Crack Detection of Solar Cell Based on Ultraviolet Fluorescence

As the core part of the entire power generation system, the quality of solar cells determines the electrical energy conversion efficiency. Currently, electroluminescence images (EL images) are widely used in the industry to detect hidden cracks in modules. Since EL technology needs to apply a reverse voltage to the solar cell, it requires a lot of resources and manpower to collect data. At this time, the non-contact ultraviolet fluorescence detection (UVF) technology has more advantages in collecting images. When using UVF to collect images on site, it is only necessary to use ultraviolet light to illuminate the solar module in the dark. Since the detection of cracks using UVF images is mostly manual, and the algorithms currently applied to it cannot detect cracks effectively, there is a lack of exploration of algorithms for separating damaged areas. In this paper, a CNN-based object detection algorithm is proposed to detect cracks. The experiments are conducted using the YOLOv5 model as the base architecture and by adding an attention mechanism to it for crack detection. Our experimental results demonstrate that our scheme for crack detection outperforms other frameworks. This work is the first attempt to use attention mechanism to solve crack detection problem in UVF images without much manpower and resources.