Scholarlink

<p>Aiming at the problems of low detection accuracy and poor generalization ability of underwater disease targets in inland waterways, an underwater disease target detection algorithm for inland waterways based on improved YOLOv5 is designed, which is denoted as YOLOv5-GBCE. Firstly, Bi-directional Feature Pyramid Network (BiFPN) is used to strengthen feature fusion and improve the accuracy of small target recognition. Secondly, the coordinate attention (CA) module is introduced to allocate attention resources to key areas, so as to reduce the interference of complex background in underwater environment. Then, EIoU is used as the frame loss function to speed up the network convergence rate and solve the problem of difficult and easy sample imbalance. Finally, the Ghost convolutional network is used to reduce the complexity and computation of the model. In order to verify the feasibility of the algorithm, based on the underwater disease data set collected by the project team, a variety of algorithms were selected for comparison and ablation experiments were designed to study the improvement and improvement effect of each module. The research results show that compared with the YOLOv5 s algorithm, the improved algorithm YOLOv5-GBCE improves the average accuracy (IOU = 0.5) of the algorithm by 6.2%, reaching 89.8%, and the detection speed reaches 78.25 FPS. The amount of model calculation is reduced by 25.9%, and the detection energy is detected in complex environments and small target scenarios.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Research on underwater disease target detection method of inland waterway based on deep learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年4月23日

查看全文
  • Tao Yu,
  • Yu Xie,
  • Jinsong Luo,
  • Wei Zhu,
  • Jie Liu

摘要

Aiming at the problems of low detection accuracy and poor generalization ability of underwater disease targets in inland waterways, an underwater disease target detection algorithm for inland waterways based on improved YOLOv5 is designed, which is denoted as YOLOv5-GBCE. Firstly, Bi-directional Feature Pyramid Network (BiFPN) is used to strengthen feature fusion and improve the accuracy of small target recognition. Secondly, the coordinate attention (CA) module is introduced to allocate attention resources to key areas, so as to reduce the interference of complex background in underwater environment. Then, EIoU is used as the frame loss function to speed up the network convergence rate and solve the problem of difficult and easy sample imbalance. Finally, the Ghost convolutional network is used to reduce the complexity and computation of the model. In order to verify the feasibility of the algorithm, based on the underwater disease data set collected by the project team, a variety of algorithms were selected for comparison and ablation experiments were designed to study the improvement and improvement effect of each module. The research results show that compared with the YOLOv5 s algorithm, the improved algorithm YOLOv5-GBCE improves the average accuracy (IOU = 0.5) of the algorithm by 6.2%, reaching 89.8%, and the detection speed reaches 78.25 FPS. The amount of model calculation is reduced by 25.9%, and the detection energy is detected in complex environments and small target scenarios.