基于图像梯度的数据增广方法

doi:10.3969/j.issn.0255-8297.2021.02.012

摘要/Abstract

摘要： 卷积神经网络用于图像识别的分类任务，需要大规模的图像数据集进行训练。因需要采集目标图像数量和设备条件的限制，采用常规方法难以获取足够多的图像样本，且耗时耗力耗财。目前已提出了多种多样的样本增广方法来解决图像样本不足的问题，本文介绍了数据增广的研究背景和意义。以提高卷积神经网络的图像识别的准确率为目的，针对图像数据增广提出了基于图像梯度的数据增广方法。选取最大图像梯度值，通过精准裁剪方法增加图像样本，扩增图像数据集，使用增广后的数据集对卷积神经网络进行训练。应用Tensorflow深度学习框架和VGG16网络模型，选取PlantVillage的部分数据集，将训练集数据增广至原来的6倍，对扩增前后的训练集进行训练和对比。实验结果表明：使用数据增广后训练集训练的模型的准确率提升4.18%。

关键词: 数据增广, 图像梯度, 卷积神经网络, Tensorflow深度学习框架, PlantVillage数据集

Abstract: As used in classification of image recognition, convolutional neural network requires large-scale image data set for training. Due to the limitation of the number of target images to be collected and the conditions of image acquisition equipment, it is difficult to obtain enough image samples by conventional methods because of time-consuming, laborconsuming and money-consuming. In order to solve the insufficiency of image samples, a variety of sample enlargement methods have been proposed. This paper introduces the research background and significance of data augmentation. For the purpose of improving the accuracy of image recognition of convolutional neural network, a data augmentation method based on image gradient is proposed. The image gradient is selected to increase image sample and enlarge image data set by precise clipping method, and the convolutional neural network is trained with the expanded data set. By using Tensorflow deep learning framework and VGG16 network model, and selecting some data sets of PlantVillage, the training set data can be expanded to 6 times of the original. The training set before and after the expansion is trained and compared. Experimental results show that the accuracy rate of the model trained by the training set after data augmentation is increased by 4.18%.

Key words: data augmentation, image gradient, convolutional neural network, TensorFlow deep learning framework, PlantVillage dataset

中图分类号:

TP181

刘之瑜, 张淑芬, 刘洋, 罗长银, 李敏. 基于图像梯度的数据增广方法[J]. 应用科学学报, 2021, 39(2): 302-311.

LIU Zhiyu, ZHANG Shufen, LIU Yang, LUO Changyin, LI Min. Data Augmentation Method Based on Image Gradient[J]. Journal of Applied Sciences, 2021, 39(2): 302-311.

参考文献

[1] 张慧, 王坤峰, 王飞跃. 深度学习在目标视觉检测中的应用进展与展望[J]. 自动化学报, 2017, 43(8):1289-1305. Zhang H, Wang K F, Wang F Y. Advances and perspectives on applications of deep learning in visual object detection[J]. Acta Automatica Sinica, 2017, 43(8):1289-1305. (in Chinese)
[2] Krizhevsky A, Sutskever I, Hinton G. Imagenet classification with deep convolutional neural networks[J]. Advances in neural information processing systems, 2012, 25(2):1097-1105.
[3] Russakovsky O, Deng J, Su H, et al. ImageNet large scale visual recognition challenge[J]. International Journal of Computer Vision, 2015, 115(3):211-252.
[4] Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[DB/OL]. 2014[2020-08-28]. https://arxiv.org/abs/1409.1556.
[5] Szegedy C, Liu W, Jia Y Q, et al. Going deeper with convolutions[DB/OL]. 2014[2020-08-28]. https://arxiv.org/abs/1409.4842.
[6] He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C]//IEEE Conference on Computer Vision & Pattern Recognition. IEEE Computer Society, 2016.
[7] 王建仁, 马鑫, 段刚龙, 等. 边缘智能背景下的手写数字识别[J]. 计算机应用, 2019, 39(12):3548-3555. Wang J R, Ma X, Duan G L, et al. Handwritten digit recognition based on edge intelligence[J]. Journal of Computer Applications, 2019, 39(12):3548-3555. (in Chinese)
[8] 尚瑛杰, 董丽亚, 何虎. 基于脉冲神经网络的迁移学习算法与软件框架[J]. 计算机工程, 2020, 46(3):53-59. Shang Y J, Dong L Y, He H. Transfer learning algorithm and software framework based on spiking neuron network[J]. Computer Engineering, 2020, 46(3):53-59. (in Chinese)
[9] 雷雨, 韩德俊, 曾庆东, 等. 基于高光谱成像技术的小麦条锈病病害程度分级方法[J]. 农业机械学报, 2018, 49(5):226-232. Lei Y, Han D J, Zeng Q D, et al. Grading method of disease severity of wheat stripe rust based on hyperspectral imaging technology[J]. Transactions of the Chinese Society for Agricultural Machinery, 2018, 49(5):226-232. (in Chinese)
[10] 李双峰. TensorFlow lite:端侧机器学习框架[J]. 计算机研究与发展, 2020, 57(9):1839-1853. Li S F. Tensorflow lite:end side machine learning framework[J]. Journal of Computer Research and Development, 2020, 57(9):1839-1853. (in Chinese)
[11] 李新叶, 龙慎鹏, 朱婧. 基于深度神经网络的少样本学习综述[J]. 计算机应用研究, 2020, 37(8):2241-2247. Li X Y, Long S P, Zhu J. Survey of few-shot learning based on deep neural network[J]. Application Research of Computers, 2020, 37(8):2241-2247. (in Chinese)
[12] 徐龙飞, 郁进明. 不同优化器在高斯噪声下对LR性能影响的研究[J]. 计算机技术与发展, 2020, 30(3):7-12. Xu L F, Yu J M. Study on influence of different optimizers on performance of LR under gaussian noise[J]. Computer Technology and Development, 2020, 30(3):7-12. (in Chinese)
[13] 胡文涛, 陈秀宏. 基于局部保持投影的鲁棒稀疏子空间学习[J/OL]. 计算机工程与应用, (2020-06-02)[2020-08-28]. http://kns.cnki.net/kcms/detail/11.2127.TP.20200602.1322. 006.html.Hu W T, Chen X H. Robust sparse subspace learning based on local preserving projection[J/OL]. Computer Engineering and Applications, (2020-06-02)[2020-08-28]. http://kns.cnki.net/kcms/detail/11.2127.TP.20200602.1322.006.html. (in Chinese)
[14] 张晓峰, 吴刚. 基于生成对抗网络的数据增强方法[J]. 计算机系统应用, 2019, 28(10):201-206. Zhang X F, Wu G. Data augmentation method based on generative adversarial network[J]. Computer Systems Applications, 2019, 28(10):201-206. (in Chinese)
[15] 林懿伦, 戴星原, 李力, 等. 人工智能研究的新前线:生成式对抗网络[J]. 自动化学报, 2018, 44(5):775-792. Lin Y L, Dai X Y, Li L, et al. The new frontier of AI research:generative adversarial networks[J]. Acta Automatica Sinica, 2018, 44(5):775-792. (in Chinese)
[16] 王坤峰, 苟超, 段艳杰, 等. 生成式对抗网络GAN的研究进展与展望[J]. 自动化学报, 2017, 43(3):321-332. Wang K F, Gou C, Duan Y J, et al. The research progress and prospect of GAN in the generated counter network[J]. Acta Automatica Sinica, 2017, 43(3):321-332. (in Chinese)
[17] 梁俊杰, 韦舰晶, 蒋正锋. 生成对抗网络GAN综述[J]. 计算机科学与探索, 2020, 14(1):1-17. Liang J J, Wei J J, Jiang Z F. Survey of generation countermeasure network GAN[J]. Journal of Frontiers of Computer Science and Technology, 2020, 14(1):1-17. (in Chinese)
[18] Chatfield K, Simonyan K, Vedaldi A, et al. Return of the devil in the details:delving deep into convolutional nets[DB/OL]. 2014[2020-08-28]. https://arxiv.org/abs/1405.3531.
[19] Raitoharju J, Riabchenko E, Meissner K, et al. Data enrichment in fine-grained classification of aquatic macroinvertebrates[C]//2016 ICPR 2nd Workshop on Computer Vision for Analysis of Underwater Imagery (CVAUI). IEEE Computer Society, 2017:43-48.
[20] Wang L M, Xiong Y J, Wang Z, et al. Towards good practices for very deep two-stream convnets[DB/OL]. 2015[2020-08-28]. https://arxiv.org/abs/1507.02159.
[21] Cubuk E D, Zoph B, Mane D, et al. AutoAugment:learning augmentation strategies from data[DB/OL]. 2019[2020-08-28]. https://arxiv.org/abs/1805.09501.
[22] Derrance T, Taylor G W. Improved regularization of convolutional neural networks with cutout[DB/OL]. 2017[2020-08-28]. https://arxiv.org/abs/1708.04552.
[23] 蒋梦莹, 林小竹, 柯岩. 基于优化分类的数据增广方法[J]. 计算机工程与设计, 2018, 39(11):3559-3563. Jiang M Y, Lin X Z, Ke Y. Data augmented method based on optimized classification[J]. Computer Engineering and Design, 2018, 39(11):3559-3563. (in Chinese)
[24] 郑远攀, 李广阳, 李晔. 深度学习在图像识别中的应用研究综述[J]. 计算机工程与应用, 2019, 55(12):20-36. Zheng Y P, Li G Y, Li Y. A survey of the application of deep learning in image recognition[J]. Computer Engineering and Applications, 2019, 55(12):20-36. (in Chinese)
[25] 李猛, 李艳玲, 林民. 命名实体识别的迁移学习研究综述[J/OL]. 计算机科学与探索,[2020-08-28]. http://kns.cnki.net/kcms/detail/11.5602.TP.20200925.1522.010.html. Li M, Li Y L, Lin M. A survey of transfer learning in named entity recognition[J/OL]. Journal of Frontiers of Computer Science and Technology,[2020-08-28]. http://kns.cnki.net/kcms/detail/11.5602.TP.20200925.1522.010.html. (in Chinese)
[26] Mehrotra R, Ansari M A, Agrawal R, et al. A transfer learning approach for AI-based classification of brain tumors[J]. Machine Learning with Applications, 2020, 2:100003.