Hanim Suraya Mohd Mokhtar, Aimi Salihah Abdul Nasir, Mohammad Faridun Naim Tajuddin, Muhammad Hafeez Abdul Nasir, Kumuthawathe Ananda Rao,
Volume 21, Issue 2 (6-2025)
Abstract
The rapid growth of photovoltaic (PV) systems has highlighted the need for efficient and reliable defect detection to maintain system performance. Electroluminescence (EL) imaging has emerged as a promising technique for identifying defects in PV cells; however, challenges remain in accurately classifying defects due to the variability in image quality and the complex nature of the defects. Existing studies often focus on single image enhancement techniques or fail to comprehensively compare the performance of various image enhancement methods across different deep learning (DL) models. This research addresses these gaps by proposing an in-depth analysis of the impact of multiple image enhancement techniques on defect detection performance, using various deep learning models of low, medium, and high complexity. The results demonstrate that mid-complexity models, especially DarkNet-53, achieve the highest performance with an accuracy of 94.55% after MSR2 enhancement. DarkNet-53 consistently outperformed both lower-complexity models and higher-complexity models in terms of accuracy, precision, and F1-score. The findings highlight that medium-depth models, enhanced with MSR2, offer the most reliable results for photovoltaic defect detection, demonstrating a significant improvement over other models in terms of accuracy and efficiency. This research provides valuable insights for optimizing defect detection systems in photovoltaic applications, emphasizing the importance of both model complexity and image enhancement techniques for robust performance.
