Enhancing defect classification in solar panel with Electroluminescence imaging and advanced machine learning strategies

Abstract

Electroluminescence (EL) imaging provides an effective non-destructive approach for identifying internal defects in photovoltaic (PV) cells and solar panels that may not be visible through conventional inspection techniques. However, accurate classification of EL images remains challenging due to complex defect structures, variations in defect severity, low-intensity patterns, and similarities between different defect categories. This paper proposes an advanced deep learning framework for automated solar panel defect classification using EL images. The proposed approach investigates convolutional neural network (CNN) architectures integrated with deep residual and densely connected featurelearning mechanisms. ResNet is employed to extract hierarchical and discriminative visual representations through residual learning, while DenseNet enhances feature reuse and information propagation across convolutional layers. The extracted representations are subsequently processed through fully connected classification layers to categorize photovoltaic cells into different defect conditions. The framework is designed to distinguish defective and non-defective cells and further identify specific defect patterns such as cracks, inactive regions, and other cell-level abnormalities. A comprehensive evaluation is performed using accuracy, precision, recall, F1-score, and confusion-matrix analysis to assess classification effectiveness across multiple defect categories. The proposed framework focuses on improving feature discrimination and classification reliability while maintaining an efficient deep learning architecture suitable for automated photovoltaic inspection. Experimental analysis demonstrates the potential of combining residual and dense feature representations for robust EL-based solar panel defect classification. The proposed methodology can support automated quality assessment, reduce dependence on manual inspection, and improve the reliability of photovoltaic manufacturing and maintenance processes.