摘要
[目的]为提升智能船舶在自动检测水面漂浮物体方面的稳定性与精确度,提出一种基于改进YOLOv8算法的智能船舶水面多目标视觉检测方法。[方法]为增强模型对不同尺度物体的检测能力,设计DDCNv4模块以替代骨干网络Backbone中的C2f模块。在与Neck层的连接处,引入Bi-Level Routing Attention注意力机制,以强化模型在抗干扰和纹理特征提取方面的性能,优化对小尺寸物体的检测效果。此外,在Neck层进一步集成DSConv模块,以实现模型的轻量化,并兼顾模型性能与计算效率。[结果]实验结果显示,相较于传统的YOLOv5,YOLOv7和YOLOv8等算法,改进YOLOv8算法的检测准确率达90.4%,相比原YOLOv8算法的mAP@0.5提升5.6%,参数量减少0.1×10^(6),模型推理速度每秒提高了1.98帧。[结论]研究表明所提算法在复杂多变的水面环境中能够保持高水平的检测性能,可为目标检测在智能船舶领域的应用提供理论参考。
[Objective]To enhance the stability and accuracy of intelligent ships in the automatic detection of floating objects on the water surface,this study proposes a water surface multi-target detection method of intelligent ship based on an improved YOLOv8 algorithm.[Method]To improve the model's detection capability across varying object scales,a DDCNv4 module is designed to replace the C2f module in the Backbone network.At the connection with the Neck layer,a bi-level routing attention(BRA)mechanism is introduced to enhance the model's anti-interference performance and texture feature extraction,particularly optimizing the detection of small-scale objects.In addition,the DSConv module is further integrated into the Neck layer to achieve a lightweight model architecture that balances the model performance and computational efficiency.[Results]Experimental results show that the proposed improved YOLOv8 algorithm achieves a detection accuracy of 90.4%,outperforming YOLOv5,YOLOv7,and the original YOLOv8.This represents a 5.6%improvement in mAP@0.5 over the original YOLOv8 algorithm,while reducing the number of parameters by 0.1×10^(6) and increasing the inference speed by 1.98 frames per second(FPS).[Conclusion]The prosed algorithm maintains high detection performance in the complex water surface environments,providing a theoretical reference for the application of target detection in the field of intelligent ships.
作者
李昊鸣
衣正尧
朱嘉晟
袁浩宇
石博博
曹杰
LI Haoming;YI Zhengyao;ZHU Jiasheng;YUAN Haoyu;SHI Bobo;CAO Jie(School of Navigation and Naval Architecture,Dalian Ocean University,Dalian 16023,China)
出处
《中国舰船研究》
北大核心
2025年第S1期169-179,共11页
Chinese Journal of Ship Research
基金
2024年辽宁省联合基金资助项目(2023-MSLH-014)。
关键词
智能船舶
卷积神经网络
YOLOv8模型
BRA注意力机制
多目标检测
intelligent ships
convolutional neural network
YOLOv8
bi-level routing attention(BRA)mechanism
multi-target detection
