Variant Histogram Equalization-Based Enhancement to Deep Transfer Learning for Automatic Detection of Photovoltaic Cell Defects in Electroluminescence Images

Increasing the use of renewable energy sources, specifically solar photovoltaic (PV) modules, is essential for meeting the world's energy needs. However, the prevalence of defects, particularly microcracks, in PV modules significantly hampers their efficiency, lifespan, and safety. Detection of these defects traditionally relies on time-consuming and expertise-intensive electroluminescence (EL) imaging. This study introduces an innovative approach utilizing EL images and deep learning techniques for automated PV module defect detection. The primary objective is the development of an automated visual defect inspection system employing EL images and deep learning models. EL image data, sourced from a public dataset, were augmented through techniques like rotation, enhancing dataset diversity and robustness. Pre-trained convolutional neural network (CNN) models based on Darknet53 were deployed for defect classification. Various enhancement techniques, including DOTHE, ETHE, ESIHE, and MMSICHE were employed to evaluate model performance in defect detection. Comparative analysis of Darknet53 models enhanced with various techniques revealed an outstanding overall accuracy of 88.50% when employing ETHE. This system effectively addresses the need for reliable and efficient PV cell defect detection methods. Furthermore, it holds the potential to enhance PV module efficiency and reliability, thus augmenting the performance and longevity of solar energy facilities. Hence, the automated visual defect inspection system demonstrated in this paper contributes to the development of methodologies for defect detection in PV modules.