Panel Surface Segmentation in Thermal Photovoltaic Images Using YOLO11n-Seg and Its Assessment as a Preprocessing Step for Thermal Analysis

Abstract

Reliable separation of photovoltaic (PV) panel surfaces from their background is an early preprocessing step for temperature-based inspection of PV systems. This study addresses the automatic segmentation of PV panel surfaces in thermal solar-panel images using YOLO11n-Seg, a lightweight instance segmentation model. The public dataset comprises 410 images acquired with a UTi165A thermal camera, split into 304 training, 70 validation and 36 test images; the annotations enclose entire panel surfaces under a single class (PV) rather than hotspots. Images were resized to 640x640, and the model was initialised from COCO-pretrained weights and trained for three epochs on CPU only. Evaluation was carried out on a held-out test split provided with the public dataset of 36 images (60 panel instances). For the mask metrics, which are the primary output of this work, the model reached an mAP@0.5 of 0.9729 and an mAP@0.5:0.95 of 0.8312, with mask precision and recall of 0.9199 and 0.9566, respectively; the box mAP@0.5:0.95 was 0.7519. Inference took about 283 ms per image on CPU (approximately 3.5 images per second). The results indicate that panel boundaries are generally delineated well across IoU thresholds and that the model captures most of the true panel regions, although a few low-confidence extra predictions were observed. The resulting panel masks can serve as a region-of-interest step for subsequent thermal-anomaly and hotspot analyses. Because the findings were obtained from a single three-epoch training run, a single class and a relatively small test set, and under CPU-based inference, they should be regarded as a preliminary assessment that requires further validation before generalisation to different field conditions. The model segments panel surfaces only; it does not classify or separately segment hotspots within a panel.

Keywords

• Thermal imaging
• Photovoltaic panel
• Segmentation
• YOLO11n-Seg
• Deep learning

Ethical Statement

Ethics committee approval was not required for this study because there was no study on animals or humans.

References

1. Abuhussein, M. F. A. (2024). Thermal imaging dataset for hotspot detection on solar panels: Impact of bird droppings on efficiency (Version 1) [Data set]. Mendeley Data. https://doi.org/10.17632/8sxfmrpfpv.1
2. Buerhop, C., Bommes, L., Schlipf, J., Pickel, T., Fladung, A., & Peters, I. M. (2022). Infrared imaging of photovoltaic modules: A review of the state of the art and future challenges facing gigawatt photovoltaic power stations. Progress in Energy, 4(4), Article 042010. https://doi.org/10.1088/2516-1083/ac890b
3. Elmessery, W. M., Habib, A., Shams, M. Y., Abd El-Hafeez, T., El-Messery, T. M., Elsayed, S., Fodah, A. E. M., Abdelwahab, T. A. M., Ali, K. A. M., Osman, Y. K. O. T., Abdelshafie, M. F., Abd El-wahhab, G. G., & Elwakeel, A. E. (2024). Deep regression analysis for enhanced thermal control in photovoltaic energy systems. Scientific Reports, 14, Article 30600. https://doi.org/10.1038/s41598-024-81101-x
4. He, K., Gkioxari, G., Dollár, P., & Girshick, R. (2017). Mask R-CNN. In Proceedings of the IEEE International Conference on Computer Vision (ICCV) (pp. 2961–2969). IEEE. https://doi.org/10.1109/ICCV.2017.322
5. Høiaas, I., Grujic, K., Imenes, A. G., Burud, I., Olsen, E., & Belbachir, N. (2022). Inspection and condition monitoring of large-scale photovoltaic power plants: A review of imaging technologies. Renewable and Sustainable Energy Reviews, 161, Article 112353. https://doi.org/10.1016/j.rser.2022.112353
6. Jocher, G., & Qiu, J. (2024). Ultralytics YOLO11 (Version 11.0.0) [Computer software]. Ultralytics. https://github.com/ultralytics/ultralytics
7. Khanam, R., & Hussain, M. (2024). YOLOv11: An overview of the key architectural enhancements [Preprint]. https://doi.org/10.48550/arXiv.2410.17725
8. Rahaman, S. A., Urmee, T., & Parlevliet, D. A. (2022). Investigate the impact of environmental and operating conditions of infrared (IR) imaging on infrared thermography of PV modules to identify defects. Solar Energy, 245, 231–253. https://doi.org/10.1016/j.solener.2022.09.003

9. Redmon, J., Divvala, S., Girshick, R., & Farhadi, A. (2016). You only look once: Unified, real-time object detection. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (pp. 779–788). IEEE. https://doi.org/10.1109/CVPR.2016.91