A Hybrid 2D/3D Convolutional Neural Network for Hyperspectral Image Classification

DOI

10.2991/assehr.k.200401.057How to use a DOI?

Keywords

hyperspectral image classification, deep learning, 3D convolutional neural network

Abstract

Hyperspectral image classification is an important and yet challenging task. With the success of deep learning, the 2D or 3D convolutional neural network-based approaches have been proposed to capture either the spectral, or the spatial data embedded in hyperspectral images. However, existing approaches fail to model the spectral-spatial data simultaneously. To cope with this issue, we proposed this novel hybrid Convolutional Neural Network (H-CNN) model which contains a module of 2D/3D CNNs, and a data interaction module to fuse the spectral- spatial data. Rigorous experimental evaluations have been performed on one benchmark dataset. Our experimental results demonstrate that the H-CNN is superior to the state-of-the-art 2D or 3D CNN models in hyperspectral image classification with respect to three widely adopted evaluation criteria, i.e., average accuracy, F1 score and Kappa coefficient.

Copyright

© 2020, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Xiaofei Yang
AU  - Xiaofeng Zhang
AU  - Shaokai Wang
AU  - Weihuang Yang
PY  - 2020
DA  - 2020/04/06
TI  - A Hybrid 2D/3D Convolutional Neural Network for Hyperspectral Image Classification
BT  - Proceedings of the International Conference on Education, Economics and Information Management (ICEEIM 2019)
PB  - Atlantis Press
SP  - 265
EP  - 269
SN  - 2352-5398
UR  - https://doi.org/10.2991/assehr.k.200401.057
DO  - 10.2991/assehr.k.200401.057
ID  - Yang2020
ER  -