| 联合YOLOv4和迁移学习的侧扫声纳图像沉船检测 |
| |
| 引用本文： | 于永灿,李永奎,龚权华,李应超.联合YOLOv4和迁移学习的侧扫声纳图像沉船检测[J].海洋测绘,2021,41(4):38-42. |
| |
| 作者姓名： | 于永灿 李永奎 龚权华 李应超 |
| |
| 作者单位： | 武汉大学 测绘学院,湖北 武汉 430079;武汉大学 海洋研究院,湖北 武汉 430079;交通运输部 东海航海保障中心 上海海事测绘中心,上海 200090;中交三航(上海)新能源工程有限公司,上海 200137;海军士官学校 航海系,安徽 蚌埠 233000 |
| |
| 基金项目： | 国家自然科学基金(41576107)；中交三航局A类课题(2020-04) |
| |
| 摘 要： | 为解决侧扫声纳(SSS)图像沉船检测中样本不足、代表性弱等带来的检测精度低的问题,提出了一种联合YOLOv4和迁移学习的SSS图像沉船检测方法。首先,基于SSS成像机理实现了SSS沉船图像样本扩增,解决样本少而无法开展沉船检测模型构建的难题;然后,利用迁移学习,将公共数据集上学习到的权重和沉船通用性特征引入YOLOv4网络,构建高性能沉船检测模型。试验表明,构建的沉船检测模型取得了85.5%的类平均精度(mAP),将传统方法的检测精度提升了7.7%,在少样本情况下实现了沉船的高精度检测。
|
| 关 键 词： | 水下目标检测 侧扫声纳图像 深度学习 迁移学习 数据增强 |
Shipwreck detection in side-scan sonar images using YOLOv4 with transfer learning |
| |
| YU Yongcan,LI Yongkui,GONG Quanhu,LI Yingchao.Shipwreck detection in side-scan sonar images using YOLOv4 with transfer learning[J].Hydrographic Surveying and Charting,2021,41(4):38-42. |
| |
| Authors: | YU Yongcan LI Yongkui GONG Quanhu LI Yingchao |
| |
| Abstract: | In order to solve the problem of low accuracy caused by insufficient samples and weak representativeness in SSS image shipwreck detection,a shipwreck detection method in SSS images combining YOLOv4 with transfer learning is proposed in this paper.Firstly,based on the SSS imaging mechanism,SSS images amplification was implemented to solve the problem that the shipwreck detection model could not be built due to the small number of samples.Then,transfer learning was used to introduce the weights and universality features of shipwrecks learned from the public data sets into YOLOv4 network to build a high-performance shipwreck detection model.The experiment results show that the built shipwreck detection model achieves 85.5% mean average precision (MAP) and 7.7% higher than that of the traditional method,which achieves high detection precision with fewer samples. |
| |
| Keywords: | |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
| 点击此处可从《海洋测绘》浏览原始摘要信息 |
| 点击此处可从《海洋测绘》下载免费的PDF全文 |
|
