Lightweight algorithm of insulator identification applicable to electric power engineering
Zhiqiang Xing, Xi Chen · Energy Reports · 2022
In order to reach actual-time viewing of insulator identification, this paper comes up with a lightweight model based on MobileNet-YOLOv4, which replaces the YOLOv4 backbone network CSPDarknet53, making the network structure lightweight enough for easier deployment on mobile devices Meanwhile, Gaussian filtering and K-Means++ clustering, Mosaic data augmentation are applied for pre-processing on the datasets to improve detection accuracy. This model is adopted to identify and test the insulators on the transmission line, and the results show that the average insulator identification precision of this model is 97.78%. Average detection speed of the model is 0.168s faster than YOLOv4, and model size is 20.91% lower than YOLOv4, which can meet the purpose of real-time monitoring.