A Deep Learning Approach to Denoise Electroluminescence Images of Solar Cells

Grace Liu, Priya Dwivedi, Thorsten Trupke, Ziv Hameiri University of New South Wales, Sydney, Australia

Luminescence imaging is essential for solar cell performance and reliability analysis. It is used to identify spatial defects and extract key electrical parameters. To reliably identify defects, high quality images are desirable; however, acquiring such images implies a higher cost and lower throughput as they require better imaging systems and longer exposure times. Reducing the exposure time or using cheaper cameras increases the amount of noise. Therefore, this study proposes a deep learning model to significantly reduce the noise in electroluminescence images, hence, improving the quality of their analysis without the need for additional hardware expenses. The proposed deep learning approach improved noisy images by 30.4% and 39.3% in terms of the peak signal-to-noise ratio and structural similarity index, respectively.