Deep cross residual network for HEp-2 cell staining pattern classification

作者：

Highlights：

• A cross connection based residual block to enhance information flow in CNN was proposed.

• A deep cross residual (DCR) network model for HEp-2 cell classification was designed.

• The proposed DCR network achieved the best result on the ICPR 2012 dataset, was winner of the most recent ICPR 2016 task 1 contest and the accuracy is higher than all of the top performers in the ICIP 2013 and the ICPR 2014 contests.

摘要

•A cross connection based residual block to enhance information flow in CNN was proposed.•A deep cross residual (DCR) network model for HEp-2 cell classification was designed.•The proposed DCR network achieved the best result on the ICPR 2012 dataset, was winner of the most recent ICPR 2016 task 1 contest and the accuracy is higher than all of the top performers in the ICIP 2013 and the ICPR 2014 contests.

论文关键词：Convolutional neural network,Cross connection,Deep cross residual network,HEp-2 classification

论文评审过程：Received 12 June 2017, Revised 20 March 2018, Accepted 5 May 2018, Available online 7 May 2018, Version of Record 15 June 2018.

论文官网地址：https://doi.org/10.1016/j.patcog.2018.05.005