交叉连接的少层残差卷积神经网络

最近的研究表明,卷积神经网络的性能可以通过采用跨层连接来提高,典型的残差网络(Res Net)便通过恒等映射方法取得了非常好的图像识别效果.但是通过理论分析,在残差模块中,跨层连接线的布局并没有达到最优设置,造成信息的冗余和层数的浪费,为了进一步提高卷积神经网络的性能,文章设计了两种新型的网络结构,分别命名为C-FnetO和C-FnetT,它们在残差模块的基础上进行优化并且具有更少的卷积层层数,同时通过在MNIST,CIFAR-10,CIFAR-100和SVHN公开数据集上的一系列对比实验表明,与最先进的卷积神经网络对比,C-FnetO和C-FnetT网络获得了相对更好的图像识别效果,其中C-FnetT网络的性能最佳,在四种数据集上均取得了最高的准确率.

Residual Convolutional Neural Networks with Cross Connections and Fewer Layers

Recent works show that the performance of Convolutional Neural Networks could be improved by applying skip connections, such as ResNet and their variants.However,the layout of the skip connections in the residual module do not achieve optimal setting through theoretical analysis,resulting in the redundancy of information and the waste of the number of layers.In this paper,in order to further improve the performance of Convolutional Neural Networks,two novel convolutional neural network architectures are proposed,which are called C-FnetO and C-FnetT respectively.They are optimized on the basis of residual modules and have fewer convolution layers.Experiential results demonstrate that,compare with other Convolutional Neural Networks,proposed models(C-FnetO and C-FnetT) could achieve better performance on four classical competitive object recognition benchmark tasks(MNIST,CIFAR-10,CIFAR-100 and SVHN),especially for C-FnetT,which has the highest test accuracy.