HIGH VOLTAGE ELECTRIC POLE DETECTION USING DEEP LEARNING

Abstract

High voltage electric pole infrastructure is an asset on which several lives depend. Due to extreme weather conditions these poles deteriorate over time. Hence power companies need to carry out their timely maintenance. There have been instances of natural disasters where lack of database of utility infrastructure has led to fatalities due to slow inspection and recovery processes. Thus, timely maintenance is important for uninterrupted operation and utility infrastructure planning purposes. Conventional methods of inspection are slow, inefficient and involve high computational cost. Some methods involve manual identification of poles in an image, and since the electric poles are spread over wide areas, it is suggested to automate the process. Other methods that have come up are feature-based that use shape and color of the poles to distinguish it among other objects. These methods suffer from problems such as different backgrounds, noise, shapes, sizes etc. of the electric poles in an image. Deep learning techniques have proved to be quite robust while handling such scenarios. Thus, in this project deep learning techniques were explored to identify a pole in oblique images taken from drone and google earth pro. Another advantage of using these images is that a single image covers a wider area and, in an image, multiple poles could be detected, therefore inspection over an area could be done with a smaller number of images. Through this study, it could be inferred that detection of electric poles using drone images is a viable method for autonomous identification of poles.