Superpixel-based 3D deep neural networks for hyperspectral image classification

作者：

Highlights：

• Superpixel is first introduced into DNNs for HSI classification.

• Spatial feature image is constructed to increase the spectral-spatial similarity.

• 3D superpixel-based samples are designed to improve boundary misclassification.

• 3D RCNNs are constructed to further improve spatial continuity and suppress noise.

摘要

•Superpixel is first introduced into DNNs for HSI classification.•Spatial feature image is constructed to increase the spectral-spatial similarity.•3D superpixel-based samples are designed to improve boundary misclassification.•3D RCNNs are constructed to further improve spatial continuity and suppress noise.

论文关键词：Hyperspectral image classification,Superpixel,3D deep neural networks

论文评审过程：Received 4 January 2017, Revised 29 May 2017, Accepted 5 September 2017, Available online 13 September 2017, Version of Record 28 October 2017.

论文官网地址：https://doi.org/10.1016/j.patcog.2017.09.007