FURI | Spring 2025

Machine Vision for Identifying Defects in Perovskite Thin Films

This research applies machine vision and deep learning to automatically detect defects in perovskite solar cells using CNNs and K-means clustering. The model is trained on optical images of perovskite films with various types and quantities of defects, including pinholes and cracks. Robustness of the model is improved using data augmentation techniques to make the model generalize across different patterns, shapes, and sizes of defects. Automated processing simplifies the detection of defects and reduces reliance on manual inspection, enabling more consistency in quality control.