Oil-surface Thermometers Recognition Method Using Multi-Task Cascade Neural Network  

作　　者：Weisheng Lao Bin Sun Yuxiao Zhao Chaoping Sun Xu Dai 

机构地区：[1]China Jiliang University,Hangzhou,Zhejiang 310018,China [2]Huzhou Quality and Technical Supervision and Testing Research Institute,Huzhou,Zhejiang 313000,China

出　　处：《Instrumentation》2024年第4期13-20,共8页仪器仪表学报(英文版)

基　　金：funded by the Zhejiang Provincial Administration for Market Regulation's Early Career Development Program,grant number CY2023323.

摘　　要：Currently,the oil-surface thermometer remains a crucial monitoring device in substations.However,challenges such as uneven scale distribution,inadequate model accuracy,and low reading precision persist in handling readings from various types of oilsurface thermometers.This paper presents a high-precision recognition method based on neural networks for accurately reading various types of oil-surface thermometers.The proposed recognition method mainly comprises the YOLOv5 convolutional network,the attention U2-Net semantic segmentation network,and enhanced computational methods.This detection method demonstrates ultra-high precision and can be applied to different types of oilsurface thermometers.Compared to existing methods,the proposed detection method exhibits outstanding accuracy and effectiveness.A comprehensive set of tests was performed to assess the viability and reliability of the approach.Findings reveal that the maximum error margin in its measurements remains below 0.13%.

关 键 词：oil-surface thermometer enhanced computational methods semantic segmentation object detection 

分 类 号：TN9[电子电信—信息与通信工程]