Solar panels play an important role in renewable energy by reducing greenhouse gas emissions. However, their performance is degraded due to various outer (cracks, hotspots) and environmental factors (dust, bird droppings, snow, etc), requiring continuous inspection and monitoring. A practical solution is an unmanned aerial vehicle (UAV)-based monitoring system that can fly and cover large PV (photovoltaic) farms. The current inspection methods are mainly based on color red-green-blue (RGB) or thermal cameras. However, these techniques encounter lower accuracy due to changing lighting environmental conditions, low resolutions, and loss of details. Multispectral cameras offer superior performance but are generally expensive and unsuitable for UAV platforms. To address these challenges, this paper introduces a method to generate multispectral images from standard RGB images captured by UAVs for segmentation problems. The developed end-to-end pipeline efficiently segments solar panels and classifies faults, outperforming existing methods in accuracy and computational complexity, including those designed for multispectral images. The main contributions of this work are a new reflectance-based multispectral decomposition method, efficient lightweight segmentation and fault classification networks, and a generalizable pipeline that can adapt to other domains. Strong evaluation and validation are conducted towards the proposed solution’s precision, efficacy, and universality.
Hayk Gasparyan (Mon,) studied this question.