A Novel Hybrid ML Approach for Powerline - Vegetation Encroachment Area Identification

Accurate and timely identification of vegetation encroachment is essential for powerline monitoring to mitigate outage risks of the electrical infrastructure. Vegetation growth near powerlines poses increasing hazards with proximity, thus necessitating routine inspections to identify high-risk zones. Ad-vances in unmanned aerial vehicles (UAV s) capable of capturing high-resolution aerial images, combined with machine learning (ML), have significantly improved powerline and vegetation detection. Moreover, the recent adoption of transformer-based architectures offers new venues for image segmentation tasks. In this work, we propose a novel hybrid segmentation approach that combines the strengths of traditional convolutional neural networks (CNNs) and transformer-based models, particularly U-Net and SegFormer, through meta-learning for powerline and vegetation detection and encroachment identification. Through extensive experimental evaluation, we prove that the proposed approach achieves an Intersection over Union (IoU) score of 86 % for powerline detection (resp. 94 % for vegetation detection), outperforming (resp. similar to) the IoU scores of standalone U-Net and SegFormer (IoU of standalone SegFormer). Our approach serves as an efficient preliminary assessment tool for powerline-vegetation encroachment identification.