Deep Residual Networks for Plankton Classification

被引:26
作者
Li, Xiu [1 ,2 ]
Cui, Zuoying [1 ,2 ]
机构
[1] Tsinghua Univ, Dept Automat, Beijing 100084, Peoples R China
[2] Tsinghua Univ, Grad Sch Shenzhen, Shenzhen 518055, Peoples R China
来源
OCEANS 2016 MTS/IEEE MONTEREY | 2016年
关键词
plankton classification; residual networks; Deep learning;
D O I
10.1109/OCEANS.2016.7761223
中图分类号
P75 [海洋工程];
学科分类号
0814 ; 081505 ; 0824 ; 082401 ;
摘要
In this paper, we introduce a deep residual network to classify images of plankton. The Plankton Dataset, which consists of 30,336 plankton images of 121 classes, was used for a data science competition hosted on the Kaggle platform(1). We finally achieved a top-5 accuracy of 95.8% and a nearly real-frame rate of 9.1ftps, which is close to the accuracy of the No. 1 team ( over 98%, 1.4ftps) (2) in the competition but much faster than their classification on a single image.
引用
收藏
页数:4
相关论文
共 16 条
[1]  
[Anonymous], OCEANS
[2]  
[Anonymous], P IEEE INT C COMP VI
[3]  
[Anonymous], 2015, PROC CVPR IEEE, DOI 10.1109/CVPR.2015.7299173
[4]  
[Anonymous], AISTATS
[5]  
[Anonymous], 2013, ARXIV13126120
[6]  
[Anonymous], 2015, arXiv: Learning
[7]  
[Anonymous], 2013, IEEE Comput. Soc.
[8]  
[Anonymous], 2014, NIPS
[9]  
[Anonymous], 1989, NEURAL COMPUTATION
[10]  
[Anonymous], 2015, Highway networks