Automating the Detection of Surface Defects on Solar Panels in Zambia

The increasing adoption of solar photovoltaic (PV) systems has created a growing need for effective methods of monitoring and maintaining solar panels to ensure reliable energy generation. PV panels can experience various defects and surface conditions, including bird droppings, dust accumulation, electrical damage, and physical damage, which can reduce their performance and, in some cases, lead to further damage if not detected and addressed in time. Conventional inspection methods often rely on manual visual inspection, which can be time-consuming, labour-intensive, and difficult to scale as the number of solar installations increases. This project presents the development of an automated solar photovoltaic panel defect detection and classification system using deep learning and computer vision. The system employs two image classification models based on the ResNet18 convolutional neural network architecture. The first model performs binary classification to distinguish between healthy and faulty solar panels, while the second model classifies faulty panels into four defect categories: bird-drop, dusty, electrical-damage, and physical-damage. The proposed system provides a practical approach to supporting solar panel inspection by reducing dependence on manual identification of visible defects and enabling faster preliminary assessment of panel conditions. The project demonstrates how deep learning and computer vision can be applied to support the maintenance and monitoring of photovoltaic systems, with potential relevance to the continued development of solar energy infrastructure in Zambia and other African countries.