东南大学学报（自然科学版）2025，Vol.55Issue(5)：1380-1387,8.DOI:10.3969/j.issn.1001-0505.2025.05.019
基于改进YOLOv8和ByteTrack的桥梁通航船舶识别与追踪
Ship identification and tracking for navigable bridge based on improved YOLOv8 and ByteTrack
摘要
Abstract
In response to the frequent ship-bridge collision accidents in recent years,the existing active ship-collision prevention methods for bridges were analyzed in depth for their deficiencies.A channel ship de-tection and tracking method based on the improved YOLOv8 and ByteTrack algorithms was developed.Three convolutional block attention modules(CBAM)were introduced between the backbone and neck networks of the YOLOv8 structure to enhance the model's ability to capture key features.The ByteTrack algorithm was employed to improve the accuracy and robustness of ship tracking.Comparative experiments were conducted for analysis.The results indicate that the improved model achieves a multi-object tracking accuracy(MOTA)of 79.8%and an identification accuracy(IDF1)of 84.5%,representing an approximate 5%increase in accu-racy compared with the original YOLOv8 model.It also shows a higher improvement compared with other mainstream attention mechanism modules.In terms of image processing speed,the improved method is about 56%faster than the Bot-SORT multi-object tracking algorithm.Less time is taken to process the same target images.关键词
桥梁工程/船舶追踪/深度学习/计算机视觉/YOLOv8/ByteTrackKey words
bridge engineering/ship tracking/deep learning/computer vision/YOLOv8/ByteTrack分类
交通工程引用本文复制引用
王浩,王旭,廖睿轩,茅建校,张一鸣,颜王吉..基于改进YOLOv8和ByteTrack的桥梁通航船舶识别与追踪[J].东南大学学报（自然科学版）,2025,55(5):1380-1387,8.基金项目
国家自然科学基金重点资助项目(52338011). （52338011）
