A Deep Learning Framework for Vegetation Management for Electrical Utilities Using Multispectral High-Resolution Satellite Imagery
Abstract
Vegetation management around power lines continues to be a major challenge for utility companies. Overgrown trees frequently produce short circuits, power blackouts and can potentially be a source of wildfires. Utility companies employ visual inspection such as observation from helicopters or driving along transmission paths to identify trees that are growing into danger zones. Not only is this labor intensive, but it is also a costly way of vegetation management for these companies.
In this work, we demonstrate an accurate and scalable method of vegetation management along transmission lines by combining machine learning with high-resolution multispectral satellite imagery and medium resolution multi spectral satellite imagery. In particular, we use deep learning models trained on aerial imagery from the National Agriculture Imagery Program (NAIP) to detect the tree types, estimate tree heights, and determine the health of trees in close proximity to transmission lines. The ground truth established from the NAIP data is combined with weekly medium resolution Sentinel imagery to track tree health and canopy growth. The vegetation management approach encompasses multiple steps: First, we demonstrate tree detection using the Normalized Difference Vegetation Index (NDVI), and estimate a tree's height from the diameter of its canopy. Second, we demonstrate how tree growth or health can be tracked over time using NDVI, and verify our results with a real world dataset from a section of the state of California. Third, we demonstrate tree type detection by training a Deep Belief Network. Finally, we present a novel risk scoring for trees endangering transmission and distribution line operation.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFMEP51E1873K
- Keywords:
-
- 1824 Geomorphology: general;
- HYDROLOGYDE: 1942 Machine learning;
- INFORMATICSDE: 4217 Coastal processes;
- OCEANOGRAPHY: GENERALDE: 4558 Sediment transport;
- OCEANOGRAPHY: PHYSICAL