研究目的
To improve the classification performance of defective photovoltaic module cells in electroluminescence images using GAN-based data augmentation.
研究成果
The proposed AC-PG GAN model effectively augments the EL images dataset, leading to improved classification accuracy of defective photovoltaic module cells. The maximum improvement in classification accuracy is up to 14%.
研究不足
The training of GAN networks is highly unstable and can be costly. The quality of generated samples may vary, and the method requires careful adjustment to ensure effectiveness.
1：Experimental Design and Method Selection:
The study employs a GAN architecture called AC-PG GAN for data augmentation and uses three CNN models (AlexNet, ResNet, SqueezeNet) for classification.
2：Sample Selection and Data Sources:
The dataset contains 507 single channel EL images of monocrystalline solar cells including four kinds of defects.
3：List of Experimental Equipment and Materials:
NVIDIA GeForce GTX 1080Ti graphics card, Pytorch
4：4 framework. Experimental Procedures and Operational Workflow:
The training procedure involves progressively growing GANs to generate high-resolution images, followed by classification using CNN models.
5：Data Analysis Methods:
The effectiveness of the augmented dataset is evaluated by comparing the classification accuracy before and after augmentation.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
