摘要
为了有效管理海上交通、实施海上救援和保护海洋环境,需要精确地掌握海上船只目标的位置和分布情况,但传统的检测方法(如CFAR)往往会出现船只细节丢失和小目标漏检的情况。为了解决以上问题,将YOLOv5模型进行改进。首先通过数据增强,提升数据的多样性,进而提高模型的泛化能力;之后加入SE注意力机制和小目标检测层来增强模型对船只的特征提取能力。实验结果表明,加入SE注意力机制和小目标检测层后,平均准确度mAP分别提高了2%和3.1%,可以有效改善船只密集分布、沿岸分布等不同场景下的检测准确率,实现整体准确率的提高。
To effectively manage maritime traffic,implement sea rescue operations,and protect the marine environment,precise monitoring of the positions and distributions of maritime vessels is crucial.However,traditional detection methods like CFAR often result in loss of ship details and missed detections of small targets.In order to solve these issues,an improved YOLOv5 object detection method is proposed.Initially,data augmentation enhances dataset diversity,thereby improving model generalization.Subsequently,SE attention mechanism and small object detection layer are incorporated to enhance the model’s capability to extract ship features.Experimental results demonstrate that integrating SE attention mechanism and small object detection layer increases the average precision mAP by 2%and 3.1%,respectively.This approach effectively enhances detection accuracy in scenarios such as dense ship clustering and coastal distribution,leading to overall accuracy improvement.
作者
龙莹莹
余华云
杨武
殷俊凯
LONG Yingying;YU Huayun;YANG Wu;YIN Junkai(Yangtze University,Jingzhou,Hubei,China 434023)
出处
《湖南邮电职业技术学院学报》
2024年第3期56-60,共5页
Journal of Hunan Post and Telecommunication College
基金
2024年度湖北省重点实验室开放基金项目“机器学习在页岩气井井筒积液预测及泡沫排水适用性诊断中的应用研究”(项目编号:YQZC202402)。
