摘要: |
通过改进动态路由和压缩函数的方式对Hinton等的胶囊网络模型进行改进。运用了加强数据集的方法,增加了数据集的大小,在一定程度上避免了过拟合现象的发生。通过实验表明,改进后的胶囊网络模型在结构上有了简化,在效率上比未改进的模型有了明显的提高。在改进的胶囊网络模型基础上,提出了将改进后的胶囊网络与卷积神经网络相结合的网络模型。该模型训练准确率达到97.56%,模型评估准确率达88%。 |
关键词: 胶囊网络 动态路由 特征向量 压缩函数 卷积神经网络 |
DOI: |
|
基金项目:福建省交通运输厅科技项目(201409);福建省职业院校智能装备应用技术协同创新中心建设项目(闽教科[2016]7号) |
|
An object recognition model combining capsule network and convolutional neural network |
LIN Shaodan,HONG Chaoqun,CHEN Yuxue |
(1a.Department of Information Engineering,Fujian Chuanzheng Communications College,Fuzhou 350007,China;2.School of Computer and Information Engineering,Xiamen University of Technology,Xiamen 361024,China;1b.Library,Fujian Chuanzheng Communications College,Fuzhou 350007,China) |
Abstract: |
The capsule network model of Hinton et al is improved by improving dynamic routing and squashing function.The method of strengthening data sets is used to increase the size of data sets and avoid the occurrence of over-fitting to a certain extent.Experiments show that the improved capsule network model is simplified in structure,its efficiency is obviously improved compared with the unmodified model.Based on the improved capsule network model,a network model combining the improved capsule network and convolution neural network is proposed.The training accuracy and evaluation accuracy of the model reaches 97.56% and 88%,respectively. |
Key words: capsule network dynamic routing feature vectors squashing function convolutional neural network |