山东电力技术2023，Vol.50Issue(12)：20-25,44,7.DOI:10.20097/j.cnki.issn1007-9904.2023.12.003
基于交互式分割的电力设备红外图像自动标注方法
Interactive Segmentation Based Automatic Infrared Images Annotation for Electrical Equipment
摘要
Abstract
Power equipment failure is a key factor affecting the safety of the power system.During fault diagnosis,a significant step is to segment the targets in infrared images.Aiming at the problem that traditional methods are not robust enough and deep learning methods require a large amount of labeled data for training,this paper proposed an interactive segmentation approach based on a deep learning.This approach is able to utilize a small set of pixel-level image labeling data and simulate the interactive clicks of users to train a coarse-fine integrated interactive segmentation network,thus obtaining a accurate segmentation of power equipment areas in infrared images.Experiment results demonstrate that the proposed method is not only capable of segmenting electric equipment precisely and quickly,but also make corrections in the segmentation of the model by providing more clicks.Moreover,the proposed approach can help users to label power equipment images very efficiently and create a larger amount dataset that is required by other applications.关键词
红外图像/电力设备/图像分割/交互式分割Key words
infrared image/electrical equipment/image segmentation/interactive segmentation分类
计算机与自动化引用本文复制引用
林颖,张峰达,李壮壮,孙艺玮,于文牮..基于交互式分割的电力设备红外图像自动标注方法[J].山东电力技术,2023,50(12):20-25,44,7.基金项目
国网山东省电力公司科技项目"输变电设备运检人工智能开放服务平台关键技术研究"(520626220010).Science and Technology Project of State Grid Shandong Electric Power Company"Research on Artificial Intelligence Open Service Platform for Transmission and Distribution Equipment Operation and Maintenance"(520626220010). （520626220010）
