Deep Convolutional Neural Networks for Automated Defective Cell Detection and Classification in Photovoltaic Electroluminescence Images

Abstract

As photovoltaic systems are threatened by microstructural degradation, there is a need to shift from traditional manual inspection methods to automated diagnostic frameworks. This study explores the application of deep convolutional neural networks in the accurate identification and classification of multidimensional defects in electroluminescent images of silicon solar cells. An improved ResNet architecture with a dual attention module is deployed to enhance sensitivity in recognizing subtle intensity gradients. However, the algorithm's potential bias in low-contrast environments, influenced to some extent by the specific contrast characteristics of the dataset, suggests that further research is needed to ensure its universal applicability across different manufacturing standards.