GMSDenseNet:基于组多结构卷积的轻量级DenseNet

doi:10.12068/j.issn.1005-3026.2020.01.003

东北大学学报:自然科学版 ›› 2020, Vol. 41 ›› Issue (1): 12-16.DOI: 10.12068/j.issn.1005-3026.2020.01.003

GMSDenseNet:基于组多结构卷积的轻量级DenseNet

于长永，何鑫，祁欣，马海涛

(东北大学秦皇岛分校计算机与通信工程学院，河北秦皇岛066004)

收稿日期:2019-03-10 修回日期:2019-03-10 出版日期:2020-01-15 发布日期:2020-02-01
通讯作者: 于长永
作者简介:于长永(1981-)，男，辽宁海城人，东北大学副教授.
基金资助:
国家自然科学基金资助项目(61772124，61332014，61401080，61402087)；河北省自然科学基金资助项目(F2015501049)；河北省教育厅项目(QN2014339)；中央高校基本科研业务费专项资金资助项目(N150402002).

GMSDenseNet:Lightweight DenseNet Based on Group Multi-structure Convolution

YU Chang-yong， HE Xin， QI Xin， MA Hai-tao

School of Computer and Communication Engineering， Northeastern University at Qinhuangdao， Qinhuangdao 066004， China.

Received:2019-03-10 Revised:2019-03-10 Online:2020-01-15 Published:2020-02-01
Contact: HE Xin
About author:-
Supported by:
-

摘要/Abstract

摘要： 提出了一种简单且高效的轻量级DenseNet模型，优化了原DenseNet模型中存在的参数冗余以及高额浮点运算数(FLOPs)的问题.首先，分析了现有流行的卷积计算单元的细节以及特性，其次，应用具有良好特性的卷积单元组合来设计组多结构卷积单元，构建轻量级DenseNet模型，进一步分析了该模型与原DenseNet模型的复杂度.通过实验结果给出所构建网络结构的最优配置，并得到使用DenseNet-40模型约18.8%的FLOPs以及28.4%的模型参数的情况下，准确率仅下降≤0.4%的结果.

关键词: 组卷积, 深度可分离卷积, 组多结构卷积单元, 轻量级, DenseNet

Abstract: A simple and efficient lightweight DenseNet model was proposed， which optimized the parameter redundancy and high FLOPs(floating point operations) in the original DenseNet model. Firstly， the details and characteristics of existing popular convolutional computing units were analyzed. Secondly， the convolutional unit combination with good characteristics was applied to design the multi-structure convolution unit which was used to construct the lightweight DenseNet model. The complexity of this lightweight model and the original DenseNet model was further analyzed. The optimal configuration of the constructed network structure was given according to the experimental results. The results of using the DenseNet-40 model with about 18.8% of FLOPs and 28.4% of the model parameters are obtained， and the accuracy is only reduced by 0.4% at most.

Key words: group convolution, depth separable convolution, group multi-structure convolution unit, lightweight, DenseNet

中图分类号:

TP183

于长永，何鑫，祁欣，马海涛. GMSDenseNet:基于组多结构卷积的轻量级DenseNet[J]. 东北大学学报:自然科学版, 2020, 41(1): 12-16.

YU Chang-yong， HE Xin， QI Xin， MA Hai-tao. GMSDenseNet:Lightweight DenseNet Based on Group Multi-structure Convolution[J]. Journal of Northeastern University Natural Science, 2020, 41(1): 12-16.

参考文献

[1]Zoph B，Vasudevan V，Shlens J，et al.Learning transferable architectures for scalable image recognition［C］//Conference on Computer Vision and Pattern Recognition.Salt Lake City，2018:8697-8710.
[2]Cai Z，Vasconcelos N.Cascade R-CNN:delving into high quality object detection［C］//Conference on Computer Vision and Pattern Recognition.Salt Lake City，2018:6154-6162.
[3]Ke T W，Hwang J J，Liu Z，et al.Adaptive affinity fields for semantic segmentation［J］.Lecture Notes in Computer Science，2018，112(5):605-621.
[4]Krizhevsky A，Sutskever I，Hinton G.ImageNet classification with deep convolutional neural networks［J］.Communication Association for Computing Machinery，2017，60(6):84-90.
[5]Szegedy C，Liu W，Jia Y，et al.Going deeper with convolutions［C］//Conference on Computer Vision and Pattern Recognition.Boston，2015:1-9.
[6]Crowley E J，Turner J，Storkey A，et al.Pruning neural networks:is it time to nip it in the bud?［C］//Workshop on Neural Information Processing Systems.Montreal，2018:1-5.
[7]He K，Zhang X，Ren S，et al.Deep residual learning for image recognition［C］//Conference on Computer Vision and Pattern Recognition.Las Vegas，2016:770-778.
[8]Huang G，Liu Z，Maaten L V D，et al.Densely connected convolutional networks［C］//Conference on Computer Vision and Pattern Recognition.Puerto Rico，2017:2261-2269.
[9]Liu Z，Li J，Shen Z，et al.Learning efficient convolutional networks through network slimming［C］//IEEE International Conference on Computer Vision.Venice，2017:2755-2763.
[10]Zhang X，Zhou X，Lin M，et al.Shufflenet:an extremely efficient convolutional neural network for mobile devices［C］//Conference on Computer Vision and Pattern Recognition.Salt Lake City，2018:6848-6856.
[11]Wang W，Li X，Yang J，et al.Mixed link networks［C］//Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence.Stockholm，2018:2819-2825.
[12]Huang G，Liu S，Laurens V D M，et al.CondenseNet:an efficient denseNet using learned group convolutions［C］//Conference on Computer Vision and Pattern Recognition.Salt Lake City，2018:2752-2761.
[13]Krizhevsky A.Learning multiple layers of features from tiny images［D］.Ontario:University of Toronto，2009.
[14]Zhang T，Qi G J，Xiao B，et al.Interleaved group convolutions for deep neural networks［EB/OL］.(2017-07-18)［2019-03-10］.https://arXiv.org/pdf/1707.02725.
[15]Larsson G，Maire M，Shakhnarovich G.Fractalnet:ultra-deep neural networks without residuals［C］//Conference on International Conference on Learning Representations.San Juan，2016:1-11.
[15]关守平，房少纯.一种新型的区间-粒子群优化算法［J］.东北大学学报(自然科学版)，2012，33(10):1381-1384.(Guan Shou-ping，Fang Shao-chun.A new particle swarm optimization algorithm［J］.Journal of Northeastern University(Natural Science)，2012，33(10):1381-1384.)

GMSDenseNet:基于组多结构卷积的轻量级DenseNet

GMSDenseNet:Lightweight DenseNet Based on Group Multi-structure Convolution

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 3

编辑推荐

Metrics

本文评价

[1]	沈志东，林晨，佟强. 面向移动云计算的轻量级数据完整性验证方法[J]. 东北大学学报:自然科学版, 2015, 36(11): 1562-1566.
[2]	曲大鹏;王兴伟;黄敏;. WPANT:应用于移动对等网络的轻量级层次蚁群路由算法[J]. 东北大学学报(自然科学版), 2011, 32(3): 356-359.
[3]	罗玎玎;赵海;孙佩刚;张希元;. 一个具有能量管理机制WSNs的操作系统内核[J]. 东北大学学报(自然科学版), 2007, 28(11): 1552-1555.