Article Details

Abstract

In the context of clean energy transition, photovoltaic cell defect detection technology is crucial for ensuring component quality, improving power generation efficiency, and preventing safety risks. Existing detection methods based on electroluminescence images face bottlenecks such as strong image noise, low contrast, and subtle intra-class feature differences. To address these issues, this paper proposes a defect detection model integrating graph attention reasoning and gaze perception mechanisms. For spatial feature processing, inspired by biological vision, a translation-invariant pixel-focused attention module is designed, which simulates eye movement characteristics through a multi-granularity fusion strategy combining fine-grained local perception and coarse-grained global understanding. For channel feature modeling, drawing on the human brain's reasoning mechanisms, a channel reasoning module based on graph attention network is constructed to capture cross-channel long-range dependencies. Through spatial-channel dual-dimensional feature induction and coupling, the robustness and generalization ability of the model are significantly enhanced. Experiments show that the proposed method improves precision and recall by 3.5% and 2.7% respectively, which verifies the effectiveness and advancement of the framework.

Keywords

Deep Learning; Graph Neural Network; photovoltaic cell defect detection

References

1. [1] KABIR E，KUMAR P，KUMAR S，et al. Solar Energy: Potential and Future Prospects [J].Renewable and Sustainable Energy Reviews，2018，82：894-900.
2. [2] MAHMOUD A，SINGH R，OMAR M. Review of Microcrack Detection Techniques for Silicon Solar Cells [J].IEEE Journal of Photovoltaics，2014，4（1）：514-524.
3. [3] ZHAO S S，CHEN H Y，WANG C H，et al. SNCF-Net: Scale-Aware Neighborhood Correlation Feature Network for Hotspot Defect Detection of Photovoltaic Farms [J/OL].Measurement，2023，206：112342（2022-12-16）.https://doi.org/10.1016/j.measurement.2022.112342.
4. [4] QU Z Y，LI L C，ZANG J Y，et al. A Photovoltaic Cell Defect Detection Model Capable of Topological Knowledge Extraction [J/OL].Scientific Reports，2024，14（1）：21904 （2024-09-19）.https://doi.org/10.1038/s41598-024-72717-0.
5. [5] LAITH A，ZHANG J L，HUMAIDI A J，et al. Review of Deep Learning: Concepts, CNN Architectures, Challenges, Applications, Future Directions [J/OL].Journal of Big Data，2021，8（1）：53（2021-03-31）.https://doi.org/10.1186/s40537-021-00444-8.
6. [6] HAN K，XIAO A，WU E H，et al. Transformer in Transformer [J].Advances in Neural Information Processing Systems.NIPS，2021：15908-15919.
7. [7] He Xiang, Yang Aijun, Li Jiansheng, etc A data augmentation method for solar cell electroluminescence images based on cycleGAN [J]. Liquid Crystal and Display, 2024, 39 (8): 1057-1069
8. [8] Wang Fangbin, Li Wenhao Photovoltaic cell EL image defect detection based on improved lightweight U-Net model [J]. Electronic Measurement Technology, 2024, 47 (5): 102-111
9. [9] Sun Tieqiang, Zhao Chengwei, Song Chao, etc Multi scale lightweight photovoltaic cell EL defect detection [J]. Modular Machine Tool and Automation Processing Technology, 2025 (3): 177-181186
10. [10] Liu Yu, Zhu Wenzhong, He Xin, etc Improving the EL defect detection of lightweight photovoltaic cells using YOLOv8n [J]. Modular Machine Tool and Automation Processing Technology, 2025 (3): 87-93
11. [11] SHI D. Transnext: Robust Foveal Visual Perception for Vision Transformers [C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.Seattle：IEEE，2024：17773-17783.
12. [12] LIU Q，LIU M，WANG C，et al. An efficient CNN-based Detector for Photovoltaic Module Cells Defect Detection in Electroluminescence Images [J/OL].Solar Energy，2024，267：112245（2023-12-07）.https://doi.org/10.1016/j.solener.2023.112245.

About the authors

Wang Shaokang (1997-), male, Han ethnicity, from Fugou County, Henan Province, China, teaching assistant, master's degree, research direction: computer technology.