基于并行和切片的深度卷积网络设计研究

杨远飞; 曾上游; 冯燕燕; 周悦; 潘兵

Chinese

您当前的位置：

首页 >

文章列表页 >

基于并行和切片的深度卷积网络设计研究

更新时间：2026-09-04

- 基于并行和切片的深度卷积网络设计研究
- Design and Research of Deep Convolution Network Based on Paralleland Slice
- 微电子学与计算机 2018年35卷第3期页码：120-124
- 作者机构：
  
  广西师范大学机构电子工程学院,广西,桂林,541004
- 作者简介：
- 基金信息：
  
  国家自然科学基金 114650042017广西研究生教育创新计划项目 XYCSZ2017054
- DOI：
  中图分类号：
- 纸质出版：2018
- 稿件说明：
移动端阅览
杨远飞, 曾上游, 冯燕燕, 周悦, 潘兵. 基于并行和切片的深度卷积网络设计研究[J]. 微电子学与计算机, 2018, 35(3): 120-124.

YANG Yuan-fei, ZENG Shang-you, FENG Yan-yan, et al. Design and Research of Deep Convolution Network Based on Paralleland Slice[J]. 2018, 35(3): 120-124.
杨远飞, 曾上游, 冯燕燕, 周悦, 潘兵. 基于并行和切片的深度卷积网络设计研究[J]. 微电子学与计算机, 2018, 35(3): 120-124. DOI：

YANG Yuan-fei, ZENG Shang-you, FENG Yan-yan, et al. Design and Research of Deep Convolution Network Based on Paralleland Slice[J]. 2018, 35(3): 120-124. DOI：

摘要

提出了基于并行和切片的卷积神经网络模型

并行网络通过两条通道分别对上一层特征图进行特征提取

但是同时网络参数也会倍增; 切片网络是把上层输出特征图先切成两份

再分别对每份特征图进行卷积操作

最后再把两部分的卷积输出进行融合.和同规模的模型相比

这两种模型能够提取更为本质的图像特征

但是切片网络不会增加网络的参数.在实验阶段用这两种网络在cifar10和cifar100数据集上进行测试

取得了良好的效果.

Abstract

We have proposed the convolution neural network model based on parallel and slice

network of parallel extract feature of the top feature map is carried out through two channels

however

the parameters of the network will be doubled; network of slice cut the top output maps into two parts at first

convolute each map

and then fuse the convolution output of the two parts finally.Compared with the same scale model

those two models are able to extract more essential image features

but network of slice will not be added the network parameters.Those networks are tested on the cifar10 and cifar100 datasets in the experimental stage

and good results are obtained.

关键词

Keywords

references

Chen H, Qi X, Yu L, et al. DCAN: Deep contour-aware networks for object instance segmentation from histology images. [J]. Medical Image Analysis, 2017, 36: 135-146. DOI: 135

Jia Y, Shelhamer E, Donahue J, et al. Caffe: Convolutional architecture for fast feature embedding[C]//ACM International Conference on Multimedia. ACM, 2014: 675-678.

Krizhevsky A, Sutskever I, Hinton G E. ImageNet classification with deep convolutional neural networks[C]//International Conference on Neural Information Processing Systems. Curran Associates Inc, 2012: 1097-1105.

Hochreiter S. The vanishing gradient problem during learning recurrent neural nets and problem solutions[M]. World Scientific Publishing Co. Inc, 1998.

Lin M, Chen Q, Yan S. Network In Network[C]//ICLR. IEEE, 2014.

Hinton G E, Salakhutdinov R R. Reducing the dimensionality of data with neural networks[J]. Science, 2006, 313(5786): 504. DOI: 504

Glorot X, Bengio Y. Understanding the difficulty of training deep feedforward neural networks[J]. Journal of Machine Learning Research, 2010, 9: 249-256. DOI: 249

He K, Zhang X, Ren S, et al. Deep Residual Learning for Image Recognition[C]//IEEE Conference on Computer Vision and Pattern Recognition. IEEE Computer Society, 2016: 770-778.

Ioffe S, Szegedy C. Batch normalization: Accelerating deep network training by reducing internal covariate shift[J]. Computer Science, 2015.

Hinton G E, Srivastava N, Krizhevsky A, et al. Improving neural networks by preventing co-adaptation of feature detectors[J]. Computer Science, 2012, 3(4): 212-223. DOI: 212

张立民, 刘凯. 基于深度玻尔兹曼机的文本特征提取研究[J]. 微电子学与计算机, 2015, 32(2): 142-147. DOI: 142

Rumelhart D E, Hinton G E, Williams R J. Learning internal representation by back-propagation of errors[J]. Nature, 1986, 323(323): 533-536. DOI: 533

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

基于MSS-YOLO-MPHands融合架构的装配手部动作识别研究

低功耗设计的双链电源关断策略的实现与验证

AlN/GaN混合缓冲层双异质结p-GaN HEMT直流特性仿真

基于区块链的AI生成内容溯源框架设计研究

超低功耗实时时钟电路的设计