Research on Surface Defect Detection of Photovoltaic Panels Based on an Improved YOLOv8n

Authors

  • Rui Wang Xihua University, Chengdu, Sichuan 610039, China
  • Heng Xiao Xihua University, Chengdu, Sichuan 610039, China
  • Jiaqi Tang Xihua University, Chengdu, Sichuan 610039, China
  • Huang Wang Xihua University, Chengdu, Sichuan 610039, China
  • Jixue Xiao Xihua University, Chengdu, Sichuan 610039, China

DOI:

https://doi.org/10.6911/WSRJ.202611_12(11).0007

Keywords:

Boundary box regression, defect detection, electroluminescence imaging, large separable kernel attention, photovoltaic panels, YOLOv8n.

Abstract

Photovoltaic panels exposed to outdoor environments for long periods may develop defects such as cracks, micro-cracks, and short-circuited cells, which degrade their electrical performance and operational reliability. Electroluminescence (EL) imaging reveals internal structural differences of solar cells and provides an effective basis for visual defect inspection. However, micro-cracks in EL images are typically thin and of low local contrast, and the scale difference among defect categories is significant, making accurate detection and boundary localization difficult. To address these problems, this paper proposes an improved detection method based on YOLOv8n. In the baseline network, the C2f module is replaced with C2fHGBlockExp, the InterpIoU loss is adopted for bounding box regression, and a large separable kernel attention (LSKA) module is introduced after the multi-scale pooling features of SPPF to construct an LSKA-SPPF feature extraction module. Experiments are conducted on the public PV_Train_Val_28_12 dataset, which contains 2,781 images divided into 1,947 for training, 556 for validation, and 278 for testing, and all models are trained for 500 epochs. In the ablation study, the model combining the three improvements achieves Precision, Recall, mAP@0.5, and mAP@0.5:0.95 of 93.289%, 88.461%, 93.180%, and 69.438%, respectively, which are 9.890, 8.823, 7.889, and 11.653 percentage points higher than the baseline. The results show that the proposed combination of improvements enhances detection accuracy and localization performance under the given dataset and experimental settings.

Downloads

Download data is not yet available.

References

[1] Yaermaimaiti, Y., Deng, H., & Xie, L. (2023). Solar panel defect detection based on improved YOLOv4. Journal of South China Normal University (Natural Science Edition), 55(5), 21–30. (in Chinese)

[2] Zheng, Z., & Ma, P. (2025). EL image defect detection of photovoltaic modules based on improved YOLOv5. Acta Energiae Solaris Sinica, 46(10), 295–301. (in Chinese)

[3] Tang, Z., & Tang, D. (2025). Solar panel defect detection based on improved YOLOv8. Journal of Hunan University of Science and Technology (Natural Science Edition), (4). (in Chinese)

[4] Wang, T., & Li, Y. (2026). Photovoltaic cell defect detection algorithm based on improved YOLOv8n. Journal of Chongqing Technology and Business University (Natural Science Edition), 43(4), 42–51. (in Chinese)

[5] Chen, T., & Chen, Z. (2026). EL image based photovoltaic cell defect detection model for edge devices. Journal of Fuzhou University (Natural Science Edition), 54(1), 1–9. (in Chinese)

[6] Liu, X., & Ge, M. (2025). Fault diagnosis and health assessment of photovoltaic modules based on image processing technology. Solar Energy, (12), 51–58. (in Chinese)

[7] Xu, Y., & Li, J. (2024). A survey of YOLO series object detection algorithms. Computer Science and Exploration, 18(9), 2221–2238. (in Chinese)

[8] Shao, Y., & Zhang, D. (2022). A survey of deep learning based YOLO object detection. Journal of Electronics & Information Technology, 44(10), 3697–3708. (in Chinese)

[9] Wang, X., Chang, J., & Yu, Q. (2025). Research on small object detection methods based on YOLOv8. Journal of North University of China (Natural Science Edition), 46(6), 712–725. (in Chinese)

[10] Liu, J., & Shi, R. (2025). An improved YOLOv8-based small target detection model for UAV vision. Journal of Northeastern University (Natural Science), 46(12), 29–37. (in Chinese)

[11] Cao, J., & Qiao, G. (2025). YOLO algorithm improvement strategies for small object detection from high-altitude perspectives. Journal of Computer Applications. (in Chinese)

[12] Cheng, H., & Jiao, L. (2024). Improved YOLOv8 for remote sensing image detection. Radio Engineering, 54(5), 1155–1161. (in Chinese)

[13] Cheng, H., & Qiao, Q. (2024). Object detection in UAV aerial images based on improved YOLOv8. Radio Engineering, 54(4), 871–881. (in Chinese)

[14] Dong, H., & Pan, J. (2024). Research on bounding box regression loss functions based on the YOLOv5s model. Modern Electronic Technology, 47(3), 179–186. (in Chinese)

[15] Huo, J., & Su, H. (2025). Small target vehicle detection in road traffic based on improved YOLOv8. Computer Engineering, 51(1), 246–257. (in Chinese)

[16] Faculty of Engineering Fayoum University. (2022). PV_Train_Val_28_12 Dataset. Roboflow Universe. https://universe.roboflow.com/faculty-of-engineering-fayoum-university/pv_train_val_28_12

Downloads

Published

2026-11-17

Issue

Section

Articles

How to Cite

Wang, R., Xiao, H., Tang, J., Wang, H., & Xiao, J. (2026). Research on Surface Defect Detection of Photovoltaic Panels Based on an Improved YOLOv8n. World Scientific Research Journal, 12(11), 52-65. https://doi.org/10.6911/WSRJ.202611_12(11).0007