Mask R-CNN Powerline Detector: A Deep Learning approach with applications to a UAV

Abstract

This paper explores the use of a UAV for power-line inspection using a Deep Learning algorithm for detection. Unmanned Aerial Vehicles (UAVs) are versatile with the potential to revolutionize multiple business domains, with benefits being discovered both inside and outside the world of utility and energy industries. They can reduce operational costs, improve worker safety, asset reliability, and decision making, also having semiautonomous flight capability. Utility and Energy companies have successfully deployed UAVs in inspections. Power-line companies usually perform regular visual inspections to check the status of their transmission lines mainly using helicopters equipped with external gimbals housing infrared and an ultraviolet camera to detect hot spots and corona discharges. This solution is quite expensive and dangerous for the crew. To resolve this, a powerline-detection segmentation algorithm based on transfer learning and an improved mask regional convolutional neural network (Mask RCNN), Mask RCNN Powerline Detector is proposed and is deployed on an UAV. For this Draganfly XP-4 was used as the UAV platform. The power-line detection system used a Deep Learning Resnet50 architecture as the backbone network, combined with the Feature Pyramid Network (FPN) architecture for feature extraction. The Region Proposal Network (RPN) was trained end-to-end to create region proposals for each feature map. This paper will present the development of the MRPD system, integration, and testing of the UAV.