一种改进的LeNet网络

doi:10.16180/j.cnki.issn1007-7820.2019.03.011

电子科技 ›› 2019, Vol. 32 ›› Issue (3): 53-57.doi: 10.16180/j.cnki.issn1007-7820.2019.03.011

一种改进的LeNet网络

胡德敏,程普芳

上海理工大学光电信息与计算机工程学院,上海 200093

收稿日期:2018-03-18 出版日期:2019-03-15 发布日期:2019-03-01
作者简介:胡德敏(1963-),男,博士,副教授。研究方向:计算机网络、分布式计算、云计算。|程普芳(1991-),男,硕士研究生。研究方向:图像分类、深度学习。
基金资助:
国家自然科学基金(61170277);国家自然科学基金(61472256);上海市教委科研创新重点项目(12zz17);上海市一流学科建设项目(S1201YLXK)

An Improved LeNet Network

HU Demin,CHENG Pufang

School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China

Received:2018-03-18 Online:2019-03-15 Published:2019-03-01
Supported by:
National Natural Science Foundation of China(61170277);National Natural Science Foundation of China(61472256);Key Project of Scientific Research and Innovation of Shanghai Municipal Education Commission(12zz17);First Class Construction Project of Shanghai(S1201YLXK)

摘要/Abstract

摘要：

针对卷积神经网络中存在的学习效率低、收敛速度慢、训练时间长等问题,文中提出一种改进的LeNet卷积神经网络模型。该模型使用卷积核大小为3,步幅为2的卷积层代替原有的池化层,并在每层激活函数之前添加批量归一化层。在Mnist和Cifar-10数据集上放入实验证明,相比于传统的LeNet网络,所提出的卷积神经网络提高了分类准确率,并且具有更快的收敛速度及更短的训练时间。

关键词: 图像分类, 卷积神经网络, 批量归一化, 池化层, 卷积核, 随机梯度下降法

Abstract: Aim

ing at the problems of low learning efficiency, slow convergence and long training time in convolutional neural networks, this paper presented an improved LeNet convolutional neural network model. The model used a convolutional kernel whose convolution scale was set as 3 and stride was set as 2 instead of the original pooled layer, and added a batch normalization layer before each activation function layer. Experiments on the Mnist dataset showed that compared with the traditional LeNet network, the convolutional neural network proposed in this paper improved the accuracy rate and had faster convergence speed and shorter training time.

Key words: image classification, convolutional neural network, batch normalization, pooling layer, convolution kernel, stochastic gradient descent.

中图分类号:

TP183

胡德敏,程普芳. 一种改进的LeNet网络[J]. 电子科技, 2019, 32(3): 53-57.

HU Demin,CHENG Pufang. An Improved LeNet Network[J]. Electronic Science and Technology, 2019, 32(3): 53-57.

图/表 6

图1

表1

表2

表3

图2

图3

参考文献 16

[1]	Lecun Y, Bengio Y, Hinton G . Deep learning[J]. Nature, 2015,521(7553):436-444.
[2]	Lecun Y, Bottou L, Bengio Y , et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998,86(11):2278-2324. doi: 10.1109/5.726791
[3]	熊海朋, 陈洋洋, 陈春玮 . 基于卷积神经网络的场景图像文本定位研究[J]. 电子科技, 2018,31(1):50-53.
	Xiong Haipeng, Chen Yangyang, Chen Chunwei . Text location in image based on convolution neural network[J]. Electronic Science and Technology, 2018,31(1):50-53.
[4]	Krizhevsky A, Sutskever I, Hinton G E. ImageNet classification with deep convolutional neural networks [C]. Lake Tahoe:International Conference on Neural Information Processing Systems, 2012.
[5]	Simonyan K, Zisserman A . Very deep convolutional networks for large-scale image recognition[J]. arXiv Preprint arXiv, 2014,1409(1556):1-14.
[6]	Szegedy C, Liu W, Jia Y, et al. Going deeper with convolutions [C].Boston:Computer Vision and Pattern Recognition IEEE, 2015.
[7]	He K, Zhang X, Ren S, et al. Deep residual learning for image recognition [C]. Las Vegas:IEEE Conference on Computer Vision and Pattern Recognition, 2016.
[8]	Springenberg J T, Dosovitskiy A, Brox T , et al. Striving for simplicity: the all convolutional net[J]. arXiv Preprint arXiv, 2014,1412(6806):1-14.
[9]	Goodfellow I J, Warde-Farley D, Mirza M , et al. Maxout networks[J]. arXiv Preprint arXiv, 2013,1302(4389):1-9.
[10]	Lin M, Chen Q, Yan S . Network in network[J]. arXiv preprint arXiv, 2013,1312(4400):1-10.
[11]	Srivastava N, Salakhutdinov R. Discriminative transfer learning with tree-based priors [C].Lake Tahoe:Advances in Neural Information Processing Systems, 2013.
[12]	Zeiler M D, Fergus R . Stochastic pooling for regularization of deep convolutional neural networks[J]. arXiv Preprint arXiv, 2013,1301(3557):1-9.
[13]	Lee C Y, Xie S, Gallagher P , et al. Deeply-supervised nets[J]. arXiv Preprint arXiv, 2015,1409(5185):1-9.
[14]	Ioffe S, Szegedy C . Batch normalization:accelerating deep network training by reducing internal covariate shift [C]. Lille:International Conference on Machine Learning, 2015.
[15]	周飞燕, 金林鹏, 董军 . 卷积神经网络研究综述[J]. 计算机学报, 2017,40(6):1229-1251.
	Zhou Feiyan, Jin Linpeng, Dong Jun . Review of convolutional neural network[J]. Chinese Journal of Computers, 2017,40(6):1229-1251.
[16]	Sutskever I, Martens J, Dahl G, et al. On the importance of initialization and momentum in deep learning [C].Atlanta: International Conference on Machine Learning, 2013.

参数	数值
处理器	英特尔第四代酷睿 i5-4210H @2.90 GHz
内存	12.0 GB
操作系统	WIN10 64位
编程软件	Python3.5 + TensorFlow框架(纯CPU模式)

LeNet层	LeNet参数	本文层级	本文结构参数
C1	156	C1	156+2+6*2
S2	0	C2	0+(33+1)6+2+6*2
C3	1516	C3	1516+2+16*2
S4	0	C4	0+(33+1)16+2+16*2
C5	48 120	C5	1 516+2+1*2
F6	10 164	F6	10 164+2+1*2

结构	Mnist准确率	Cifar-10准确率
LeNet	90.58%	73.46%
只去除池化层	91.80%	75.78%
只加BN	94.36%	78.69%
本文结构	96.42%	80.08%

一种改进的LeNet网络

An Improved LeNet Network

RichHTML

PDF (PC)

赞

可视化

摘要/Abstract

引用本文

使用本文

图/表 6

参考文献 16

相关文章 15

Metrics

本文评价

推荐阅读 10

[1]	李辉,王一丞. 基于CNNCIFG-Attention模型的文本情感分类[J]. 电子科技, 2022, 35(2): 46-51.
[2]	程晓雅,张雷. 基于深度学习的有遮挡人脸识别方法研究[J]. 电子科技, 2022, 35(1): 35-39.
[3]	战荫伟,朱百万,杨卓. 一种车脸识别算法的研究与应用[J]. 电子科技, 2021, 34(8): 1-7.
[4]	廖璐明,张伟. 基于改进VGG16网络的混合批量训练交通标志识别[J]. 电子科技, 2021, 34(8): 8-13.
[5]	马立新,豆晨飞,宋晨灿,杨天笑. 基于特征融合卷积神经网络的绝缘子无损检测[J]. 电子科技, 2021, 34(7): 26-30.
[6]	顾伟,李菲菲,陈虬. 基于多特征融合的行人检测方法[J]. 电子科技, 2021, 34(5): 29-34.
[7]	薛永杰,巨志勇. 基于改进AlexNet的鱼类识别算法[J]. 电子科技, 2021, 34(4): 12-17.
[8]	闫书豪,乔美英. 基于一维WConv-BiLSTM的轴承故障诊断算法[J]. 电子科技, 2021, 34(4): 75-82.
[9]	葛靖,刘子龙. 基于CNN和LSTM的睡眠呼吸暂停检测算法[J]. 电子科技, 2021, 34(2): 21-26.
[10]	朱斌,刘子龙. 基于新型初始模块的卷积神经网络图像分类方法[J]. 电子科技, 2021, 34(2): 52-56.
[11]	张灿,陈玮,尹钟. 基于弱监督宫颈细胞图像的语义分割方法[J]. 电子科技, 2021, 34(12): 68-74.
[12]	张莹,刘子龙,万伟. 基于Faster R-CNN的无人机车辆目标检测[J]. 电子科技, 2021, 34(11): 11-20.
[13]	李诚,刘昊,蒋希峰,吴军法,韩文刚,高建国. 基于VGG网络的发电机定转子智能诊断算法[J]. 电子科技, 2021, 34(11): 62-66.
[14]	董波,周燕,王永雄. 基于渐进结构感受野和全局注意力的显著性检测[J]. 电子科技, 2021, 34(1): 23-30.
[15]	张翔,史志才,陈良. 基于SWA优化级联网络的表情识别方法[J]. 电子科技, 2020, 33(9): 16-20.