指挥控制与仿真2025，Vol.47Issue(5)：49-57,9.DOI:10.3969/j.issn.1673-3819.2025.05.007
面向无人艇视角的海上光学图像目标检测算法研究
Research on marine optical image target detection algorithm from the perspective of USV
王玉松 1纪延琚 2陈勤辉3
作者信息
- 1. 海军装备部某中心,北京 100000
- 2. 江苏自动化研究所,江苏 连云港 222061
- 3. 湖南大学电气与信息工程学院,湖南 长沙 410000
- 折叠
摘要
Abstract
During the application of unmanned surface vehicles(USVs)in natural sea areas,issues such as low-quality op-tical image imaging,localized target features and depth-directional distortion caused by side-viewing angles,lens deformation due to saltwater droplets,and the coexistence of extremely large and small targets occur.These phenomena lead to a decline in the performance of deep learning-based image target detection algorithms,resulting in high rates of missed detections and false alarms.Consequently,this causes the collision avoidance decision-making algorithms to diverge or trigger frequent e-mergency alarms.To enhance object detection accuracy for USV optical images,this paper proposes an optimized framework based on the YOLOv8 model.First,the Copy-Paste algorithm and unpaired image style transfer algorithm are introduced to generate multi-view target images and low-quality images caused by adverse environmental conditions,addressing the imbal-ance in the proportion of low-quality images in the dataset.Second,a dedicated detection head for small targets is added to the detection network,and the loss function is optimized to enhance the detection capability for small targets while preserving the original model's performance on larger objects.The newly developed object detection model,trained on the augmented dataset,achieves an average precision of 96.2%on the test set.关键词
无人艇/YOLOv8/数据增广/目标检测Key words
USV/YOLOv8/data augmentation/target detection分类
军事科技引用本文复制引用
王玉松,纪延琚,陈勤辉..面向无人艇视角的海上光学图像目标检测算法研究[J].指挥控制与仿真,2025,47(5):49-57,9.
