非对称交叉平均教师_一种半监督语义分割算法及其在变电站表计读数识别中的应用.pdfVIP

第 43 卷 第 8 期 中 国 电 机 工 程 学 报 Vol.43 No.8 Apr. 20, 2023 2023 年 4 月 20 日 Proceedings of the CSEE ©2023 Chin.Soc.for Elec.Eng. 2979 DOI ：10.13334/j.0258-8013.pcsee.213145 文章编号：0258-8013 (2023) 08-2979-11 中图分类号：TM 93 文献标识码：A 非对称交叉平均教师：一种半监督语义分割算法及 其在变电站表计读数识别中的应用 滕国龙，郑易，闫云凤，齐冬莲* (浙江大学电气工程学院，浙江省 杭州市 310027) Asymmetric Cross Mean Teacher: A Semi-supervised Semantic Segmentation Algorithm and Its Application on Substation Meter Reading Recognition * TENG Guolong, ZHENG Yi, YAN Yunfeng, QI Donglian (College of Electrical Engineering, Zhejiang University, Hangzhou 310027, Zhejiang Province, China) ABSTRACT: Aiming at the problem of expensive dataset 当前主流算法。 annotation costs, a semi-supervised semantic segmentation 关键词：半监督语义分割；平均教师；变电站表计；读数 algorithm is proposed. This algorithm is able to utilize a small 识别 set of labeled data and a large set of unlabeled data to train a high-performance segmentation model; therefore a lot of 0 引言 manual annotation costs can be saved. The proposed method 随着电力系统数字化转型的不断深入，智能变 improves the classic mean teacher algorithm by extending the 电站实施并应用了许多与传统变电站不同的运维 single teacher-student branch to a dual-branch structure, where the student models are trained in a cross-supervision manner. In 技术，如视频监控技术、智能机器人巡检技术等[1-2] 。 this way, the model coupling issue of the vanilla mean teacher 在电力设备状态监测方面，以在线监测为核心的状 algorithm is effectively solved.