Deep Kernel Extreme-Learning Machine for the Spectral–Spatial Classification of Hyperspectral Imagery

Extreme-learning machines (ELM) have attracted significant attention in hyperspectral image classification due to their extremely fast and simple training structure. However, their shallow architecture may not be capable of further improving classification accuracy. Recently, deep-learning-based alg...

Full description

Saved in:
Bibliographic Details
Published in:Remote sensing (Basel, Switzerland) Switzerland), 2018-12, Vol.10 (12), p.2036
Main Authors: Li, Jiaojiao, Xi, Bobo, Du, Qian, Song, Rui, Li, Yunsong, Ren, Guangbo
Format: Article
Language:English
Subjects:
Accuracy
Algorithms
Artificial intelligence
Artificial neural networks
Classification
deep layer
Feature extraction
hyperspectral classification
Hyperspectral imaging
Image classification
kernel-based ELM
Kernels
Learning algorithms
Machine learning
Neural networks
Random variables
Remote sensing
spectral and spatial features
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Extreme-learning machines (ELM) have attracted significant attention in hyperspectral image classification due to their extremely fast and simple training structure. However, their shallow architecture may not be capable of further improving classification accuracy. Recently, deep-learning-based algorithms have focused on deep feature extraction. In this paper, a deep neural network-based kernel extreme-learning machine (KELM) is proposed. Furthermore, an excellent spatial guided filter with first-principal component (GFFPC) is also proposed for spatial feature enhancement. Consequently, a new classification framework derived from the deep KELM network and GFFPC is presented to generate deep spectral and spatial features. Experimental results demonstrate that the proposed framework outperforms some state-of-the-art algorithms with very low cost, which can be used for real-time processes.
ISSN:2072-4292
2072-4292
DOI:10.3390/rs10122036