Overhead power line detection from aerial images using segmentation approaches

Ensuring the optimal efficiency of electrical networks requires vigilant surveillance and preventive maintenance. While traditional methods, such as human patrols and helicopter inspections, have been longstanding practices for grid control by electrical power distribution companies, the emergence of Unmanned Aerial Vehicles (UAV) technology offers a more efficient and technologically advanced alternative. The proposed comprehensive pipeline integrates various elements, including preprocessing techniques, deep learning (DL) models, classification algorithms (CA), and the Hough transform, to effectively detect powerlines in intricate aerial images characterized by complex backgrounds. The pipeline begins with Canny edge detection, progresses through morphological reconstruction using Otsu thresholding, and concludes with the development of the RsurgeNet model. This versatile model performs binary classification and feature extraction for power line identification. The Hough transform is employed to extract semantic powerlines from intricate backgrounds. Comparative assessments against three existing architectures and classification algorithms highlight the superior performance of RsurgeNet. Experimental results on the VL-IR dataset, encompassing both visible light (VL) and infrared light (IR) images validate the effectiveness of the proposed approach. RsurgeNet demonstrates reduced computational requirements, achieving heightened accuracy and precision. This contribution significantly enhances the field of electrical network maintenance and surveillance, providing an efficient and precise solution for power line detection.