Hybrid CNN architecture for hot spot detection in photovoltaic panels using fast R-CNN and GoogleNet
Due to the continuous increase in global energy demand, photovoltaic solar energy generation and associated maintenance requirements have significantly expanded. One critical maintenance challenge in photovoltaic installations is detecting hot spots, localized overheating defects in solar cells that...
| Autores: | , , |
|---|---|
| Tipo de documento: | artigo |
| Data de publicação: | 2025 |
| País: | España |
| Recursos: | Universidad Autónoma de Madrid |
| Repositório: | Biblos-e Archivo. Repositorio Institucional de la UAM |
| Idioma: | inglês |
| OAI Identifier: | oai:repositorio.uam.es:10486/724500 |
| Acesso em linha: | https://hdl.handle.net/10486/724500 https://dx.doi.org/10.32604/cmes.2025.069225 |
| Access Level: | Acceso aberto |
| Palavra-chave: | photovoltaic panel convolutional neural network deep learning hot spots thermal imaging unmanned aerial vehicle inspection GoogleNet fast regions with convolutional neural networks automated defect detection transfer learning aerial thermography Telecomunicaciones |
| Resumo: | Due to the continuous increase in global energy demand, photovoltaic solar energy generation and associated maintenance requirements have significantly expanded. One critical maintenance challenge in photovoltaic installations is detecting hot spots, localized overheating defects in solar cells that drastically reduce efficiency and can lead to permanent damage. Traditional methods for detecting these defects rely on manual inspections using thermal imaging, which are costly, labor-intensive, and impractical for large-scale installations. This research introduces an automated hybrid system based on two specialized convolutional neural networks deployed in a cascaded architecture. The first convolutional neural network efficiently detects and isolates individual solar panels from high-resolution aerial thermal images captured by drones. Subsequently, a second, more advanced convolutional neural network accurately classifies each isolated panel as either defective or healthy, effectively distinguishing genuine thermal anomalies from false positives caused by reflections or glare. Experimental validation on a real-world dataset comprising thousands of thermal images yielded exceptional accuracy, significantly reducing inspection time, costs, and the likelihood of false defect detections. This proposed system enhances the reliability and efficiency of photovoltaic plant inspections, thus contributing to improved operational performance and economic viability |
|---|