广西师范大学学报(自然科学版) ›› 2019, Vol. 37 ›› Issue (3): 50-59.doi: 10.16088/j.issn.1001-6600.2019.03.006

• • 上一篇    下一篇

多尺度并行融合的轻量级卷积神经网络设计

范瑞,蒋品群*,曾上游,夏海英,廖志贤,李鹏   

  1. 广西师范大学电子工程学院,广西桂林541004
  • 出版日期:2019-07-12 发布日期:2019-07-12
  • 通讯作者: 蒋品群(1970—),男,广西全州人,广西师范大学副教授,博士。E-mail: pqjiang@mailbox.gxnu.edu.cn
  • 基金资助:
    国家自然科学基金(11465004,61762014);桂林市科学研究与技术开发计划项目(20170113-4)

Design of Lightweight Convolution Neural Network Based on Multi-scale Parallel Fusion

FAN Rui, JIANG Pinqun*, ZENG Shangyou, XIA Haiying, LIAO Zhixian, LI Peng   

  1. College of Electronic Engineering, Guangxi Normal University, Guilin Guangxi 541004,China
  • Online:2019-07-12 Published:2019-07-12

摘要: 针对传统深度卷积神经网络分类精度不佳,参数量巨大,难以在内存受限的设备上进行部署的问题,本文提出了一种多尺度并行融合的轻量级卷积神经网络架构PL-Net。首先,将上层输出特征图分别送入两种不同尺度的深度可分离卷积层;然后对并行输出特征信息进行交叉融合,并加入残差学习,设计了一种并行轻量型模块PL-Module;同时,为了更好地提取特征信息,利用尺度降维卷积模块SR-Module来替换传统池化层;最后将上述两个模块相互堆叠构建轻量级网络。在CIFAR10、Caltech256和101_food数据集上进行训练与测试,结果表明:与同等规模的传统CNN、MobileNet-V2网络及SqueezeNet网络相比,PL-Net在减少网络参数的同时,提升了网络的分类精度,适合在内存受限的设备上进行部署。

关键词: 卷积神经网络, 深度可分离卷积, 残差学习, 并行卷积

Abstract: Aiming at the problem that the traditional deep convolutional neural network has poor classification accuracy and large amount of parameters, which is difficult to deploy in memory-constrained devices, a multi-scale parallel fusion lightweight convolutional neural network architecture PL-Net is proposed. Firstly, the upper output feature map is sent to two different scales of the depth separable convolution layer, and then the parallel output is cross-fused with the feature information, and with the residual learning, a parallel lightweight module PL-Module is designed. To better extract the feature information, the scale-dimensional reduction convolutional module(SR-Module) is proposed to replace the traditional pooling layer. Finally, the above two modules are stacked on each other to construct a lightweight network. In the experimental phase, training and testing are performed on the CIFAR10,Caltech 256 and 101_food data sets. The results show that compared with the traditional CNN,MobileNet-V2 and Squeezenet networks of the same scale, PL-Net improves the classification accuracy of the network while reducing the amount of network parameters, and is suitable for deployment on memory-constrained devices.

Key words: convolutional neural network, depthwise separable convolutions, residual learning, parallel convolution

中图分类号: 

  • TP183
[1] 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.
[2] XIE Saining,GIRSHICK R,DOLLAR P,et al.Aggregated residual transformations for deep neural networks [C]//2017 IEEE Conference on Computer Vision and Pattern Recognition.Los Alamitos,CA:IEEE Computer Society,2017:5987-5995.DOI:10.1109/CVPR.2017.634.
[3] KRIZHEVSKY A,SUTSKEVER I,HINTON G E.ImageNet classification with deep convolutional neural networks[C]//PEREIRA F,BURGES C J C,BOTTOU L,et al.Advances in Neural Information Processing Systems 25.Red Hook,NY:Curran Associates,Inc.,2012:1097-1105.
[4] SIMONYAN K,ZISSERMAN A.Very deep convolutional networks for large-scale image recognition[EB/OL]. (2014-9-4)[2018-11-05].https://arxiv.org/pdf/1409.1556.pdf.
[5] SZEGEDY C,LIU Wei,JIA Yangqing,et al.Going deeper with convolutions[C]//2015 IEEE Conference on Computer Vision and Pattern Recognition.Piscataway,NJ:IEEE Press,2015:1-9.DOI:10.1109/CVPR.2015.7298594.
[6] HE Kaiming,ZHANG Xiangyu,REN Shaoqing,et al.Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition.Piscataway,NJ:IEEE Press,2016:770-778.DOI: 10.1109/CVPR.2016.90.
[7] HUANG Gao,LIU Zhuang,van der MAATEN L,et al.Densely connected convolutional networks[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition.Piscataway,NJ:IEEE Press,2017:2261-2269.DOI: 10.1109/CVPR.2017.243.
[8] SZEGEDY C,VANHOUCKE V,IOFFE S,et al.Rethinking the inception architecture for computer vision[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition.Piscataway,NJ:IEEE Press,2016: 2818-2826.DOI:10.1109/CVPR.2016.308.
[9] IOFFE S,SZEGEDY C.Batch normalization:Accelerating deep network training by reducing internal covariate shift[C]//Proceedings of the 32nd International Conference on Machine Learning.Brookline,MA: Microtome Publishing,2015:448-456.
[10]SZEGEDY C,IOFFE S,VANHOUCKE V, et al.Inception-v4,inception-resnet and the impact of residual connections on learning[EB/OL].(2016-2-23)[2018-11-05].https://arxiv.org/pdf/1602.07261.pdf.
[11]IANDOLA F N,HAN Song,MOSKEWICZ M W,et al.SqueezeNet:AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size[EB/OL].(2016-2-24)[2018-11-05].https://arxiv.org/pdf/1602.07360.pdf.
[12]HOWARD A G,ZHU Menglong,CHEN Bo,et al.MobileNets:Efficient convolutional neural networks for mobile vision applications[EB/OL].(2017-4-17)[2018-11-05].https://arxiv.org/pdf/1704.04861.pdf.
[13]SANDLER M,HOWARD A,ZHU Menglong,et al.MobileNetV2:Inverted residuals and linear bottlenecks[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Piscataway,NJ:IEEE Press,2018: 4510-4520.DOI:10.1109/CVPR.2018.00474.
[14]ZHANG Xiangyu,ZHOU Xinyu,LIN Mengxiao,et al.ShuffleNet:An extremely efficient convolutional neural network for mobile devices[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway,NJ:IEEE Press,2018:6848-6856.DOI:10.1109/CVPR.2018.00716.
[15]MARTINEL N,PICIARELLI C,MICHELONI C.A supervised extreme learning committee for food recognition[J]. Computer Vision and Image Understanding,2016,148:67-86.DOI:10.1016/j.cviu.2016.01.012.
[16]HERAVI E J,AGHDAM H H,PUIG D.An optimized convolutional neural network with bottleneck and spatial pyramid pooling layers for classification of foods[J].Pattern Recognition Letters,2018,105:50-58.DOI: 10.1016/j.patrec.2017.12.007.
[17]ZEILER M D,FERGUS R.Stochastic pooling for regularization of deep convolutional neural networks[EB/OL]. (2013-1-16)[2018-11-05].https://arxiv.org/pdf/1301.3557.pdf.
[18]GOODFELLOW I J,WARDE-FARLEY D,MIRZA M,et al.Maxout networks[EB/OL].(2013-2-18)[2018-11-05].https:// arxiv.org/pdf/1302.4389.pdf.
[1] 严浩, 许洪波, 沈英汉, 程学旗. 开放式中文事件检测研究[J]. 广西师范大学学报(自然科学版), 2020, 38(2): 64-71.
[2] 武文雅, 陈钰枫, 徐金安, 张玉洁. 基于高层语义注意力机制的中文实体关系抽取[J]. 广西师范大学学报(自然科学版), 2019, 37(1): 32-41.
[3] 薛洋,曾庆科,夏海英,王文涛. 基于卷积神经网络超分辨率重建的遥感图像融合[J]. 广西师范大学学报(自然科学版), 2018, 36(2): 33-41.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
[1] 李钰慧, 陈泽柠, 黄中豪, 周岐海. 广西弄岗熊猴的雨季活动时间分配[J]. 广西师范大学学报(自然科学版), 2018, 36(3): 80 -86 .
[2] 覃盈盈, 漆光超, 梁士楚. 凤眼莲组织浸提液对靖西海菜花种子萌发的影响[J]. 广西师范大学学报(自然科学版), 2018, 36(3): 87 -92 .
[3] 韦宏金, 周喜乐, 金冬梅, 严岳鸿. 湖南蕨类植物增补[J]. 广西师范大学学报(自然科学版), 2018, 36(3): 101 -106 .
版权所有 © 广西师范大学学报(自然科学版)编辑部
地址:广西桂林市三里店育才路15号 邮编:541004
电话:0773-5857325 E-mail: gxsdzkb@mailbox.gxnu.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发