Dual CNN for photovoltaic electroluminescence images microcrack detection

Abstract

Accurate detection of microcracks in photovoltaic (PV) cells is crucial for ensuring the efficiency and longevity of solar panels. This study proposes a dual convolutional neural network (Dual-CNN) architecture to enhance microcrack detection in electroluminescence (EL) PV images. By integrating shallow feature extraction with deep semantic analysis, the proposed model effectively captures both fine-grained local textures and high-level structural patterns, addressing the limitations of conventional single-stream CNN models that primarily focus on coarse-grained features. Experimental evaluations on an EL image dataset demonstrate that the Dual-CNN approach significantly improves defect localization and classification with an accuracy of 85.33%, a recall of 71.71% and an F1 score of 73.9%, paving the way for more robust and automated PV inspection systems in the solar energy sector.