Extraction of Overhead Transmission Towers from UAV Images

Abstract

To ensure the integrity of power lines, electrical transmission towers must be monitored. Monitoring vegetation encroachment, which can lead to power outages, is a significant challenge. The majority of current monitoring techniques rely on manual labor and traditional methods of observation such as unmanned aerial vehicles (UAV) and airborne photography. Monitoring large areas with these methods, however, is expensive and time consuming. Our paper describes a method for monitoring power line corridors with UAV images. A two-stage procedure is proposed. Background clustering was performed using Fuzzy C-means in the first stage. Our second step was to detect the presence of transmission towers using state-of-the-art deep learning technologies AlexNet and DenseNet-121. By comparing the two deep learning architectures, the proposed methodology detects the transmission tower from VAV images with an accuracy of 94.8% for AlexNet and 98.6% for DenseNet - 121 with better precision, recall, and F1-score.