Robust denoising methodology for outdoor electroluminescence images of photovoltaic modules using deep learning

This study presents a novel deep learning-based approach for denoising photovoltaic (PV) module luminescence images, addressing the critical need to enhance image quality in outdoor inspection imaging. A simplified ResNet-based architecture, termed SimpleResNet, was developed and evaluated against conventional denoising techniques. The proposed method demonstrates superior performance in both quantitative metrics and qualitative visual assessments, particularly in preserving the fine details and structural integrity of PV module images. It is significantly faster than conventional techniques and, therefore, highly suitable for high-throughput applications. A comprehensive preprocessing pipeline incorporating perspective distortion correction, denoising, and sharpness enhancement is implemented to address real-world outdoor imaging challenges. The pipeline's effectiveness has been validated using actual outdoor electroluminescence images, exhibiting consistent performance across diverse module types and lighting conditions. These developed capabilities contribute to enhancing the accuracy and efficiency of automated inspection systems for utility-scale PV plants. • This study presents a novel deep learning-based approach for denoising photovoltaic (PV) module luminescence images. • This work helps to address the critical need to enhance image quality in outdoor inspection imaging. • The proposed method demonstrates superior performance against conventional denoising techniques. • It is also significantly faster than conventional techniques, so is highly suitable for high-throughput applications. • This study contributes to enhancing the accuracy and efficiency of automated inspection systems for utility-scale PV plants.