Improved YOLOv10 model for detecting surface defects on solar photovoltaic panels

Abstract

Surface defects greatly affect the performance and service life of photovoltaic (PV) modules. Detecting these defects is important to improve the management, repair and maintenance of PV panels. With the development of artificial intelligence, computer vision brings higher accuracy and lower labor costs than traditional inspection methods. This paper introduces an improved PV you only look once v10 (YOLOv10) model for detecting surface defects of PV modules. The improvement includes adding an exponential moving average (EMA) attention mechanism to the neck, using a cycle generative adversarial network (GAN) to enhance the data, and replacing the YOLOv10 head with a YOLOv9 head to retain non-maximum suppression (NMS). Experiments show that the proposed model outperforms state-of-the-art methods such as YOLOv10s, n, x, b, l, and e, achieving superior detection accuracy. Despite the increased computational cost, the proposed method improved mAP@0.5 and mAP@0.5:0.95 by 5.1% and 6.5% over the original YOLOv10s.