Deep Learning for Localization of Damaged Photovoltaic Panels with Hotspot Detection

The use of photovoltaic panels has been increasing over the last decade resulting in a growing need for advanced and robust condition monitoring systems. Implementation of photovoltaic panels in a large-scale farm requires predictive maintenance to ensure their reliability and effectiveness. In this study, a deep learning model has been developed for hotspot detection in photovoltaic panels through the processing of thermal images captured by a thermal imaging drone. An advanced convolutional neural network model is used wherein the task of the model is to detect hotspots on the photovoltaic panel, identify the region of interest of the hotspot, and provide a confidence level to determine the quality of hotspots. The associated GPS data will then be employed to geolocate and align these thermal images, creating a comprehensive thermal representation of the entire solar farm. This innovative approach allows for the precise localization of damaged photovoltaic panels within the solar farm, facilitating targeted maintenance and repair interventions, thereby optimizing farm performance and efficiency. This study leverages cutting-edge artificial intelligence and autonomous drone technology to revolutionize solar farm maintenance. Through a well-trained deep learning model, we can achieve precise hotspot detection in photovoltaic panels, minimizing physical interaction for workers and eliminating the need for extensive manual interventions. A deep learning model has been developed that provides hotspot localization with accuracy using bounding box predictions and confidence intervals. The model can reliably identify hotspots in thermal images with accuracy of 80% and confidence level above 90%.