基于YOLOv5算法的多尺度小目标船舶识别方法  

Multi-Scale Small Target Ships Recognition Method Based on YOLOv5 Algorithm

作　　者：杨俊秀 王荣杰[1,2] 林安辉 王亦春[1,2] 曾广淼 蒋德松[1,2] YANG Junxiu;WANG Rongjie;LIN Anhui;WANG Yichun;ZENG Guangmiao;JIANG Desong(School of Marine Engineering,Jimei University,Xiamen 361021,China;Fujian Provincial Key Laboratory of Naval Architecture and Ocean Engineering,Xiamen 361021,China)

机构地区：[1]集美大学轮机工程学院,福建厦门361021 [2]福建省船舶与海洋工程重点实验室,福建厦门361021

出　　处：《集美大学学报（自然科学版）》2024年第4期344-357,共14页Journal of Jimei University：Natural Science

基　　金：机器视觉检测安徽省重点实验室开放基金项目“基于跨模态迁移的近海无人艇水面目标识别方法研究”(KLMVI-2023-HIT-14);福建省自然科学基金项目“船用真空制淡/制冷一体机优化匹配及智能调控研究”(2020J01689);福建省教育厅项目“基于深度强化学习的船舶自主靠泊运动控制方法研究”(JAT220173)。

摘　　要：为提高海面多尺度小目标船舶的识别性能,提出一种数据集划分方法,并在YOLOv5算法中改进数据增强方法,融合注意力机制,改进损失函数。实验结果表明,该方法能更好地识别海面上的多尺度小目标船舶,平均精度(mAP)、精确率(P)、召回率(R)分别为99.1%,98.5%,97.5%,识别性能比经典深度学习算法和近几年的方法都高。A dataset division method was proposed to improve the recognition performance of multi-scale small target ships on the sea surface.The data augmentation method,which contains the YOLOv5 algorithm,was improved by integrating the attention mechanism and improving the loss function.Experimental results showed that this method could better identify multi-scale small target ships on the sea surface,with the average precision(mAP),precision(P)and recall(R)are 99.1%,98.5%and 97.5%,respectively.It suggested that the detection performance was higher than that from classical deep learning algorithms and methods in the past several years.

关 键 词：船舶 多尺度小目标 图像识别 数据增强 YOLOv5算法 

分 类 号：U675.79[交通运输工程—船舶及航道工程] TP391.4[交通运输工程—船舶与海洋工程]