 |
International Society of Science and Applied Technologies |
| Deep Learning-based Solar Panel Defect Detection and Localization Using Electroluminescent Images | ||||
| Author | Di Zhou
|
|||
| Co-Author(s) | Chuqi Yang; Xuan Zhang; Jianmin Sun; Xiao Zhuang
|
|||
| Abstract | Automatic defect detection in solar cell electroluminescence (EL) images is a challenging task due to the similarity between defect features and complex background features. To solve this problem, we propose an improved method YOLOv5n-DSC based on YOLOv5 neural network for solar panel surface defects detection. We introduce Dynamic Snake Convolution on top of YOLOv5, which automatically changes the shape of convolution according to the shape of defects. When targeting long defects, the Dynamic Snake Convolution can provide a significant improvement in feature extraction compared to the normal convolution. Finally, some experimental results on a large-scale EL dataset (including 4500 images) show that the proposed method performs better than the original one in terms of defect classification and detection results of the original solar cell EL images.
|
|||
| Keywords | Deep Learning; Solar Panel; Defect Detection | |||
| Article #: RQD2024-303 | ||||
Proceedings of 29th ISSAT International Conference on Reliability & Quality in Design |