1520 IEEE SIGNAL PROCESSING LETTERS, VOL. 25, NO. 10, OCTOBER 2018
Hyperspectral Image Classification via Superpixel
Spectral Metrics Representation
Bing Tu , Member, IEEE, Wenlan Kuang, Student Member, IEEE, Guangzhe Zhao , Member, IEEE,
and Hongyan Fei
, Member, IEEE
Abstract—This letter proposes a new hyperspectral classifica-
tion method that fuses superpixel spectral metrics and joint sparse
representation (JSR), which is termed as superpixel spectral met-
rics representation (SSMR). Recently, superpixel segmentation has
proven to be a powerful tool to exploit the spatial information
of hyperspectral images (HSIs), since the size and shape of each
superpixel can be adaptively changed in different structural tex-
tures. Moreover, spectral information divergence (SID) has su-
periority compared to other distance-based similarity measures,
particularly when using with a JSR classifier. Taking the aforeme-
mentioned advantages into account, superpixel segmentation, SID,
and JSR are availably combined to effectively utilize the spectral-
spatial information of the HSI. The proposed SSMR method in-
cludes the following main steps. First, superpixel segmentation is
utilized to divide the original map into several superpixels. Second,
similarity metric SID among test samples in all superpixels and
training samples are calculated. Next, the JSR model is employed
to obtain the reconstruction residuals of each class. Then, a reg-
ularization parameter λ is introduced to attain balance between
JSR and SID. Finally, pixel’s label is determined by the minimal
total residual. Experimental results on the Indian Pines dataset
show better performance than several well-known classification
methods.
Index Terms—Hyperspectral image (HSI), joint sparse repre-
sentation (JSR), superpixel segmentation, spectral information
divergence (SID).
I. INTRODUCTION
T
HE classification of hyperspectral images (HSIs) has
played an essential role in many application domains,
such as environment monitoring, urban planning, and preci-
sion agriculture. Various classification techniques [1] including
Manuscript received April 29, 2018; revised May 25, 2018; accepted May
30, 2018. Date of publication August 17, 2018; date of current version August
27, 2018. This work was supported in part by the National Natural Science
Foundation of China under Grant 51704115 and Grant 61462089; in part by
the Key Laboratory Open Fund Project of Hunan Province University under
Grant 17K040; and in part by the Science and Technology Program of Hunan
Province under Grant 2016TP1021. The associate editor coordinating the review
of this manuscript and approving it for publication was Prof. Joao Paulo Papa.
(Corresponding author: Guangzhe Zhao.)
B. Tu, W. Kuang, and H. Fei are with the School of Information Sci-
ence and Engineering, Hunan Institute of Science and Technology, Yueyang
414000, China (e-mail:, tubing@hnist.edu.cn; wenlan_kuang@foxmail.com;
hy_fei@hotmail.com).
G. Zhao is with the College of Electrical and Information Engineering, Bei-
jing University of Civil Engineering and Architecture, Beijing 100044, China
(e-mail:,zhaoguangzhe@bucea.edu.cn).
Color versions of one or more of the figures in this letter are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/LSP.2018.2865687
parametric and nonparametric classifiers have been presented
to efficiently exploit the discriminative information contained
in these spectral channels. The nearest neighbor (NN) classifier
[2], which is one of the simplest yet most effective classifica-
tion methods, is commonly utilized in the HSI processing. The
principle of NN-based classifiers is usually to find a predefined
number (e.g., k) of training samples closest to the test samples
by employing an Euclidean distance and assign the majority
category label based on its k-nearest training samples. Mean-
while, many classical pixel-wise classification methods, such
as, support vector machines (SVMs) [3], neural networks, and
artificial immune networks, have been extensively applied to the
HSI classification and obtained promising performance.
Recently, sparse representation (SR) has demonstrated to be a
powerful image processing tool for addressing problems such as
remote sensing image fusion, target detection, and face recogni-
tion. It depends on the underlying assumption that a test sample
can be linearly represented by a few training samples from the
same class. For example, Lai and Jiang [4] proposed an SR-
based classifier (SRC) for robust face recognition that can pro-
vide superior performance in face recognition problems. How-
ever, the traditional SRC just takes the spectral information into
consideration and ignores spatial neighbors around the test pixel
based on the assumption that pixels within a local region usually
contain similar spectral materials and characteristics. Chen et al.
[5] proposed a joint sparse representation classifier (JSRC) that
makes full use of spatial information where a fixed-size square
window is first used for each test pixel. Although the JSRC
method obtains an improved performance, a homogeneous area
may need a larger window compared to the heterogeneous area.
Fu et al. [6] s ought to address this problem by proposing a new
shape-adaptive joint sparse representation classification method
that could sufficiently utilize the spatial information compared
to the fixed-size window.
Similarly, numerous superpixel segmentation methods
[7]–[9] have been widely used in the image processing field
on the basis of the merit that the size and shape of superpixels
can be adaptively adjusted according to the spatial structures in
a local region. Meanwhile, a superpixel can obtain redundant
information for images, it can also provide convenient origi-
nal images for computing image features and greatly reduce
the complexity of subsequent image processing tasks. In [10],
k-mean clustering is used for classification while not consid-
ering correlations among superpixels. Analogously, the corre-
lation coefficient (CC) [11] is a common similarity index that
1070-9908 © 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications
standards/publications/rights/index.html for more information.