Abstract

Since insulators play an essential part in transmission lines, insulator defect detection could be an important assignments for intelligent inspection of high-voltage transmission lines. In this paper, an improved YOLOv5 algorithm for insulator defect detection task for aerial images with various backgrounds is proposed. We use collected aerial images of insulators with one or more defects in different scenarios, perform data augmentation of exposure and noise on the images to expand the sample (expanded to 2125 images), and establish a dataset by combining aerial images of normal working insulators. By adjusting the CSP(Cross-Stage-partial-connections) and CBL(Convalution-BatchNorm-Leaky_relu) modules in the YOLOv5 model to change the depth and width of the model, change model parameters, and build five different scale YOLOv5 models to further meet the real-time task. ResNet and DarkNet are used for the transfer learning of the YOLOv5 model, and various optimization methods are used in the Backbone structure, Neck structure and output of the model, then the established data set is trained and tested on each YOLOv5 model. Among them, the YOLOv5n model has the fastest detection speed, which can reach 10ms, and the precision also reaches 95%. The YOLOv5x model has the highest precision, reaching 97%, and the detection speed is 21ms. These models are all able to satisfy the accuracy and real-time mission in the process of aerial photography and analysis among which YOLOv5n can achieve lightweight tasks while being efficient enough.

Full Text
Paper version not known

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call