Multiscale ship detection algorithm incorporating global attention mechanism
HUANG Yuan
CAO Minghua
LÜ Zhisheng
ZHANG Lingyun
FENG Bin
Abstract：The aim of this research is to address the remaining issues with small target leakage and poor multi-target classification performance that are still present during navigation in complex sea areas.To this end,a global attention-guided multi-scale ship detection algorithm,named GAG-YOLO,has been proposed.The global attention mechanism(GAM)has been introduced into the YOLOv5 target detection network in order to capture the global contextual information,thereby enabling the model to better dis-criminate the ship target from the complex contextual differences.Furthermore,the anchor frame parame-ters are optimized to better match the shape of the ship and to more accurately locate and identify ship tar-gets at different scales.Finally,the loss function of the YOLOv5 network is improved to focus on high-quality samples using Focal-EIOU for more accurate bounding box regression by assigning different weights.The experimental results demonstrate that GAG-YOLO outperforms the listed YOLO series al-gorithms in terms of detection accuracy and robustness.Specifically,the mean average precision(mAP)at 0.5∶0.95 is improved by 4.6％.The detection algorithm improves the overall detection speed and effec-tively reduces the leakage and false detection rate,providing a faster and more accurate solution for ship monitoring in complex sea areas.
Keywords：YOLOv5sglobal attention mechanismFocal-EIOUanchor boxmulti-scale ship detection
Publication Date：2025-12-31
Online Publishing Date：2026-08-28(First online date of this platform, not the publication date of the document)
Pages：11( 853-863 )