改进YOLOv8的无人机小目标检测方法
Detection method of UAV at small target based on improved YOLOv8
刘付刚 1刘巾瑞 1祝永涛2
作者信息
- 1. 黑龙江科技大学 电子与信息工程学院,哈尔滨 150022
- 2. 黑龙江龙煤双鸭山矿业有限公司,黑龙江 双鸭山 155199
- 折叠
摘要
针对无人机视角拍摄的小目标具有分布聚集、数量繁多、类别不均衡等特点,致使小目标检测精度较低,容易出现漏检和误检的问题,提出了基于改进YOLOv8 的目标检测算法.通过添加小目标特征融合网络的方式优化网络结构,引入可变形卷积来提升模型对感兴趣区域的关注能力,采用MPDIoU损失函数,提高边界框回归的准确性.结果表明,改进后的YOLOv8 检测算法在VisDrone2019 数据集上的精度提升了6.1%,模型参数量减少了25.3%,在轻量化网络的同时有效提高了小目标检测精度.
Abstract
This paper aims to address the low detection accuracy,missed detection and false detec-tion at the small targets photographed by UAV with the characteristics of distribution clustering,large number,and unbalanced categories,and proposes a target detection algorithm based on improved YOLOv8.The study involves optimizing the network structure by adding a small target feature integrated network;introducing the deformable convolution to improve the ability of the model at the region focused;and improving the accuracy of bounding box regression by using MPDIoU loss function.The results show that the improved YOLOv8 detection algorithm improves the accuracy of the VisDrone2019 dataset by 6.1%,and the model parameters are reduced by 25.3%,as which effectively improves the accuracy of small target detection while lightweighting the network.
关键词
小目标检测/YOLOv8/可变形卷积/损失函数Key words
small object detection/YOLOv8/deformable convolution/loss function引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据