Power Transmission Line Classification from Images using Pre-trained Deep Learning Models

Abstract

Identifying the type of a power transmission line from images is a fascinating field. It can lead to applications like tower detection, line inspection, location detection, multi-fitting detection, fault detection, foreign object detection, etc. The data collection involves 348 real-time images captured from the UAV and 696 synthetic images. Based on the geometry of the line, the whole data is classified into three classes: simple, real, and Duplo lines. This paper feeds real and synthetic images to the pretrained deep neural networks (DNN) for the line inspection application. Comprehensively top-32 pretrained models were tested on the dataset and evaluated the classification performance. In addition, the DNN algorithms are also tested on the Raspberry Pi hardware platform to know the run-time feasibility of this real-time application. © 2022 IEEE.