ELM-BASED SPECTRAL–SPATIAL CLASSIFICATION OF HYPERSPECTRAL IMAGES
USING BILATERAL FILTERING INFORMATION ON SPECTRAL BAND-SUBSETS
Yu Shen, Jinhuan Xu, Heng Li, Liang Xiao*
School of Computer Science and Engineering,
Nanjing University of Science and Technology, Nanjing, China
*xiaoliang@mail.njust.edu.cn
ABSTRACT
As single-layer feed-forward neural networks, extreme
learning machine (ELM) has recently been used with success
for the classification of hyperspectral images (HSIs).
However, the results of pure pixel-wise spectral classifiers
often appear very noisy with limited training samples. To
further improve the accuracy, we propose a novel spectral-
spatial information integrating scheme for pixel-wise kernel
ELM-based classifier. In particular, we show that a spatial
bilateral filtering information on spectral band-subsets can
significantly improve the accuracy of the pixel-wise kernel
ELM based classifier. The benefits of the proposed method
are twofold: 1) spectral structural similarity guided band-
subsets partition and 2) incorporating the spectral-spatial
information by bilateral filtering. Experiments on the widely
used real HSI demonstrate that the proposed approach
outperforms several well-known classification methods in
terms of classification accuracy and low computational cost.
Index Terms— hyperspectral image classification,
ELM, band-subset partition, bilateral filtering, SVM
1. INTRODUCTION
As a supervised learning technique, Extreme Learning
Machine (ELM) [1] belongs to a class of feed-forward
neural networks with random weights that has been used
with success for the classification problems. In particular,
ELM randomly generates the hidden node parameters and
analytically determines the output weights instead of
iterative tuning, which makes the learning extremely fast.
ELM is not only computationally efficient but also tends to
achieve similar or even better generalization performance
than SVMs.
Currently, several ELM-based classifiers has been
proposed in the literature. In [2], ELM was used for land
cover classification, which achieved comparable
classification accuracies to a back-propagation neural
network on two datasets considered. KELM was used in [3]
for hyperspectral remote-sensing images classification. The
results indicate that KELM is similar to, or more accurate
than, SVMs in terms of classification accuracy and offers
notably low computational cost. Considering both spectral
and spatial information has been verified to improve the HSI
classification accuracy significantly [4]. One major way is to
directly use pixels in a small neighborhood for joint
classification assuming that these pixels usually share the
same class membership. In [5], three-dimensionality (3-D)
Gabor filters were applied to hyperspectral images to extract
3-D Gabor features. In [6], a preprocessing algorithm named
multi-hypothesis (MH) prediction was successfully used for
hyperspectral image classification. From the
abovementioned methods, we observe that exploiting both
spectral and spatial information is the key issue to improve
the performance of ELM-based classifiers for HSIs.
In this work, a novel spectral-spatial information
enforcing kernel ELM (KELM) based classifier is proposed
for HSI classification. The main novelty of the paper is a
new robust spectral signature creation procedure named
Bilateral Filtering on Spectral Band-Subsets. The procedure
includes two stages. First, spectral-adaptive band-subsets
partition [7] is introduced to group highly correlated spectral
bands and separate low-correlated ones. In each band subset,
the highly correlated spectral bands have continuous and
close spectral characteristics. Second, bilateral filtering [8]
is exploited on each subset to incorporate the spectral
signature in spatial neighborhood. All subsets are merged
again as a new HSI, with each pixel containing more robust
spectral signatures. Finally, a pixel-wise KELM classifier is
applied for the merged HSI data to compute the final
classification results.
2. SUPERVISED CLASSIFICATION VIA KELM
ELM [1] was originally developed from feed-forward neural
networks. Recently, KELM generalizes ELM from explicit
activation function to implicit mapping function, which can
produce better generalization in most applications.
Firstly, for G classes of a HSI, let us define
,
. A row vector
1
[ ,..., ,..., ]
kG
z z zz
indicates the class that a sample belongs to. For example, if
and other elements in
are zero, then the sample
belongs to the kth class. Given
training samples
497978-1-5090-3332-4/16/$31.00 ©2016 IEEE IGARSS 2016