Deep Learning Based Defect Detection in Photovoltaic Cell: Performance Evaluation Across CNN Architectures

Photovoltaic (PV) panels are critical to renewable energy generation but are prone to defects such as black core, thick line, cracks, dislocations, finger interruptions, and material anomalies, which affect efficiency and reliability. This study explores the use of convolutional neural networks (CNNs), including Alex Net, Squeeze Net, and ResNet18, to identify and categorize defects utilizing the PVEL-AD dataset. Binary classification was employed to distinguish between healthy and anomalous cells, while multiclass classification identified six specific defect types. The models were trained using the Adam optimizer with fine-tuned hyperparameters, achieving robust results. ResNet18 demonstrated the best performance, achieving a binary mAP@50 of 96.7% and a multiclass mAP@50 of 97.8 %. The results highlight the effectiveness of CNN-based approaches for automating PV panel defect detection, offering a scalable solution for real-time inspection and maintenance.