文献类型：会议论文

英文题名：Marine Ship Detection Under Fog Conditions Based on an Improved Deep-Learning Approach

作者：Xu, Guokang[1];Yin, Jianchuan[1,2,3];Zhang, Zeguo[1]

机构：[1]Guangdong Ocean Univ, Naval Architecture & Shipping Coll, Zhanjiang, Peoples R China;[2]Sea Marine Ranching, Guangdong Prov Key Lab Intelligent Equipment Sout, Zhanjiang 524088, Guangdong, Peoples R China;[3]Guangdong Prov Engn Res Ctr Ship Intelligence & S, Zhanjiang 524088, Guangdong, Peoples R China

会议论文集：5th International Conference on Neural Computing for Advanced Applications (NCAA)

会议日期：JUL 05-07, 2024

会议地点：Asia Pacific Assoc Cognit Intelligence, Guilin, PEOPLES R CHINA

主办单位：Asia Pacific Assoc Cognit Intelligence

语种：英文

外文关键词：Ship detection; fog weather; deformable attention mechanism; deep learning

外文摘要：Target detection plays an essential role in automatic ship identification, ocean border monitoring, and fishing ships supervision, as well as providing environmental perception capability for booming unmanned marine vehicles. However, the maritime setting poses significant challenges, with frequent occurrences of adverse weather conditions like thick fog and dense mist. These harsh environmental factors can significantly impede ship detection efforts, as vessel images may suffer from diminished visual clarity in real-world scenarios. Prior research has primarily relied on image pre-processing methods, such as defogging and low-light enhancement, to enhance image quality before conducting target detection. However, the additional image pre-processing is time-consuming and would hinder us from realizing fast and efficient ship detection, particularly in adverse environmental conditions. To enhance the real-time ship image detection performance under fog conditions, a deformable attention mechanism is proposed to pay attention to relevant locations with more flexibility, which is immune to time-consuming image processing. Firstly, the publicly available new dataset SeaShips was aggregated by using a synthetic central fog algorithm, which can provide realistic visibility conditions for ship detection. Secondly, a deformable attention mechanism was used in the YOLOv8n model to focus on relevant locations more flexibly. Finally, the mechanism was validated based on the new dataset SeaShips_fog. The experimental findings illustrate that the suggested method attains outstanding detection performance even under foggy atmospheric conditions.