Real-Time Solar Cell Defect Detection Using Optimized YOLOv5 with Attention and Multi-Scale Augmentation

Abstract

This paper presents an optimized YOLOv5
based model for accurate detection of solar cell surface 
defects. The model incorporates advanced data 
augmentation techniques and a Channel Attention 
(CA) mechanism to enhance feature extraction and 
robustness. Additionally, a decoupled detection head 
is introduced to improve classification and localization 
performance. Experimental results demonstrate 
significant improvement in detection accuracy and 
real-time performance compared to the standard 
YOLOv5 model.