression. Experimental results show significant improvements in key performance metrics compared to other models, demonstrating enhanced accuracy and efficiency in dense object detection. This approach is particularly effective in complex maritime environments, where it has been shown to improve ship detection accuracy and reduce model convergence time.