Real-Time Powerline Detection System for an Unmanned Aircraft System

This paper will explore the development of a powerline detection system using Deep Learning to autonomously recognize powerlines in real-time using an Unmanned Aerial System. Additionally, the detection system can identify the individual components of the powerline and electric utility pole such as the cross-arms, insulators, transformers, and primary wires. This proposed model has fast recognition speed and high accuracy, which allows the Unmanned Aerial System to inspect quickly; decreasing both time and cost and increasing safety. To achieve this the Real-Time Powerline Detection System leverages the capability of the YOLACT algorithm, a recently proposed real-time instance segmentation model that achieves 29.8 minimum average precision on the MSCOCO Dataset at 33.5 fps. The results using this approach is promising and perform much better when compared to the Mask R-CNN algorithm. This paper demonstrates the application and results of the Real-Time Powerline Detection System in combination with the YOLACT algorithm for detecting individual components of the powerline using instance segmentation.