Please wait a minute...
浙江大学学报(工学版)
Journal of ZheJiang University (Engineering Science)  2020, Vol. 54 Issue (4): 684-693    DOI: 10.3785/j.issn.1008-973X.2020.04.007
Computer Technology, Inf ormation Engineering     
Solar cell defect enhancement method based on generative adversarial network
Kun LIU1(),Xi WEN1,Min-ming HUANG2,Xin-xin YANG1,Jing-kun MAO3
1. College of Artificial Intelligence, Hebei University of Technology, Tianjin 300131, China
2. College of Big Data and Information Engineering, Guizhou University, Guiyang 550025, China
3. Tianjin Xinsi Technology Company with Limited Liability, Tianjin 300450, China
Download: HTML     PDF(2002KB) HTML
Export: BibTeX | EndNote (RIS)      

Abstract  

A solar cells defect sample enhancement method which is negative sample-guided generative adversarial network was proposed in order to solve the problem of sample imbalance of solar cells. The representation ability of positive samples features was promoted and the diversity of generated samples was improved by introducing many negative samples and mixing negative sample guidance loss in the generative adversarial network. An adaptive weight constraint method was designed to balance the representation ability of generators and discriminators, and the quality of generated samples was improved. The experimental results show that the proposed method outperforms deep convolutional generative adversarial network (DCGAN), Wasserstein generative adversarial network-gradient penalty (WGAN-GP) and first-order generative adversarial network (FOGAN) in generation quality and detection accuracy on electroluminescence (EL) defect data sets of solar cells. F-measure of the method was 10%, 8% and 5% higher than DCGAN, WGAN-GP and FOGAN respectively, which showed better data enhancement performance. The performance of the proposed method is better than DCGAN, WGAN-GP and FOGAN on strip steel surface defect dataset and DAGM 2007 public dataset, which shows certain generalization ability.

Key words： data imbalance      data enhancement      generative adversarial network (GAN)      solar cell      negative sample guidance     
Received: 10 November 2019      Published: 05 April 2020
CLC:  TP 391  
Service
E-mail this article
Add to my bookshelf
Add to citation manager
E-mail Alert
RSS
Articles by authors
Kun LIU
Xi WEN
Min-ming HUANG
Xin-xin YANG
Jing-kun MAO
Cite this article:

Kun LIU,Xi WEN,Min-ming HUANG,Xin-xin YANG,Jing-kun MAO. Solar cell defect enhancement method based on generative adversarial network. Journal of ZheJiang University (Engineering Science), 2020, 54(4): 684-693.

URL:

http://www.zjujournals.com/eng/10.3785/j.issn.1008-973X.2020.04.007     OR     http://www.zjujournals.com/eng/Y2020/V54/I4/684


基于生成对抗网络的太阳能电池缺陷增强方法

为了解决太阳能电池样本不均衡问题，提出负样本引导生成对抗网络的太阳能电池缺陷样本增强方法. 通过在生成对抗模型中引入大量负样本和增加负样本引导损失，促进模型对正样本特征的表达，提升生成样本的多样性；设计自适应的权值约束方法，平衡生成器和判别器的表达能力，提升生成样本的质量. 实验结果表明，在太阳能电池电致发光(EL)缺陷数据集上，提出方法的生成质量和检测精度优于深度卷积生成对抗网络（DCGAN）、梯度惩罚Wasserstein距离生成对抗网络（WGAN-GP）和一阶导数生成对抗网络（FOGAN）；该方法的F测度较DCGAN、WGAN-GP和FOGAN分别最高提升了10%、8%和5%，具有较好的数据增强性能. 在带钢表面缺陷数据集及DAGM 2007公共数据集上，提出方法的性能优于DCGAN、WGAN-GP和FOGAN，具有一定的泛化能力.


关键词： 样本不均衡,  数据增强,  生成对抗网络（GAN）,  太阳能电池,  负样本引导 
Fig.1 Model structure of NSG-GA
Fig.2 Feature visualization of different samples
Fig.3 EL image samples of solar cells
Fig.4 F-measure curves of different numbers of negative samples
Fig.5 Generated EL image samples of solar cells by DCGAN，WGAN-GP，FOGAN and NSGGAN
方法 s p1-NN MMD WD
DCGAN 20 0.81 0.35 4.3
DCGAN 50 0.79 0.33 4.0
DCGAN 100 0.76 0.31 3.8
DCGAN 200 0.75 0.30 3.6
WGAN-GP 20 0.73 0.32 4.2
WGAN-GP 50 0.73 0.31 3.9
WGAN-GP 100 0.72 0.28 3.7
WGAN-GP 200 0.69 0.27 3.5
FOGAN 20 0.68 0.29 3.8
FOGAN 50 0.65 0.25 3.4
FOGAN 100 0.61 0.20 3.1
FOGAN 200 0.58 0.18 2.9
NSGGAN 20 0.53 0.20 3.3
NSGGAN 50 0.53 0.18 3.2
NSGGAN 100 0.53 0.17 3.1
NSGGAN 200 0.52 0.15 2.8
Tab.1 Generated quality of each model on solar cell EL dataset
方法 GAN训练集 CNN训练集 CNN测试集
s n s n s n
没有数据增强 ? ? s 10 000 300?s 6 000
传统数据增强方法 s ? 10 000 10 000 300?s 6 000
DCGAN增强方法 s ? 10 000 10 000 300?s 6 000
WGAN-GP增强方法 s ? 10 000 10 000 300?s 6 000
FOGAN增强方法 s ? 10 000 10 000 300?s 6 000
NSGGAN增强方法 s 20 000 10 000 10 000 300?s 6 000
Tab.2 Samples sets of solar cell EL image
Fig.6 F-measure curves of solar cell EL dataset for different methods under different positive samples
对比方法 F
DCGAN 0.85
DCGAN + NSG 0.89
DCGAN + clip 0.91
DCGAN + adaptive_clip 0.92
DCGAN+NSG + clip 0.93
NSGGAN 0.96
Tab.3 F-measure index of each method on solar cell EL dataset
对比方法 p1-NN MMD WD
DCGAN 0.81 0.35 4.3
DCGAN + NSG 0.75 0.29 3.8
DCGAN + clip 0.74 0.27 3.5
DCGAN +adaptive_clip 0.71 0.25 3.5
DCGAN + NSG+ clip 0.69 0.22 3.3
NSGGAN 0.53 0.18 3.1
Tab.4 Generated quality of each method on solar cell EL defect dataset
Fig.7 Generated samples by DCGAN，WGAN-GP，FOGAN and NSGGAN
方法 s p1-NN MMD WD
DCGAN 20 0.83 0.42 5.1
DCGAN 50 0.80 0.40 4.9
DCGAN 100 0.78 0.37 4.5
DCGAN 200 0.76 0.36 4.3
WGAN-GP 20 0.78 0.41 4.7
WGAN-GP 50 0.76 0.40 4.5
WGAN-GP 100 0.75 0.39 4.4
WGAN-GP 200 0.73 0.36 4.2
FOGAN 20 0.85 0.46 5.3
FOGAN 50 0.74 0.42 4.8
FOGAN 100 0.70 0.37 4.3
FOGAN 200 0.67 0.35 4.0
NSGGAN 20 0.59 0.28 3.6
NSGGAN 50 0.58 0.27 3.5
NSGGAN 100 0.58 0.24 3.2
NSGGAN 200 0.55 0.23 3.0
Tab.5 Generated quality of each model on strip steel surface defect dataset
方法 s p1-NN MMD WD
DCGAN 20 0.88 0.53 5.4
DCGAN 50 0.87 0.51 5.2
DCGAN 100 0.85 0.50 5.5
DCGAN 200 0.83 0.47 4.8
WGAN-GP 20 0.85 0.51 4.9
WGAN-GP 50 0.83 0.47 4.6
WGAN-GP 100 0.82 0.45 4.5
WGAN-GP 200 0.80 0.44 4.3
FOGAN 20 0.90 0.56 5.8
FOGAN 50 0.86 0.50 5.5
FOGAN 100 0.83 0.46 5.0
FOGAN 200 0.78 0.43 4.5
NSGGAN 20 0.75 0.46 4.4
NSGGAN 50 0.73 0.43 4.2
NSGGAN 100 0.70 0.42 4.0
NSGGAN 200 0.68 0.40 3.9
Tab.6 Generated quality of each model on DAGM2007 dataset
Fig.8 F-measure curves of different methods under different numbers of positive samples
[1]   TSAI D M, LI G N, LI W C, et al Defect detection in multi-crystal solar cells using clustering with uniformity measures[J]. Advanced Engineering Informatics, 2015, 29 (3): 419- 430
doi: 10.1016/j.aei.2015.01.014
[2]   林剑春, 杨爱军, 沈熠辉 电致发光缺陷检测仪的成像性能评估[J]. 光学 精密工程, 2017, 25 (6): 1418- 1424
LIN Jian-chun, YANG Ai-jun, SHEN Yi-hui Evaluation of imaging performance for electroluminescence defect detector[J]. Optics and Precision Engineering, 2017, 25 (6): 1418- 1424
doi: 10.3788/OPE.20172506.1418
[3]   TSAI D M, LIN M C Machine-vision-based identify-cation for wafer tracking in solar cell manufacturing[J]. Robotics and Computer Integrated Manufacturing, 2013, 29 (5): 312- 321
doi: 10.1016/j.rcim.2013.01.009
[4]   夏英, 李刘杰, 张旭, 等 基于层次聚类的不平衡数据加权过采样方法[J]. 计算机科学, 2019, 46 (4): 28- 33
XIA Ying, LI Liu-jie, ZHANG Xu, et al Weighted oversampling method based on hierarchical clustering for unbalanced data[J]. Computer Science, 2019, 46 (4): 28- 33
[5]   王宪保, 李洁, 姚明海, 等 基于深度学习的太阳能电池片表面缺陷检测方法[J]. 模式识别与人工智能, 2014, 27 (6): 517- 523
WANG Xian-bao, LI Jie, YAO Ming-hai, et al Solar cells surface defects detection based on deep learning[J]. Pattern Recognition and Artificial Intelligence, 2014, 27 (6): 517- 523
[6]   SHI J J, PING W, PEI Y J, et al. Research on data augmentation for image classification based on convolution neural networks [C] // 2017 Chinese Automation Congress. Beijing: IEEE, 2017: 4165- 4170.
[7]   GOODFELLOW I, POUGET-ABADIE J, MIRZA M, et al. Generative adversarial nets [C] // Advances in Neural Information Processing Systems. Montréal: MIT Press, 2014: 2672-2680.
[8]   SANDFORT V, YAN K, PICKHARDT P J, et al Data augmentation using generative adversarial networks (CycleGAN) to improve generalizability in CT segmentation tasks[J]. Scientific Reports, 2019, 9 (1): 1- 9
doi: 10.1038/s41598-018-37186-2
[9]   ZHONG Z, ZHENG L, ZHENG Z, et al. Camera style adaptation for person re-identification [C] // Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Utah: IEEE, 2018: 5157-5166.
[10]   LIU L, PAN Z, QIU X, et al. SAR target classification with cycle-gan transferred simulated samples [C] // 2018 IEEE International Geoscience and Remote Sensing Symposium. Valencia: IEEE, 2018: 4411-4414.
[11]   LI Z, WANG Y, YU J. Reconstruction of thin-slice medical images using generative adversarial network [C] // International Workshop on Machine Learning in Medical Imaging. Berlin: Springer, 2017: 325-333.
[12]   GUAN J, LI R, YU S, et al. Generation of synthetic electronic medical record text [C] // 2018 IEEE International Conference on Bioinformatics and Biomedicine. Madrid: IEEE, 2018: 374-380.
[13]   WANG Q, ZHOU X, WANG C, et al WGAN-based synthetic minority over-sampling technique: improving semantic fine-grained classification for lung nodules in CT images[J]. IEEE Access, 2019, 7 (1): 18450- 18463
[14]   HAN C, HAYASHI H, RUNDO L, et al. GAN-based synthetic brain MR image generation [C] // 2018 IEEE 15th International Symposium on Biomedical Imaging. Washington: IEEE, 2018: 734-738.
[15]   REZAEI M, YANG H, HARMUTH K, et al. Conditional generative adversarial refinement networks for unbalanced medical image semantic segmentation [C] // 2019 IEEE Winter Conference on Applications of Computer Vision. Washington: IEEE, 2019: 1836-1845.
[16]   FANG W, ZHANG F, SHENG V S, et al A method for improving CNN-based image recognition using DCGAN[J]. Computers Materials and Continua, 2018, 57 (1): 167- 178
[17]   LIN T, HU Z, GUO X. Sparsemax and relaxed wasserstein for topic sparsity [C] // Proceedings of the 12th ACM International Conference on Web Search and Data Mining. Melbourne: ACM, 2019: 141-149.
[18]   ARJOVSKY M, CHINTALA S, BOTTOU L. Wasserstein generative adversarial networks [C] // Proceedings of the 34th International Conference on Machine Learning. Sydney: ACM, 2017: 214-223.
[19]   GULRAJANI I, AHMED F, ARJOVSKY M, et al. Improved training of Wasserstein GANs [C] // Advances in Neural Information Processing Systems. Long Beach: MIT Press, 2017: 5767-5777.
[20]   SEWARD C, UNTERTHINER T, BERGMANN U, et al First order generative adversarial networks[J]. Proceedings of the 35th International Conference on Machine Learning, 2018, 80 (1): 4567- 4576
[21]   SHMELKOV K, SCHMID C, ALAHARI K. How good is my GAN [C]// Proceedings of the European Conference on Computer Vision. Munich: Springer, 2018: 213-229.
[22]   SU B Y, CHEN H Y, ZHU Y, et al Classification of manufacturing defects in multicrystalline solar cells with novel feature descriptor[J]. IEEE Transactions on Instrumentation and Measurement, 2019, 68 (12): 4675- 4688
doi: 10.1109/TIM.2019.2900961
[23]   WANG H Y, ZHANG J W, TIAN Y, et al A simple guidance template-based defect detection method for strip steel surfaces[J]. IEEE Transactions on Industrial Informatics, 2018, 15 (5): 2798- 2809
[1] Hui-ya HU,Shao-yan GAI,Fei-peng DA. Face frontalization based on generative adversarial network[J]. Journal of ZheJiang University (Engineering Science), 2021, 55(1): 116-123.
[2] RAO Lei, JI Chun lei. Triple junction solar cells characteristics under variable temperature and concentration at AM0[J]. Journal of ZheJiang University (Engineering Science), 2015, 49(12): 2269-2275.
[3] ZHANG Dan-Yan, TUN Xiao-Bei, DIAO Meng-Lian, et al. Integrated circuit design for solar cell maximum power point tracking[J]. Journal of ZheJiang University (Engineering Science), 2009, 43(11): 2000-2005.