| Title |
Solar Cell Defect Detection System through Image Augmentation with Filtering |
| Authors |
김태경(Tae-kyeong Kim) ; 조현종(Hyun-chong Cho) |
| DOI |
https://doi.org/10.5370/KIEE.2023.72.3.428 |
| Keywords |
Filtering on Augment Image; Image Augmentation; Solar cell; Vision Transformer |
| Abstract |
The need for renewable energy is increasing due to problems such as environmental pollution and resource depletion. Solar energy is the largest type of renewable energy and is increasing. However, when a micro-crack is formed in a solar cell, it greatly affects the photovoltaic power generation system. This degrades the performance of the solar power system and can be easily damaged. In this paper, solar cell defects were detected through Electroluminescent (EL) images. The Vision Transformer-Huge model was used to extract solar cell defect features. In addition, the CIFAR-10 augmentation policy and Imagenet augmentation policy included in Auto-augment were used to increase the performance of the model for the small dataset. When using the augmentation policy, the characteristics of the solar cell defect may disappear from the data, so filtering was performed with the model learned from the original data. For filtering, only images with a prediction probability of 70% or more of augmented images through the Soft-max function were used. When comparing the performance of the proposed methods and the method learned with the original data, it showed excellent performance. As a result, the proposed model showed a performance improvement of about 5% based on accuracy, and a recall value of about 93% was obtained. |