Data augmentation method based on improved deep convolutional generative adversarial networks

doi:10.11772/j.issn.1001-9081.2020071059

Abstract

Abstract: In order to solve the training difficulty of small sample data in deep learning and increase the training efficiency of DCGAN (Deep Convolutional Generative Adversarial Network), an improved DCGAN algorithm was proposed to perform the augmentation of small sample data. In the method, Wasserstein distance was used to replace the loss model in the original model at first. Then, spectral normalization was added in the generation network, and discrimination network to acquire a stable network structure. Finally, the optimal noise input dimension of sample was obtained by the maximum likelihood estimation and experimental estimation, so that the generated samples became more diversified. Experimental result on three datasets MNIST, CelebA and Cartoon indicated that the improved DCGAN could generate samples with higher definition and recognition rate compared to that before improvement. In particular, the average recognition rate on these datasets were improved by 8.1%, 16.4% and 16.7% respectively, and several definition evaluation indices on the datasets were increased with different degrees, suggesting that the method can realize the small sample data augmentation effectively.

Key words: small sample, data augmentation, Deep Convolutional Generative Adversarial Network （DCGAN), Wasserstein distance, spectral normalization, intrinsic dimension

摘要： 针对小样本数据在深度学习中训练难的问题，为提高DCGAN训练效率，提出了一种改进的DCGAN算法对小样本数据进行增强。首先，使用Wasserstein距离替换原模型中的损失模型；其次，在生成网络和判别网络中加入谱归一化，以得到稳定的网络结构；最后，通过极大似然估计算法和实验估算得到样本的最佳噪声输入维度，从而提高生成样本的多样性。在MNIST、CelebA和Cartoon这三个数据集上的实验结果表明：改进后的DCGAN所生成样本的清晰度以及识别率比改进前均得到了明显提高，其中平均识别率在这几个数据集上分别提高了8.1%、16.4%和16.7%，几种清晰度评价指标在各数据集上均有不同程度的提高。可见该方法能够有效地实现小样本数据增强。

关键词: 小样本, 数据增强, DCGAN, Wasserstein距离, 谱归一化, 内在维数

CLC Number:

TP391.4

GAN Lan, SHEN Hongfei, WANG Yao, ZHANG Yuejin. Data augmentation method based on improved deep convolutional generative adversarial networks[J]. Journal of Computer Applications, 2021, 41(5): 1305-1313.

甘岚, 沈鸿飞, 王瑶, 张跃进. 基于改进DCGAN的数据增强方法[J]. 计算机应用, 2021, 41(5): 1305-1313.

References

[1] SCHMIDHUBER J. Deep learning in neural networks:an overview[J]. Neural Networks,2015,61:85-117.
[2] LI D,WEN I. A genetic algorithm-based virtual sample generation technique to improve small data set learning[J]. Neurocomputing, 2014,143:222-230.
[3] HA M,LI J,TIAN J,et al. The key theorem of learning theory on g/sub/spl lambda//measure spaces[C]//Proceedings of the 2004 International Conference on Machine Learning and Cybernetics. Piscataway:IEEE,2004:1904-1907.
[4] 高智勇, 黄金镇, 杜程刚. 基于特征金字塔网络的肺结节检测[J]. 计算机应用,2020,40(9):2571-2576.(GAO Z Y,HUANG J Z,DU C G. Pulmonary nodule detection based on feature pyramid networks[J]. Journal of Computer Applications,2020,40(9):2571-2576.)
[5] SALAMON J,BELLO J P. Deep convolutional neural networks and data augmentation for environmental sound classification[J]. IEEE Signal Processing Letters,2017,24(3):279-283.
[6] KRIZHEVSKY A, SUTSKEVER I, IHINTON G E. ImageNet classification with deep convolutional neural networks[C]//Proceedings of the 25th International Conference on Neural Information Processing Systems. Red Hook, NY:Curran Associates Inc.,2012:1097-1105.
[7] SZEGEDY C,LIU W,JIA Y,et al. Going deeper with convolutions[C]//Proceedings of the 2015 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE,2015:1-9.
[8] SIMONYAN K,ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[EB/OL].[2020-09-14]. https://arxiv.org/pdf/1409.1556.pdf.
[9] HE K,ZHANG X,REN S,et al. Deep residual learning for image recognition[C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE, 2016:770-778.
[10] GOODFELLOW I J,POUGET-ABADIE J,MIRZA M,et al. Generative adversarial nets[C]//Proceedings of the 27th International Conference on Neural Information Processing Systems. Cambridge:MIT Press,2014:2672-2680.
[11] RADFORD A, METZ L, CHINTALA S. Unsupervised representation learning with deep convolutional generative adversarial networks[EB/OL].[2020-09-13]. https://arxiv.org/pdf/1511.06434.pdf.
[12] 王娟, 徐志京. HR-DCGAN方法的帕金森声纹样本扩充及识别研究[J]. 小型微型计算机系统,2019,40(9):2026-2032. (WANG J,XU Z J. Study on augmentation and recognition of Parkinson's voiceprint samples by HR-DCGAN method[J]. Journal of Chinese Computer Systems,2019,40(9):2026-2032.)
[13] ALYAFI B,DIAZ O,ELANGOVAN P,et al. Quality analysis of DCGAN-generated mammography lesions[C]//Proceedings of the 15th International Workshop on Breast Imaging. Bellingham, WA:SPIE,2020:No. 115130B.
[14] 戚永军, 顾军华, 栗位勋, 等. 基于生成对抗网络的肺结节图像数据增强方法[C]//中国计算机用户协会网络应用分会第二十三届网络新技术与应用年会论文集. 重庆:《计算机科学》编辑部,2019:199-203.(QI Y J,GU J H,LI W X,et al. Lung nodule image data enhancement method based on generative confrontation network[C]//Proceedings of the 23rd Annual Meeting of New Network Technology and Application of Network Application Branch of China Computer Users Association. Chongqing:Editorial Board of Computer Science, 2019:199-203.)
[15] ZHANG Y,FANG Q,QIAN S,et al. Knowledge-aware attentive Wasserstein adversarial dialogue response generation[J]. ACM Transactions on Intelligent Systems and Technology,2020,11(4):No. 37.
[16] HERNÁNDEZ JUAN C,RODRÍGUEZ J M,SIGARRETA J M. Mathematical properties of the hyperbolicity of circulant networks[J]. Advances in Mathematical Physics, 2015, 2015:No. 723451.
[17] MIYATO T, KATAOKA T, KOYAMA M, et al. Spectral normalization for generative adversarial networks[EB/OL].[2020-05-08]. https://arxiv.org/pdf/1802.05957.pdf.
[18] 孙钰, 李林燕, 叶子寒, 等. 多层次结构生成对抗网络的文本生成图像方法[J]. 计算机应用,2019,39(11):3204-3209.(SUN Y,LI L Y,YE Z H,et al. Text-to-image synthesis method based on multi-level structure generative adversarial networks[J]. Journal of Computer Applications,2019,39(11):3204-3209.)
[19] 张变兰, 路永钢, 张海涛. 基于KL散度和近邻点间距离的球面嵌入算法[J]. 计算机应用,2017,37(3):680-683,690. (ZHANG B L,LU Y G,ZHANG H T. Spherical embedding algorithm based on Kullback-Leibler divergence and distances between nearest neighbor points[J]. Journal of Computer Applications,2017,37(3):680-683,690.)
[20] 丁月, 汪学明. 基于改进特征加权的朴素贝叶斯分类算法[J]. 计算机应用研究,2019,36(12):3597-3600,3627.(DING Y, WANG X M. Naive Bayes classification algorithm based on improved feature weighting[J]. Application Research of Computers,2019,36(12):3597-3600,3627.)
[21] GUTJAHR W J. ACO algorithms with guaranteed convergence to the optimal solution[J]. Information Processing Letters,2002,82(3):145-153.
[22] PALMER C N A,IRVINE A D,TERR-ONKWIATKOWSKI A,et al. Common loss-of-function variants of the epidermal barrier protein filaggrin are a major predisposing factor for atopic dermatitis[J]. Nature Genetics,2006,38(4):441-446.
[23] TOROUS J,LIPSCHITZ J,NG M,et al. Dropout rates in clinical trials of smartphone apps for depressive symptoms:a systematic review and meta-analysis[J]. Journal of Affective Disorders, 2020,263:413-419.
[24] YU K,ZHANG T,GONG Y. Nonlinear learning using local coordinate coding[C]//Proceedings of the 22nd International Conference on Neural Information Processing Systems. Red Hook, NY:Curran Associates Inc.,2009:2223-2231.
[25] LEVINA E,BICKEL P J. Maximum likelihood estimation of intrinsic dimension[C]//Proceedings of the 17th International Conference on Neural Information Processing Systems. Cambridge:MIT Press,2004:777-784.
[26] 代强, 程曦, 王永梅, 等. 基于轻量自动残差缩放网络的图像超分辨率重建[J]. 计算机应用,2020,40(5):1446-1452.(DAI Q, CHENG X, WANG Y M, et al. Light-weight automatic residual scaling network for image super-resolution reconstruction[J]. Journal of Computer Applications,2020,40(5):1446-1452.)
[27] CHERN N N K,NEOW P A,ANG M H. Practical issues in pixelbased autofocusing for machine vision[C]//Proceedings of the 2001 IEEE International Conference on Robotics and Automation. Piscataway:IEEE,2001:2791-2796.
[28] ZHANG H, GOODFELLOW I, METAXAS D, et al. Selfattention generative adversarial networks[C]//Proceedings of the 36th International Conference on Machine Learning. New York:JMLR. org,2019:7354-7363.