基于YOLOv8的民用船舶影像分类方法研究  

Research on Civil Ship Image Classification Method Based on YOLOv8

作　　者：张潇艺 杨胜龙[2] 

机构地区：[1]上海海事大学物流工程学院,上海201306 [2]农业农村部渔业遥感重点试验室,中国水产科学研究院东海水产研究所,上海200090

出　　处：《工业控制计算机》2024年第4期72-73,76,共3页Industrial Control Computer

基　　金：崂山实验室专项经费(LSKJ202201804)。

摘　　要：航运水平不断发展,船舶种类逐渐细化,面对用途不同船舶,准确掌握船舶类型为人们更好地管理船舶提供条件。选用FGSCR数据集,筛选其中的起重船舶、超级游艇、货船、集装箱船、拖船、小型游艇、运砂船、油船等8种类型,基于YOLOv8的实现民用船舶影像分类。对YOLOv8的YOLOv8n、YOLOv8s、YOLOv8m、YOLOv8l、YOLOv8x的5个模型进行训练,结果显示模型大小从2.83 MB增大到107 MB,其中Top1准确率最高的是YOLOv8m,达到96.97%。用深度学习法识别出船舶类型,可用于遥感数据中船舶类型的识别。The research on civilian ship image classification methods based on YOLOv8 continues to improve the level of shipping.The types of ships are gradually refined,and in the face of different uses of ships,accurate understanding of ship types provides conditions for people to better manage ships.This paper selects the FGSCR dataset.Eight types of crane ships,superyacht,cargo ships,container ships,tugs,small yachts,sand carriers and oil tankers are selected.Implement civil ship image classification based on YOLOv8.Train the 5 models of YOLOv8,YOLOv8n,YOLOv8s,YOLOv8m,YOLOv8l,and YOLOv8x.The training results show that the model size has increased from 2.83 MB to 107MB,with YOLOv8m having the highest Top1 accuracy,reaching 96.97%.The deep learning method can be used to identify ship types in remote sensing data.

关 键 词：YOLOv8 目标分类 船舶类型 遥感影像 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] U675.79[自动化与计算机技术—计算机科学与技术] U692[交通运输工程—船舶及航道工程]