基于改进YOLOv8的船舶目标检测算法研究OA
Research on ship target detection algorithm based on improved YOLOv8
针对复杂海洋环境下船舶目标检测存在的小目标漏检率高、背景干扰严重以及多尺度目标检测精度不足等问题,本文提出了一种改进的 YOLOv8 船舶检测算法.首先,在原有检测头基础上增加 160×160分辨率的 P2 微小目标检测层;其次,在颈部网络的特征融合节点嵌入 CBAM(Convolutional Block Attention Module)注意力机制;最后,引入 Focal-EIoU 损失函数替代传统损失函数,通过动态调节样本权重提升模型对困难样本的学习能力,改善边界框定位精度.在 SeaShips 数据集上的实验结果表明,改进算法的mAP@0.5 达到 90.2%,mAP@0.5:0.95 达到 68.4%,相比原始 YOLOv8 分别提升 3.4 和 4.6 个百分点,其中小目标检测精度提升达 8.5 个百分点.实验验证了该方法在保持实时性的同时显著提升了船舶检测的准确率与鲁棒性,为海事安全监管提供了有效的技术支持.
To address the challenges of high missed detection rates for small targets,severe background interference,and insufficient multi-scale detection accuracy in ship detection under complex marine environments,this paper proposes an improved YOLOv8-based ship detection algorithm.First,a P2 micro-scale detection layer with a resolution of 160×160 is added to the original detection heads.Second,a Convolutional Block Attention Module(CBAM)is embedded into the feature fusion nodes of the neck network to enhance target feature representation.Finally,the traditional loss function is replaced with the Focal-EIoU loss function,which dynamically adjusts sample weights to strengthen the model's ability to learn from hard examples and improve bounding box localization accuracy.Experimental results on the SeaShips dataset show that the improved algorithm achieves an mAP@0.5 of 90.2%and an mAP@0.5:0.95 of 68.4%,representing improvements of 3.4 and 4.6 percentage points over the original YOLOv8,respectively,with an 8.5 percentage point increase in small-target detection accuracy.These results demonstrate that the proposed method significantly enhances the accuracy and robustness of ship detection while maintaining real-time performance,providing effective technical support for maritime safety supervision.
何海辉
广州蓝海豚游船有限公司,广州 510045
信息技术与安全科学
船舶检测YOLOv8CBAM注意力机制小目标检测深度学习
ship detectionYOLOv8CBAM attention mechanismsmall target detectiondeep learning
《船电技术》 2026 (10)
19-23,5
评论