针对太阳电池缺陷图像数据少导致深度学习模型过拟合,从而造成小样本缺陷检测困难的问题,该文提出一种融合元学习和双路径注意力的生成对抗网络（MAGAN）作为数据增强策略。所设计的元学习调参模块（MTM）优化生成器中权重参数;所设计的双路径注意力模块（DPAT）在特征提取过程中更关注图像中微小微弱缺陷特征;在改进网络构架的同时提出一种聚类约束损失函数解决训练过程中梯度消失问题。实验和研究结果表明,所提方法能够针对小样本缺陷生成有效目标数据集并优于其他生成对抗网络,最后通过分类准确率验证了该网络的有效性。
Abstract
Due to over-fitting of deep learning models brought on by the absence of picture data containing solar cell defects, few-shot defect identification has become challenging. As a method of enhancing the data, this study proposes employing meta-attention generative adversarial networks （MAGAN）, which incorporate meta-learning and dual-path attention. The dual-path attention module （DPAT） is designed to pay more attention to the small and faint defect features in the image during the feature extraction process. A clustering constrained loss function is proposed to solve the gradient disappearance problem during the training process while improving the network architecture. The meta-learning tuning module （MTM） is designed to optimize the weight parameters in the generator. The study and experiment discoveries demonstrate that the suggested strategy outperforms previous generative adversarial networks and may produce useful target datasets for modest sample faults.
关键词
太阳电池 /
缺陷检测 /
生成对抗网络 /
元学习 /
双路径注意力 /
数据增强
Key words
solar cells /
defect detection /
generative adversarial networks /
meta-learning /
dual-path attention /
data augmentation
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 张治, 邹鹏辉, 刘志锋, 等. 发射极掺杂工艺对产业化IBC太阳电池性能的影响[J]. 太阳能学报, 2022, 43(3): 158-162.
ZHANG Z, ZOU P H, LIU Z F, et al.Influence of emitter doping process on performance of industrial IBC solar cell[J]. Acta energiae solaris sinica, 2022, 43(3): 158-162.
[2] 仝勖峰, 金嘉祺, 袁晓军, 等. 基于机器视觉的区域太阳直接辐射动态预测方法研究[J]. 太阳能学报, 2021, 42(6): 247-255.
TONG X F, JIN J Q, YUAN X J, et al.Research on regional solar direct normal irradiance dynamic forecasting based on machine vision[J]. Acta energiae solaris sinica, 2021, 42(6): 247-255.
[3] EL HAJ Y, EL-HAG A H, GHUNEM R A. Application of deep-learning via transfer learning to evaluate silicone rubber material surface erosion[J]. IEEE transactions on dielectrics and electrical insulation, 2021, 28(4): 1465-1467.
[4] CUBUK E D, ZOPH B, MANÉ D, et al.AutoAugment: learning augmentation strategies from data[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Long Beach, CA, USA, 2020: 113-123.
[5] REIPSCHLAGER P, FLEMISCH T, DACHSELT R.Personal augmented reality for information visualization on large interactive displays[J]. IEEE transactions on visualization and computer graphics, 2021, 27(2): 1182-1192.
[6] GOODFELLOW I J, POUGRT A J, MIRZA M, et al.Generative adversarial nets[C]//Proceedings of the 27th International Conference on Neural Information Processing Systems, Montreal, Cambridge, UK, 2014: 2672-2680.
[7] TAO X, ZHANG D P, WANG Z H, et al.Detection of power line insulator defects using aerial images analyzed with convolutional neural networks[J]. IEEE transactions on systems, man, and cybernetics: systems, 2020, 50(4): 1486-1498.
[8] HANG R L, ZHOU F, LIU Q S, et al.Classification of hyperspectral images via multitask generative adversarial networks[J]. IEEE transactions on geoscience and remote sensing, 2021, 59(2): 1424-1436.
[9] RADFORD A, METZ L, CHINTALA S.Unsupervised representation learning with deep convolutional generative adversarial networks[J]. Computer ence, 2015. DOI:10.48550/arXiv.1511.06434.
[10] ARJOVSKY M, CHINTALA S, BOTTOU L. Wasserstein GAN[EB/OL].2017: arXiv: 1701.07875. https://arxiv.org/abs/1701.07875.
[11] ZHANG H, GOODFELLOW I, METAXAS D, et al. Self-attention generative adversarial networks[EB/OL].2018: arXiv: 1805.08318. https://arxiv.org/abs/1805.08318.
[12] CHEN Z T, FU Y W, ZHANG Y D, et al.Multi-level semantic feature augmentation for one-shot learning[J]. IEEE transactions on image processing, 2019, 28(9): 4594-4605.
[13] PENG S G.Overview of meta-reinforcement learning research[C]//2020 2nd International Conference on Information Technology and Computer Application (ITCA), Guangzhou, China, 2021: 54-57.
[14] PARK S, JANG H, SIMEONE O, et al.Learning to demodulate from few pilots via offline and online meta-learning[J]. IEEE transactions on signal processing, 2021, 69: 226-239.
[15] 叶星余, 何元烈, 汝少楠. 基于生成式对抗网络及自注意力机制的无监督单目深度估计和视觉里程计[J]. 机器人, 2021, 43(2): 203-213.
YE X Y, HE Y L, RU S N.Unsupervised monocular depth estimation and visual odometry based on generative adversarial network and self-attention mechanism[J]. Robot, 2021, 43(2): 203-213.
[16] WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[EB/OL].2018: arXiv: 1807.06521. https://arxiv.org/abs/1807.06521.
基金
国家自然科学基金（62073117; U21A20482）; 中央引导地方科技发展资金项目（206Z1701G）
PDF(2005 KB)
