基于GhostNet的改进模型轻量化方法
Improved model lightweighting method based on GhostNet
投稿时间:2022-01-16  修订日期:2022-01-16
DOI:
中文关键词: 目标检测  GhostNet  残差网络  轻量化部署
英文关键词: Object Detection  GhostNet  Residual Networks  Lightweight Deployment
基金项目:湖北省科技重大专项(2020AEA011);武汉市科技计划应用基础前沿项目(2020020601012267)
作者单位邮编
宋中山 中南民族大学 
周珊 中南民族大学 
艾勇 中南民族大学 430074
郑禄 中南民族大学 
肖博文 中南民族大学 
摘要点击次数: 220
全文下载次数: 0
中文摘要:
      为了降低深度卷积神经网络的部署成本,优化模型的检测性能,提出一种改进的轻量化结构S-GhostNet。该模块通过引入特征图生成优化的Ghost Module结构降低卷积操作的计算参数,并结合改进类残差模块提升模型的精准度。S-GhostNet具有较强的即插即用性,可以应用于多数卷积神经网络模型。实验结果表明,在目标分类以及目标检测任务中,S-GhostNet相较于MobileNet、ShuffleNet以及GhostNet,模型计算量更小,模型的精准度相似,甚至更高。
英文摘要:
      In order to reduce the deployment cost of deep convolutional neural networks and optimize the detection performance of the model, an improved lightweight structure S-GhostNet is proposed. This module reduces the computational parameters of the convolutional operation by introducing a Ghost Module structure optimized for feature map generation and improves the accuracy of the model by combining with an improved class of residual modules. S-GhostNet is highly plug-and-play and can be applied to most convolutional neural network models. Experimental results show that S-GhostNet is less computationally intensive than MobileNet, ShuffleNet and GhostNet in target classification and target detection tasks, and the accuracy of the model is similar or even higher.
View Fulltext   查看/发表评论  下载PDF阅读器
关闭