缺陷微小微弱、数据样本规模不足等是制约太阳电池质量检测技术快速发展的瓶颈问题。因此,该文提出一种用于生成太阳电池缺陷图像的生成对抗网络模型FAGAN。该方法先在源域公开道路数据集上进行模型预训练提取跨域底层视觉特征,以提升FAGAN在目标域生成缺陷形式的多样性;然后设计了有效通道空间注意力ECSA,在空间与通道两个维度对缺陷特征进行增强,以提升微小微弱缺陷生成样本的质量。实验结果表明：该文提出的方法所生成的太阳电池缺陷图像的性能评价指标FID优于现有的梯度惩罚Wasserstein距离生成对抗网络（WGAN-GP）、深度卷积生成对抗网络（DCGAN）、循环生成对抗网络（CycleGAN）和样式生成对抗网络（StyleGAN）。
Abstract
Small and weak defects and insufficient dataset are the bottlenecks problems which restrict the rapid development of solar cells quality detection technology. Therefore, this paper proposes FAGAN for generating defect images of solar cells. This method firstly performs model pre-training on the source domain open road dataset to extract cross-domain underlying visual features, so as to improve the diversity of defect forms generated by FAGAN on the target domain; then, ECSA is designed to enhance the defect features in two dimensions of space and channel, so as to improve the quality of small and weak defect samples. The experiments show that the FID of the solar cell defect images generated by the proposed method is better than those of the existing gradient penalty Wasserstein distance generative adversarial network （WGAN-GP）, deep convolution generative adversarial network （DCGAN）, cycle generative adversarial network （CycleGAN） and style generative adversarial network （StyleGAN）.
关键词
迁移学习 /
太阳电池 /
图像处理 /
卷积神经网络 /
生成对抗网络 /
注意力
Key words
transfer learning /
solar cells /
image processing /
convolutional neural networks /
generative adversarial network /
attention
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 李琼, 吴文宝, 刘斌, 等. 基于迁移学习的光伏组件鸟粪覆盖检测[J]. 太阳能学报, 2022, 43(2): 233-237.
LI Q, WU W B, LIU B, et al.Bird droppings coverage detection of photovoltaic module based on transfer learning[J]. Acta energiae solaris sinica, 2022, 43(2): 233-237.
[2] ZHONG Z, ZHENG L, KANG G L, et al.Random erasing data augmentation[C]//Proceedings of the AAAI Conference on Artificial Intelligence. New York, NY, USA, 2020.
[3] ZHANG H Y, CISSE M, DAUPHIN Y N, et al.Mixup: beyond empirical risk minimization[C]//International Conference on Learning Representations. Vancouver, Canada, 2018.
[4] GOODFELLOW I J, POUGET-ABADIE J, MIRZA M, et al.Generative adversarial networks[C]//Proceedings of the 27th International Conference on Neural Information Processing Systems. Montreal, Canada, 2014.
[5] NIU S L, LI B, WANG X G, et al.Region-and strength-controllable GAN for defect generation and segmentation in industrial images[J]. IEEE transactions on industrial informatics, 2022, 18(7): 4531-4541.
[6] NIU S L, LI B, WANG X G,et al.Defect image sample generation with GAN for improving defect recognition[J]. IEEE transactions on automation science and engineering, 2020, 17(3): 1611-1622.
[7] WANG Y X, GONZALEZ-GARCIA A, BERGA D, et al.MineGAN: effective knowledge transfer from GANs to target domains with few images[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). Seattle, WA, USA, 2020: 9332-9341.
[8] ZHANG H, GOODFELLOW I, METAXAS D, et al.Self-attention generative adversarial networks[C]//International Conference on Machine Learning, Long Beach, CA, USA, 2019: 7354-7363.
[9] 王晋宇, 张长弓, 杨海涛, 等. 基于注意力残差网络的卫星图像翻译方法[J]. 激光与光电子学进展, 2022, 59(2): 232-242.
WANG J Y, ZHANG C G, YANG H T, et al.Satellite image translation method based on attention residual network[J]. Laser & optoelectronics progress, 2022, 59(2): 232-242.
[10] ZHU J Y, PARK T, ISOLA P, et al.Unpaired image-to-image translation using cycle-consistent adversarial networks[C]//2017 IEEE International Conference on Computer Vision(ICCV), Venice, Italy, 2017: 2223-2232.
[11] WANG Q L, WU B G, ZHU P F, et al.ECA-Net: efficient channel attention for deep convolutional neural networks[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). Seattle, WA, USA, 2020.
[12] WANG Z, BOVIK A C, SHEIKH H R, et al.Image quality assessment: from error visibility to structural similarity[J]. IEEE transactions on image processing, 2004, 13(4): 600-612.
[13] HEUSEL M, RAMSAUER H, UNTERTHINER T, et al.GANs trained by a two time-scale update rule converge to a local Nash equilibrium[C]//Neural Information Processing Systems. Long Beach, CA, USA, 2017: 6626-6637.
基金
国家自然科学基金（62173124）; 河北省自然科学基金（F2019202305）
PDF(1817 KB)
