Drone-Based Aerial Inspection of Solar Panel Defects using an Improved YOLO Model

DOI: https://doi.org/10.62517/jike.202604136

Author(s)

Tao Chen, Mengmei Wang*, Tianxiang Hou, Yang Cao

Affiliation(s)

Jiangsu Normal University KeWen College, Xuzhou, Jiangsu, China *Corresponding Author

Abstract

As the scale of photovoltaic power plants continues to expand, traditional manual inspection methods are inefficient, costly, and pose significant safety risks, making them inadequate for actual operation and maintenance needs. The integration of drone aerial photography and computer vision technology provides an efficient and safe solution for detecting defects in photovoltaic panels. To address the problem of insufficient detection accuracy caused by the small size of defect targets, complex backgrounds, and unclear features in aerial images, this paper proposes a defect detection model based on an improved YOLOv8. The model incorporates a coordinate attention mechanism into the backbone network to enhance the localisation of tiny defects; it replaces the original structure with a weighted bidirectional feature pyramid network to improve multi-scale feature fusion; and it uses a SIoU loss function to optimise bounding box regression accuracy. Experimental results show that the improved model achieves an average precision of 92.7% on a self-built dataset, an increase of 4.3 percentage points over the baseline model, and its detection speed meets real-time requirements, providing effective technical support for intelligent operation and maintenance of photovoltaic power plants.

Keywords

UAV Aerial Photography; PV Panel Defect Detection; YOLOv8; Attention Mechanism; Feature Fusion

References

[1] Cui Jianwei, Wang Yueming. A Defect Detection and Localization System for Photovoltaic Panels Based on YOLOv11. Electronic Measurement Technology, 1–9 [2026-03-23]. [2] Ji Ruirui, Mei Yuan, Yang Sifan, et al. An Infrared Hotspot Detection Algorithm for Photovoltaic Modules Based on Improved Faster R-CNN. Laser & Infrared, 2024, 54(04): 584–592. [3] Sun Qiyao, Sun Junhai, Zhu Xianyuan, et al. A Method for Defect Detection in Infrared Photovoltaic Modules Based on an Improved YOLOv8. Journal of Solar Energy, 2026, 47(02): 148–154. DOI: 10.19912/j.0254-0096.tynxb.2024-1871. [4] Li Zhe, Yang Jie, Zhang Yi, et al. Precise Identification of Aircraft Arresting Gear Towing Status Based on Coordinate Attention and Weighted Bidirectional Feature Pyramid Network . Chinese Ship Research, 2025, 20(04): 124-133. DOI:10.19693/j.issn.1673-3185.04005. [5] Xiang Linying, Shao Shao, Ma Lepeng, et al. An Image Restoration Algorithm Based on SE Modules and Hollow Convolution. Journal of Command and Control, 2025, 11(05): 649–654. [6] Wu Yifan, Ma Lingkun, Yan Meiyi. Research on a Novel Variable Step Size Algorithm Based on the Sigmoid Function. Applied Electronics, 2026, 52(02): 62–65. DOI:10.16157/j.issn.0258-7998.256591. [7] Sang Zhi-lei, Qi Long, Zhou Xun-rong. Research on a PANet-Based Defect Detection Model for Toy Parts. Toy World, 2025, (06): 7–9. [8] Zhang, Jing. A Study on Document Image Binarization Methods Based on Metric Loss and Attention. Taiyuan University of Technology, 2024. DOI: 10.27352/d.cnki.gylgu.2024.002711. [9] Liu Tianbo. Design and Implementation of a Blockchain-Based Precision Poverty Alleviation System under the C2F Model. South-Central University for Nationalities, 2021. [10] Gu Ming, Zhang Zhenwei, Chen Mingming, et al. Defect Detection in Photovoltaic Modules Based on an Improved YOLOv5-RCD. Journal of Xi'an University of Engineering, 1–9 [2026-03-23].