Different from traditional defect detection methods, our method is a domain generalization model.
Electroluminescence (EL) Endogenous Shift (ELES) dataset is the first endogenous domain shift benchmark of photovoltaic manufacturing scenarios. Due to the differences in intrinsic mechanisms or hardware components of the imaging systems of different types of imaging conditions, there are subtle perturbations in EL images generated by different quality inspection production line, which cause the endogenous shift.
We collected three sets of EL data (EL group 1, EL group 2, and EL group 3) at different times during the production line update process of the same manufacturer. In this study, EL group 1 is used for training the detector, and the other two are used for the test. To efficiently test PV modules, we slice a PV module based on the unit of 2 cells and spliced it into a complete module after the detection. Combined with deep learning technology and with the assistance of multiple experts, we meticulously annotate all defects. The dataset contains a total of
Number of defects | EL group 1 | EL group 2 | EL group 3 | Total |
---|---|---|---|---|
broken gate | 2436 | 909 | 335 | 1620 |
unjoined weld | 1563 | 1167 | 275 | 1669 |
black spot | 2313 | 833 | 270 | 1440 |
crack | 1951 | 187 | 62 | 508 |
scratch | 2219 | 1011 | 257 | 1567 |
Datasets | EL group 1 | EL group 2 | EL group 3 |
---|---|---|---|
Number of defective image | 7762 | 3209 | 839 |
Number of defect-free image | 1410 | 2541 | 562 |
Total | 9172 | 5750 | 1401 |
Acquisition time | September 2020 | May 2021 | October 2021 |
Resolution | 384 x 384 | 640 x 589 | 700 x 668 |
If you want to access the dataset, please read the ELES-Dataset Request Form in this folder carefully, sign the corresponding commitment file, and send it to [email protected].
EDS dataset consists of:
-
3 different domains
-
Altogether 16,323 EL images
If you are using the code/model/data provided here in a publication, please consider citing:
@ARTICLE{3327572,
author={Zhao, Shenshen, and Chen, Haiyong, and Wang, Chuhan, and Zhong, Zhang},
journal={IEEE Transactions on Industrial Informatics},
title={SSN: Shift Suppression Network for Endogenous Shift of Photovoltaic Defect Detection},
year={2023},
volume={},
number={},
pages={1-13},
doi={10.1109/TII.2023.3327572}
}