南昌工程学院学报2025，Vol.44Issue(1)：91-97,7.
基于改进YOLOv5的河湖"四乱"目标检测算法
Four chaos target detection algorithm for rivers and lakes based on improved YOLOv5
摘要
Abstract
In response to the problem of low target detection accuracy and difficulty in detecting small targets in remote sens-ing images of river and lake four chaos,an improved YOLOv5-based algorithm for detecting four chaos targets in rivers and lakes is proposed.Firstly,based on YOLOv5s,diversified branch blocks are used to replace the original C3 modules to en-hance the feature extraction capability of the backbone network.Secondly,a dynamic detection head with a unified attention mechanism is introduced to strengthen the algorithm's target detection and localization capabilities.Finally,a new loss func-tion named ShapeIoU is designed to reduce the impact of large differences in target scales and enhance the model's general-ization ability.The proposed algorithm is compared with several other representative algorithms.The experimental results show that the proposed algorithm demonstrates good detection performance on both the self-built river and lake four chaos dataset and the RSOD and NWPU VHR-10 public datasets,which provide a certain reference value for the deployment and application of river and lake four chaos target detection.关键词
河湖"四乱"/目标检测/多样化分支块/动态检测头/损失函数Key words
river and lare four chaos/object detection/diversified branch blocks/dynamic detection head/loss function分类
信息技术与安全科学引用本文复制引用
谢敏,王文丰,周波,曾斌,王超民,陈涛..基于改进YOLOv5的河湖"四乱"目标检测算法[J].南昌工程学院学报,2025,44(1):91-97,7.基金项目
国家自然科学基金资助项目(61962036) （61962036）
江西省水利厅科技重点项目(202325ZDKT17,202426ZDKT13) （202325ZDKT17,202426ZDKT13）
