Photovoltaic Module Array Luminescence Image Preprocessing: Heuristic Algorithms for Perspective Correction and Cell Segmentation

To address the significant and growing challenge of photovoltaic (PV) module end-of-life (EoL) management, characterisation techniques including infrared (IR) imaging, or more recently electroluminescence (EL) and even photoluminescence (PL) imaging are increasingly being employed to monitor the quality of installed PV modules. However, current methods for module quality assessment are often manual and time-consuming, particularly for luminescence-based imaging, performed by domain experts on individual images. These approaches cannot meet the needs of the rapidly growing PV installations, and therefore automation is required to unlock the significant potential of these imaging techniques. Existing image preprocessing methodologies typically fail to successfully handle images with the complexity of features commonly found in real-world images. This study outlines a novel methodology for robust PV module array image preprocessing, comprising a pipeline of heuristic algorithms, enabling subsequent cell-level analysis and quality assessment of individual modules. The developed preprocessing methodology has been successfully validated on multiple separate real-world image datasets, including those comprising EL, PL, and IR images of module arrays. Further, upon integration within a novel module quality assessment pipeline, we have shown the end-to-end capability to automatically identify and classify defective cells and modules at scale with high accuracy, solely from as-measured PL images of fielded PV modules. This helps to address the significant challenge of informed PV module EoL decisions for utility-scale systems.