Applying RetinaNet Machine Learning Models for Hot-Spot Detection in Thermal Images of Photovoltaic Panel

Abstract

In recent years, the remarkable development of the solar power industry has been undeniable. This led to the necessity of monitoring and maintenance of solar power plants due to many issues and failures that affect the efficiency and lifespan of the system. One of the most common fault cases of solar systems is the hot spot in photovoltaic panels created by the mismatch of cells in the panel due to incompatible environmental conditions or physical damage causing the heat up in lower power produced cells. The hot spot is detected through thermal image capture by a thermal camera and analyzed through image processing techniques with manual adjustment and testing. With recent significant achievements and growth of machine learning methods in image processing and technical diagnosis, it is greatly possible to develop a machine learning model to perform hot spot detection for higher efficiency and automation. In this study, our research group proposes an application of RetinaNet to develop a model capable of detecting hot spots in photovoltaic panels through processing thermal images. © 2025 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.