Integrating uncrewed aerial systems and repeat station imaging for inspection and monitoring of electric utility infrastructure

Abstract

Repeat station imaging (RSI) pertains to capturing aerial images repetitively over time from the same location in the sky based on aircraft navigation and sensor triggering systems that rely on survey-grade global navigation satellite system (GNSS) signals. Time sequential images captured from the same imaging stations are co-registered, enabling fine-scale changes to be detected as anomalies on a frame-by-frame basis. This effectively replicates the situation for stationary surveillance cameras at multiple imaging stations in the sky. RSI has been demonstrated to be effective in accurately detecting people and vehicles for law enforcement and for post-hazard assessment of damage to critical infrastructure. Here we assess the effectiveness of integrating uncrewed aerial systems (UAS) and RSI for detecting ultra-fine scale defects in electric utility infrastructure. We conducted UAS-RSI imaging missions for electric distribution poles and transmission towers, with simulated damage introduced to image pairs. UAS flight plans were generated automatically and optimized to minimize the number of captured images. An automatic image registration routine and a novel deep learning algorithm for detecting defects were applied to matched RSI image frames. Registration accuracy of a few pixels is achieved, where the ground sampling distance associated with a pixel is approximately 1 mm. Defect detection accuracy with the novel, time-based deep learning algorithm is nearly 98%. We demonstrate how a graphical user interface which displays RSI image pairs and automatic damage detection results enables an inspector/analyst to accurately and efficiently detect defects for maintenance and wildfire mitigation purposes.