您的位置:首页  > 论文页面

面向移动设备的轻量级卷积神经网络研究

发表时间:2021-06-19  浏览量:203  下载量:31
全部作者: 吴铖,李剑
作者单位: 北京邮电大学智能科学与技术中心
摘 要: 在移动设备环境下,基于传统卷积的深度神经网络存在参数量和计算量大的问题。针对这一问题,提出一种新型的轻量级模块,基于复杂网络特征图存在冗余这一思想,先使用分组卷积压缩输入通道,再使用传统卷积融合信息,保证组间信息有效交流的同时简化了网络。基于CIFAR-10数据集对resnet-18网络进行了实验,改进后的网络准确率只损失了0.9%,但是参数量缩减至原网络的1/14,计算量缩减至原网络的1/13. 实验结果表明,该模块能以即插即用的方式代替传统卷积嵌入原深度神经网络中,在保持网络能力的前提下极大地减少网络的参数量和计算量。
关 键 词: 人工智能;卷积神经网络;轻量级网络;分组卷积
Title: Research on lightweight convolutional neural network for mobile devices
Author: WU Cheng, LI Jian
Organization: Intelligence Science and Technology Center, Beijing University of Posts and Telecommunications
Abstract: In the mobile environment, the deep neural network based on traditional convolution has the problem of large amounts of parameters and calculation. In order to solve this problem, a new lightweight module is proposed. Based on the idea that the feature maps of complex network are redundant, firstly, group convolution was used to compress the input channels and then the traditional convolution was used to fuse information, which simplified the network while ensuring the effective communication between groups. Experiments were conducted on the resnet-18 network based on the CIFAR-10 dataset. The improved network’s accuracy only lost 0.9%, but the amount of parameters is reduced to 1/14 and the amount of calculation is reduced to 1/13 of the original network. Experimental results show that this module supports to replace traditional convolution in the original deep neural network as a plug-and-play way, which greatly reduces the amount of network’s parameters and calculation while maintaining the network accuracy.
Key words: artificial intelligence; convolutional neural network; lightweight network; group convolution
发表期数: 2021年6月第2期
引用格式: 吴铖,李剑. 面向移动设备的轻量级卷积神经网络研究[J]. 中国科技论文在线精品论文,2021,14(2):202-207.
 
1 评论数 0
暂无评论
友情链接