Authors: Yu Liu Baocai Yin Jun Yu Zengfu Wang
Publish Date: 2016/08/01
Volume: 76, Issue: 8, Pages: 11065-11079
Abstract
In the past few years convolutional neural networks CNNs have exhibited great potential in the field of image classification In this paper we present a novel strategy named crosslevel to improve the existing networks’ architecture in which different levels of feature representation in a network are merely connected in series The basic idea of crosslevel is to establish a convolutional layer between two nonadjacent levels aiming to extract more sufficient features with multiple scales at each feature representation level The proposed crosslevel strategy can be naturally integrated into an existing network without any change on its original architecture which makes it very practical and convenient Four popular convolutional networks for image classification are employed to illustrate its implementation in detail Experimental results on the dataset adopted by the ImageNet LargeScale Visual Recognition Challenge ILSVRC verify the effectiveness of the crosslevel strategy on image classification Furthermore a new convolutional network with crosslevel architecture is presented to demonstrate the potential of the proposed strategy in future network designThe authors would like to thank the editors and anonymous reviewers for their constructive comments and valuable suggestions This work was supported by the National Natural Science Foundation of China No 61472393 and No 61303150 the National Science and Technology Major Project of the Ministry of Science and Technology of China No 2012GB102007 and the Anhui Province Initiative Funds on Intelligent Speech Technology and Industrialization No 13Z02008 The authors greatly acknowledge the support of IFLYTEK COLTD
Keywords: