Solar photovoltaic panel cells defects classification using deep learning ensemble methods
Hambal Tella, A. Hussein, Shafiqur Rehman, Bo Liu, Adil Balghonaim, Mohamed Mohandes
Abstract
Solar photovoltaic (PV) systems are essential for sustainable energy production; however, their reliability may be undermined by unfavorable weather conditions, resulting in defects in the individual cells. Conventional manual inspection techniques are labor-intensive and susceptible to human error. This study utilizes drone-acquired electroluminescence (EL) images to identify and categorize solar cell defects through an ensemble-based deep learning framework. Eight advanced models—AlexNet, SENet, GoogleNet (Inception V1), Xception, Vision Transformer (ViT), Darknet53, ResNet18, and SqueezeNet—were fine-tuned on the 2,624-sample ELPV benchmark dataset. Experimental findings indicate that the proposed voting and bagging ensembles attain accuracies of 68.36% and 68.31%, respectively, exceeding the previously documented accuracy of a hybrid model at 61.15%. Significantly, the ResNet18 model achieves an accuracy of 73.02% in a straightforward binary classification task, highlighting that individual models can surpass ensembles in particular circumstances. This study emphasizes the efficacy of integrating various deep learning architectures to augment defect detection precision in photovoltaic systems, enhancing operational reliability and enabling prompt maintenance under challenging environmental conditions.