YOLOv4-tiny的绝缘子缺陷检测算法
The Insulator Defect Detection Algorithm Based on YOLOv4-tiny
刘维娜 1钟宇宁 2余兆钗3
作者信息
- 1. 福建信息职业技术学院 物联网与人工智能学院,福建 福州 350008
- 2. 国网武平县供电公司,福建 龙岩 364300
- 3. 福建省信息处理与智能控制重点实验室,福建 福州 350121;闽江学院计算机与大数据学院,福建 福州 350121
- 折叠
摘要
提出一种基于YOLOv4-tiny的绝缘子缺陷检测算法,该算法在YOLOv4-tiny的特征提取网络中加入有效通道注意力网络,明显增强从主干网络中提取的特征质量.在特征融合阶段,将原本的FPN改进成为两条特征融合路径双向特征金字塔结构,使不同尺度特征之间能够更加充分的融合.最后在损失函数的设计上,使用能够解决检测过程中出现的正负样本数量不均衡问题的Focal损失代替二元交叉熵损失函数.实验结果表明:所提算法在平均分类精度和漏检误检方面有较大的提升,性能表现优异.
Abstract
The insulator is one of the components that are prone to failure in the power system,and the failure will seriously affect the power supply safety of the power network.Therefore,how to identify the faulty insulators has become an important condition for the safe operation of power network.However,the insulator is a small component in the power system,which causes poor results of the existing de-tection methods.To solve these problems,an YOLOv4-tiny based defect detection algorithm for insulators is presented.This algorithm ap-plies the effective channel attention network to the feature extraction network of YOLOv4-tiny,and significantly enhances the feature ex-traction capability of the backbone network.In the feature fusion stage,the original FPN is improved into a two-way feature pyramid structure of two feature fusion paths,which enables more full fusion between different scale features.Finally,choose Focal loss instead of the binary cross-entropy loss function as loss function,which can solve the problem of unbalanced number of positive and negative sam-ples in the detection process.The experimental results show that,the proposed algorithm improves the average classification accuracy and miss detection,and performs well.
关键词
绝缘子/电网运行安全/YOLOv4-tiny/特征金字塔/Focal损失Key words
insulator/safety of power network operation/YOLOv4-tiny/feature pyramid/focal loss引用本文复制引用
出版年
2024
NETL
NSTL
维普
万方数据