Scientific- Research Quarterly of Geographical Data (SEPEHR)

Scientific- Research Quarterly of Geographical Data (SEPEHR)

Automated detection and classification of solar panel defects in UAV imagery using the YOLOv8m model

Document Type : Research Paper

Authors
Mohammad Amin Safari 1
Samira Mavaddati 2
1 MSc student , Faculty of engineering and technology ,University of Mazandarn,, Babolsar, Iran
2 Associate professor, Faculty of engineering and technology, University of Mazandaran, Babolsar, Iran
10.22131/sepehr.2025.2064007.3145
Abstract
Extended Abstract
Introduction:
With the rapid global shift toward renewable energy sources, photovoltaic (PV) systems have emerged as a cornerstone for clean and sustainable power generation. Ensuring the optimal performance and long-term durability of these systems requires effective inspection and maintenance strategies. Solar panels are frequently exposed to harsh environmental factors such as dust, moisture, temperature variations, and mechanical stress, which can lead to various physical defects including cracks, hotspots, delamination, and surface contamination. If left undetected, these defects may significantly compromise the energy output and reduce the economic lifespan of the solar panel infrastructure.
Traditional inspection methods, such as manual visual inspection or infrared thermography, are often labor-intensive, expensive, and prone to inaccuracies due to human error and environmental interference. These limitations underscore the need for an automated, accurate, and scalable solution. This research presents a smart defect detection framework based on deep learning, leveraging the cutting-edge YOLOv8m model for accurate and real-time identification of defects in solar panels. In addition, spatial analysis techniques are integrated to investigate the geographic distribution of defects, aiding in the development of targeted and predictive maintenance policies.
Materials & Methods:
The core of the proposed framework is the YOLOv8m (You Only Look Once version 8, medium variant) model, an advanced object detection architecture optimized for speed, accuracy, and efficiency. YOLOv8m is known for its compact design and superior performance, making it suitable for deployment in real-time monitoring systems.
A curated dataset was constructed for training and evaluation, consisting of both colored (RGB) and grayscale images of solar panels collected under various illumination conditions, angles, and environmental settings. The images were annotated to highlight six prevalent types of defects observed in real-world installations: cracks, hot spots, delamination, bird droppings, dirt patches, and broken glass.
To improve the model’s robustness and generalization capability, several data augmentation techniques were applied, such as random flipping, rotation, brightness variation, and Gaussian noise injection. Hyperparameter tuning was conducted to optimize learning rates, batch sizes, and anchor box dimensions. The model was trained using a transfer learning approach with pre-trained weights on the COCO dataset and fine-tuned on the specific solar panel defect dataset.
Furthermore, to extract spatial insights, the metadata embedded within the image files—such as GPS coordinates and timestamps—was utilized. This data was combined with the defect detection results to perform spatial clustering and distribution mapping using GIS-based tools and statistical analysis techniques.
Results & Discussion:
The proposed YOLOv8m model achieved impressive results across several performance metrics. The model recorded a mean average accuracy (mAP) of 97.43%, a precision score of 97%, and a recall rate of 97.58% in detecting and classifying the six identified defect types. These values were consistently higher than those of traditional deep learning models such as Convolutional Neural Networks (CNN), VGG16, and ResNet50, which served as baseline comparisons. Specifically, the YOLOv8m framework demonstrated a relative improvement of 1.5% to 3.5% in detection accuracy over the baseline models, highlighting its effectiveness in handling complex visual scenarios.
Qualitative analysis further supported these results, with the YOLOv8m model accurately localizing defects with tight bounding boxes and minimal false positives. The lightweight nature of the architecture enabled near real-time inference on standard GPU hardware, making it viable for deployment in field conditions.
The integration of spatial analysis added a novel dimension to the defect detection task. By correlating the defect locations with their geographic coordinates, patterns such as defect clustering in specific panel zones or environmental exposure zones were identified. These patterns suggested that external factors like shading from nearby objects, accumulation of debris, or exposure to extreme weather conditions could contribute to defect formation. Such insights can be used to develop location-specific maintenance protocols or preventive interventions, thereby improving the overall health and longevity of the solar panel infrastructure.
Conclusion:
This research introduces a comprehensive and scalable framework for intelligent detection of solar panel defects, built on the robust and efficient YOLOv8m architecture. The system not only offers high accuracy and real-time performance but also incorporates spatial intelligence for deeper understanding of defect trends and propagation. Its performance superiority over conventional models and its operational feasibility position it as a powerful tool for smart solar farm management.
By facilitating early diagnosis and enabling condition-based maintenance, the proposed framework contributes to reducing operational costs, extending the service life of solar panels, and enhancing the efficiency of photovoltaic systems. Moreover, the integration of machine vision and spatial analytics bridges the gap between technical diagnostics and actionable maintenance strategies, paving the way for smarter and more sustainable energy infrastructures.
Keywords
Defect detection
Solar panels
Deep learning
YOLOv8m
Machine vision
Spatial analysis
Smart monitoring
Subjects
 1- Abdelsattar, M., Abdelmoety, A., Ismeil, M. A., & Emad-Eldeen, A. (2025). Automated defect detection in solar cell images using deep learning algorithms. IEEE Access, 13, 4136–4157. https://doi.org/10.1109/ACCESS.2024.3525183
2- Aktouf, L., Shivanna, Y., & Dhimish, M. (2024). High-precision defect detection in solar cells using YOLOv10 deep learning model. Solar, 4(4), 639–659. https://doi.org/10.3390/solar4040030
3- Ali, M. L., & Zhang, Z. (2024). The YOLO framework: A comprehensive review of evolution, applications, and benchmarks in object detection. Computers, 13(12), 336. https://doi.org/10.3390/computers13120336
4- Balcıoğlu, Y. S., Sezen, B., & Cubukcu Cerasi, C. (2023). Solar cell busbars surface defect detection based on deep convolutional neural network. IEEE Latin America Transactions, 21(2), 242–250. https://doi.org/10.1109/TLA.2023.10015216
5- Basheer, M. Y. I., Ali, A. M., Hamid, N. H. A., & Yusoff, M. Z. M. (2024). Solar panel defect detection and panel localization using Yolov5. International Conference on Innovation & Entrepreneurship in Computing, Engineering & Science Education (InvENT), 139–149. https://doi.org/10.2991/978-94-6463-589-8_15
6- Dakamayi, Y., Sadeghzadeh, S. M., & Dakamayi, M. (2021). Intelligent fault detection and diagnosis in solar systems using machine learning methods. 2nd International Conference on Modern Research in Electrical Engineering, Computer, Mechanics and Mechatronics in Iran and the Islamic World, Karaj, Iran. https://civilica.com/doc/1245859
7- Demirci, M. Y., Beşli, N., & Gümüşçü, A. (2024). An improved hybrid solar cell defect detection approach using generative adversarial networks and weighted classification. Expert Systems with Applications, 252(A), 124230. https://doi.org/10.1016/j.eswa.2024.124230
8- Fasihi, A., & Dakiya, A. (2024). Investigation of faults and defects in photovoltaic systems and solar panels in electricity generation. 10th International Conference on Electrical, Computer and Mechanical Engineering, Tehran, Iran. https://civilica.com/doc/2120545
9- Go, S.-E., Kim, J.-H., Chuluunsaikhan, T., Choi, W.-S., Choi, S.-H., & Nasridinov, A. (2024). Unified generative data augmentation for efficient solar panel soiling localization. Electronics, 13(24), 4859. https://doi.org/10.3390/electronics13244859
10- Hijjawi, U., Lakshminarayana, S., Xu, T., Malfense Fierro, G. P., & Rahman, M. (2023). A review of automated solar photovoltaic defect detection systems: Approaches, challenges, and future orientations. Solar Energy, 266, 112186. https://doi.org/10.1016/j.solener.2023.112186
11- Huang, J., Zeng, K., Zhang, Z., & Zhong, W. (2023). Solar panel defect detection design based on YOLO v5 algorithm. Heliyon, 9(8), e18826. https://doi.org/10.1016/j.heliyon.2023.e18826
12- Karaca Aydemir, B. K., Telatar, Z., Güney, S., & et al. (2025). Detecting and classifying breast masses via YOLO-based deep learning. Neural Computing and Applications, 37, 11555–11582. https://doi.org/10.1007/s00521-025-11153-1
13- Lakshmi, P. S., Sivagamasundari, S., & Rayudu, M. S. (2024). Solar panel fault detection using low complex convolution neural network deep learning model. International Journal of Engineering Trends and Technology, 72(8), 18–26. https://doi.org/10.14445/22315381/IJETT-V72I8P103
14- Li, J., Li, H., Wu, Y., Zhou, H., Manfredi, L., Li, P., & Zhang, H. (2023). Deep learning based defect detection algorithm for solar panels. 2023 WRC Symposium on Advanced Robotics and Automation (WRC SARA), 438–443. https://doi.org/10.1109/WRCSARA60131.2023.10261859
15- Liang, Z., Xu, M., Su, Y., Chen, H., & Jin, G. (2024). A survey of solar panel surface defect detection methods based on improved VGG-16 model. 2024 IEEE 4th International Conference on Electronic Technology, Communication and Information (ICETCI), 820–827. https://doi.org/10.1109/ICETCI61221.2024.10594221
16- Liu, Y., Wu, Y., Yuan, Y., & Zhao, L. (2024). Deep learning-based method for defect detection in electroluminescent images of polycrystalline silicon solar cells. Optics Express, 32, 17295–17317.
17- Lu, S., Wu, K., & Chen, J. (2023). Solar cell surface defect detection based on optimized YOLOv5. IEEE Access, 11, 71026–71036. https://doi.org/10.1109/ACCESS.2023.3294344
18- Mahmud, A., Shishir, M. S. R., Hasan, R., & Rahman, M. (2023). A comprehensive study for solar panel fault detection using VGG16 and VGG19 convolutional neural networks. 2023 26th International Conference on Computer and Information Technology (ICCIT), 1–6. https://doi.org/10.1109/ICCIT60459.2023.10441429
19- Mavaddati, S., & Razavi, M. (2025). An optimized YOLO-ViT hybrid model for enhanced precision in rice classification and quality assessment. International Journal of Engineering, 38(10), 2435–2450. https://doi.org/10.5829/ije.2025.38.10a.19
20- Pandey, H., & Dhavale, S. (2024). A survey on artificial intelligence based photovoltaic cell defect detection techniques. 2024 IEEE Pune Section International Conference (PuneCon), 1–6. https://doi.org/10.1109/PuneCon63413.2024.10895155
21- PythonAfroz. (n.d.). Solar panel images dataset. Kaggle. https://www.kaggle.com/datasets/pythonafroz/solar-panel-images/data
22- Rainio, O., Teuho, J., & Klén, R. (2024). Evaluation metrics and statistical tests for machine learning. Scientific Reports, 14, 6086. https://doi.org/10.1038/s41598-024-56706-x
23- Ranjbarzadeh, R., Crane, M., & Bendechache, M. (2025). The impact of backbone selection in YOLOv8 models on brain tumor localization. Iranian Journal of Computer Science. https://doi.org/10.1007/s42044-025-00198-z
24- Shafiei, A., Kameli, V., & Grailu, H. (2023). Improved defect detection in photovoltaic panels through deep learning and decision tree-based classifiers. SSRN. https://ssrn.com/abstract=4519346
25- Singh, R., Kashyap, R., & Kumar, A. (2024). Prominent solution for solar panel defect detection using AI-based computer vision technology with IoT sensors in the solar panel manufacturing industry. International Journal of Information Technology. https://doi.org/10.1007/s41870-024-02212-2
26- Sirisha, U., Praveen, S. P., Srinivasu, P. N., et al. (2023). Statistical analysis of design aspects of various YOLO-based deep learning models for object detection. International Journal of Computational Intelligence Systems, 16, 126. https://doi.org/10.1007/s44196-023-00302-w
27- Song, X., & Liu, Y. (2023). Defect detection of solar panel based on DenseNet network. Advances in Transdisciplinary Engineering, 35, 661–669. https://doi.org/10.3233/ATDE230093
28- Terven, J., Cordova-Esparza, D. M., Romero-González, J. A., et al. (2025). A comprehensive survey of loss functions and metrics in deep learning. Artificial Intelligence Review, 58, 195. https://doi.org/10.1007/s10462-025-11198-7
29- Yang, C., Sun, F., Zou, Y., Lv, Z., Xue, L., Jiang, C., Liu, S., Zhao, B., & Cui, H. (2024). A survey of photovoltaic panel overlay and fault detection methods. Energies, 17(4), 837. https://doi.org/10.3390/en17040837
30- Zyout, I., & Oatawneh, A. (2020). Detection of PV solar panel surface defects using transfer learning of the deep convolutional neural networks. 2020 Advances in Science and Engineering Technology International Conference (ASET), 1–4. https://doi.org/10.1109/ASET48392.2020.9118382