Abstract
This thesis focused on detecting defects of solar panels in a solar photovoltaic plant using deep learning models. Solar panels represent a component in the plant which are vulnerable to defects caused by various weather conditions and posing challenges for effective maintenance. Traditional visual inspection methods used for defects identification are often inadequate, especially in large installations. Hence, specifically, the study involves the segmentation of solar panels from the background in photovoltaic systems using SegFormer model. Then, the application of six YOLO variants models were applied on the defective and non-defective panels dataset. Among the YOLO variants, the YOLOS model demonstrates better performance, and further optimization is achieved through adjusting weights of the networks, giving the YOLOS-PV model. Further to the panel’s defects, the research investigates the application of twelve deep learning models to classify defects of solar cells. By leveraging electroluminescence phenomenon, these models prove to be more effective in defect classification. In summary, this work experimented the use of deep learning models for defect detection in solar panels and extending to defects classification in solar cells