摘要
针对目前绝缘子缺陷检测算法检测精度与速度不平衡以及对小目标绝缘子缺陷检测效果不佳等问题,提出一种融合多尺度特征的轻量级YOLOv7绝缘子缺陷检测算法。以YOLOv7为基础框架,使用CA-GhostNet作为主干网络;将头部预测网络中的残差卷积替换为深度可分离卷积;在颈部网络设计Light-SPPCSPC特征提取模块;在特征金字塔部分将不同尺度的特征图融合。实验结果表明,所提算法实现了精度与速度的平衡,降低了绝缘子缺陷的漏检率。
A lightweight YOLOv7 insulator defect detection algorithm that integrates multi-scale features is proposed to address the issues of imbalanced detection accuracy and speed in current insulator defect detection algorithms,as well as poor detection performance for small target insulator defects.Using YOLOv7 as the basic framework and CA HostNet as the backbone network;Replace the residual convolution in the head prediction network with deep separable convolution;Design a Light-SPPCSPC feature extraction module in the neck network;In the feature pyramid section,fuse feature maps of different scales.The experimental results show that the proposed algorithm achieves a balance between accuracy and speed,reducing the missed detection rate of insulator defects.
作者
党宏社
许勃
张选德
DANG Hongshe;XU Bo;ZHANG Xuande(College of Electrical and Control Engineering,Shaanxi University of Science and Technology,Xi'an 710021,China)
出处
《电瓷避雷器》
CAS
北大核心
2023年第6期187-195,共9页
Insulators and Surge Arresters
基金
国家自然科学基金项目(编号:61871206)
陕西省科技厅自然科学基金项目(编号:2020JM-509)。
关键词
绝缘子缺陷
轻量化
空间金字塔
多尺度特征融合
insulator defects
lightweight
pyramid of space
multi-scale feature fusion