IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 14, NO. 11, NOVEMBER 2017 1933
Independent Encoding Joint Sparse Representation
and Multitask Learning for Hyperspectral
Target Detection
Yuxiang Zhang, Wu Ke, Member, IEEE,BoDu,Senior Member, IEEE, and Xiangyun Hu
Abstract— Target detection is playing an important role in
hyperspectral image (HSI) processing. Many traditional detection
methods utilize the discriminative information within all the
single-band images to distinguish the target and the background.
The critical challenge with these methods is simultaneously
reducing spectral redundancy and preserving the discriminative
information. The multitask learning (MTL) technique has the
potential to solve the aforementioned challenge, since it can fur-
ther explore the inherent spectral similarity between the adjacent
single-band images. This letter proposes an independent encoding
joint sparse representation and an MTL method. This approach
has the following capabilities: 1) explores the inherent spectral
similarity to construct multiple sub-HSIs in order to reduce
spectral redundancy for each sub-HSI; 2) takes full advantage
of the prior class label information to construct reasonable joint
sparse repr esentation and MTL models for the target and the
background; 3) explores the great difference between the target
dictionary and background dictionary with different regulariza-
tion strategies in order to better encode the task relatedness for
two joint sparse representation and MTL models; and 4) makes
the detection decision by comparing the reconstruction residuals
under different prior class labels. Experiments on two HSIs
illustrated the effectiveness of the proposed method.
Index Terms— Hyperspectral image (HSI), multitask learn-
ing (MTL), sparse representation, target detection.
I. INTRODUCTION
H
YPERSPECTRAL imagery can provide a unique
advantage for target detection, since it can distinguish
subtle spectral differences with the characteristic of the high
spectral resolution [1], [2]. Target detection aims to separate
specific targ et pixels from various backgrounds with prior
knowledge of the targets [3 ], [4], which is essentially a binar y
classification problem [5].
The current target detection methods mainly utilize the
detailed sp ectral informatio n within all single-band images of
Manuscript received April 15, 2017; revised July 4, 2017; accepted
August 9, 2017. Date of publication September 21, 2017; date of current
version October 25, 2017. This w ork was supported in part by the National
Natural Science Foundation of China under Grant U1536204, Grant 61701452,
Grant 61471274, Grant 41630317, Grant 61372153, and Grant 61601522, in
part by the China Postdoctoral Science Foundation under Grant 2017M612533
and Grant 2015M580753, and in part by the Fundamental Research
Funds for the Central Uni versities, China Univ ersity of Geosciences
under Grant CUG170617, Grant CUGL140410, and Grant 26420160125.
(Corresponding author: Bo Du.)
Y. Zhang, W. Ke, and X. Hu are with the Hubei Subsurface Multi-scale
Imaging Key Laboratory, Institute of Geophysics and Geomatics, China
Uni versity of Geosciences, Wuhan 430074, China (e-mail: zyx_070504@
163.com).
B. Du is with the School of Computer, Wuhan University, Wuhan 430079,
China (e-mail: gunspace@163.com).
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/LGRS.2017.2739821
the hyperspectral image (HSI) to distinguish the target from
the background. According to the different techniques of u ti-
lizing spectral information for the target and the background,
a number of target detection methods have been proposed in
recent decades, such as the statistical hypothesis testing theory-
based detectors [6]–[8], filtering or projection technique-based
detectors [9]–[11], and sparse representation technique-based
detectors [12]–[15]. It can be found that these existing target
detection methods fully and uniformly utilize the discrimina-
tive information within all single-band im ages. However, with
the high spectral resolution of HSI, the adjacent single-band
images present a great spectral similarity or redundancy, and
this spectral redundancy can affect the detection performance.
Many dimension reduction methods have been, therefore,
proposed to reduce spectral redundancy for target detec-
tion [16]–[20]. Nevertheless, these methods can hardly guar-
antee that all the valuable discriminative spectral inf ormation
underlying the HSI data will be reserved with the dimension
reduction process. In brief, there exists a critical challenge to
simultaneously reduce the spectral redundancy and reserve the
discriminative information for Hyperspectral target detection.
Recently, researchers h ave tried to address the above-
mentioned challenge for hyperspectral target detection. Based
on the above-mentioned analysis, the spectral similarity char-
acteristic within the adjacent single-band images is consistent
with the multitask learnin g (MTL) technique. The MTL is
an approach that learns a problem together with other related
problems at the same time [21]–[23], which often leads to
a better model for each task, as the commonality among
the tasks is further used. Thus, MTL can work particularly
well if the tasks have some commonality. The MTL is
employed for hyperspectral target detection, labeled as the
joint sparse representation and MTL (JSR-MTL) detector
in [24]. It explores the spectral similarity between the adjacent
single-band images to construct multiple sub-HSIs, leading to
multiple related detection tasks. In this way, the redundancy
in each detection task can be effectively avoided. It further
explores the spectral similarity to analyze the latent sparse
representation of each task, and multiple sparse representation
models via the union target and background dictionary are
then integrated via a unified MTL technique [24]. Therefore,
the spectral information behind the high-dimension original
HSI data set is fully used without the loss of discriminative
information [24].
This letter proposes a novel independent encoding JSR-
MTL (IEJSR-MTL) method. It explores the p rior class label
information of the training samples to construct two reasonable
joint sparse representation and MTL models, where the test
pixel (target or background) is separately modeled via the
1545-598X © 2017 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.