SG-YOLOv8: an improved YOLOv8-based photovoltaic panel defect detection algorithm

Abstract Efficient defect detection in photovoltaic (PV) panels is essential for optimizing the efficiency and reliability of solar power systems, crucial for advancing sustainable energy solutions. Traditional defect detection methods, relying on physical inspections or electrical analyses, are not only labor-intensive but also become economically inefficient when applied on a large scale. Addressing these challenges, this study introduces SG-YOLOv8, an enhanced version of the YOLOv8 algorithm tailored for automated defect detection in PV panels through sophisticated infrared image analysis. This novel algorithm enhances detection capabilities by integrating spatial attention mechanisms and depthwise separable convolutions, which significantly optimize the feature extraction process for improved accuracy. Furthermore, SG-YOLOv8 employs an advanced soft-NMS technique, innovatively designed to prevent the loss of detection accuracy caused by improper threshold settings in traditional NMS algorithms. Experimental results confirm that SG-YOLOv8 surpasses traditional methods, showing superior accuracy and processing speed, especially in detecting minute defects and operating effectively against complex backgrounds. These findings not only demonstrate the potential of advanced machine learning techniques in the field of PV defect detection but also underscore the algorithm’s effectiveness and scalability as a crucial tool for enhancing solar power system maintenance.