Moroccan researchers develop fast solar panel labeling tool using drones and AI

Moroccan scientists have developed a new method to automatically locate and label solar panels in large photovoltaic installations using thermal images from drones, a tool that helps reduce inspection time and simplifies the training of artificial intelligence models used to monitor solar infrastructure.

The research was carried out by experts from IAV Hassan II, the Green Energy Park, and the National School of Applied Sciences in Oujda. Their findings were published in Cleaner Engineering and Technology and reported by PV Magazine.

The system uses drone metadata such as GPS, camera settings, and inertial measurements to calculate ground distances and identify the exact position of solar modules.

It applies Niblack thresholding to highlight panel shapes in the thermal images and then refines the detection using edge checking and clustering methods.

The information is converted into formats that deep learning systems can use for automatic detection.

The researchers tested the method on two sites in Morocco. One was a 22 kilowatt ground-mounted solar field at the Green Energy Park.

The other was a one megawatt rooftop plant on a data center. The tool achieved a 91 percent accuracy rate in geo-labeling panels and kept false positives low. It also had a geolocation margin of error of just 2.5 meters, which is suitable for on-site inspections.

The labeled image data was used to train several artificial intelligence models. Among them, YOLOv7 showed the highest performance with a mean average precision of 98.3 percent. It processed each image in just 15 milliseconds, making it ideal for real-time monitoring.

The researchers say this method has the potential to improve solar panel maintenance by reducing the need for manual checks and enabling faster detection of issues. It also supports real-time operation, which is essential for managing large solar parks efficiently.