PPGS-YOLO: A lightweight algorithms for offshore dense obstruction infrared ship detection

Abstract

With the increasing number of ships at sea, ship monitoring has become increasingly important. However, traditional visible light detection technologies are limited in various environments, particularly in low-light or adverse weather conditions. In contrast, infrared-based ship detection technology performs well under low light and harsh weather conditions, making it an effective alternative. However, most infrared-based ship detection methods currently focus primarily on improving detection accuracy, often at the cost of significant computational resources. To address this issue, this paper proposes a lightweight infrared ship detection algorithm, specifically designed for near-coast applications. We combine the PP-LCNet backbone network with YOLOv5, effectively reducing the model’s parameter count and computational load. Additionally, we introduce a convolution operation suitable for mobile devices, GSConv, to further enhance the algorithm’s computational efficiency, achieving higher performance without compromising accuracy. In the face of frequent ship occlusion in near-coast dense scenarios, we employ the Soft-NMS technique, significantly improving the algorithm’s target detection ability in such environments. Finally, the improved algorithm in this paper achieved a 2.4 % increase in mAP_0.50:0.95 on the dataset, while reducing Flops by 4G and parameters by 1.63 M. The effectiveness of the improved algorithm is verified through extensive experiments.