电力学报2025，Vol.40Issue(2)：75-86,12.DOI:10.13357/j.dlxb.2025.009
基于改进YOLOv8的半监督变电站仪表检测算法
A Semi-supervised Substation Instrument Detection Algorithm based on Improved YOLOv8
摘要
Abstract
Currently,substation instrument detection faces challenges such as varying target scales,blurry fea-ture information,and high costs associated with data annotation.To address these issues,a semi-supervised in-strument detection algorithm using an improved YOLOv8 is proposed to seek solutions.The algorithm optimiz-es the backbone network by replacing the C2f module with the C3K2 module to enhance feature extraction capa-bilities and introduces the ECA attention mechanism to improve feature representation and reduce the risk of overfitting.In the neck network,a small object detection layer is added to optimize information transmission and fusion,reducing the issues of missed and false detections.Additionally,the Efficient Teacher semi-super-vised learning algorithm is improved.By optimizing the loss function,the learning ability from unlabeled data is enhanced,improving detection accuracy and speed.Experimental results show that on a custom substation in-strument dataset,PmAP50 of the improved algorithm is 93.38％and FFPS is 41.26 frames per second,absolute in-crease amounts are 5.63 percentage points and 7.93 frames per second,respectively,compared to the baseline model.The algorithm proposed demonstrates good application potential in the field of substation instrument de-tection,providing a solid foundation for subsequent research on meter recognition and reading.关键词
变电站/仪表检测/改进YOLOv8算法/高效通道注意力机制/半监督学习Key words
substation/instrument detection/improved YOLOv8 algorithm/efficient channel attention(ECA)/semi-supervised learning分类
信息技术与安全科学引用本文复制引用
刘星宇,樊绍胜,罗强..基于改进YOLOv8的半监督变电站仪表检测算法[J].电力学报,2025,40(2):75-86,12.基金项目
国家自然科学基金(62473065) （62473065）
国家自然科学基金(62473065). （62473065）
