International Journal of Research in Advanced Technology - IJORAT

Vol. 1, Issue 7, JULY 2016

FIREFLY BASED BAND SELECTION FOR

HYPERSPECTRAL IMAGE
CLASSIFICATION
K. Keerthana1, G.VeeraSenthilKumar2, S. Vasuki3,
1,2,3

Department of Electronics and Communication Engineering, Velammal College of Engineering and Technology,
Madurai, Tamil Nadu, India.

A b s t r a c t : In Remote sensing applications, Hyperspectral Image(HSI) analysis is found more useful, compared
with other imaging techniques, due to high dimensionality. However, high dimensionality of hyperspectral image
leads to increase in computational complexity in their analysis. Hence, dimensionality reduction is an essential
preprocess in all hyperspectral image analysis. Dimensionality reduction can be achieved by either transform
based method or band selection method. To determine the number of bands to be selected Virtual
Dimensionality(VD) is used. This paper presents a novel approach of band selection using firefly algorithm for
hyperspectral image classification by interpreting the band selection as an optimization problem. Firefly algorithm
provides a global optimal solution much more efficiently, compared with other frequently searching strategies.
Support Vector Machine (SVM) is used in classification phase. The performance of our proposed method is
evaluated using accuracy as a measure and compared with PSO based band selection. Experimental results proved
that our proposed method outperforms the PSO based band selection by improving the classification accuracy by
2%.
Keywords: Band Selection, VD, FIREFLY, SVM

I. INTRODUCTION
Multispectral and Hyperspectral sensors have
been widely developed to observe the Earths surface in
remote sensing application ,hyperspectral images has
high spectral resolution compared to multispectral
images, vary by several number of spectral bands,
typically up to several hundreds. This unnecessary large
number of spectral bands implies high dimensionality of
hyperspectral image and causes several challenges to
process the image. Feature selection and feature
extraction are two methods involved for dimensionality
reduction. The original dimension of the hyperspectral
image has to be reduced without any loss of
information. Hence feature extraction method can be
used for band selection. VD provides the reliable
estimate of number of endmembers and hence the
number of bands to be selected [5][3]. Many band
selection technique such as Information Divergence
Band Selection (IDBS)[3], Maximum Ellipsoid Volume
(MEV)[2], Similarity based unsupervised band selection
(SUBS)[10],Affinity
propagation
based
band
selection(APBS)[16],Constrained
Band
Selection
(CBS)[4],Exemplar
Component
Analysis
(ECA)[8],Particle Swarm Optimization[5]are found in
the literature. In our work, FIREFLY is proposed in this
work for band selection because FIREFLY can find a
global optimal solution much more efficient than any
other searching strategies.
The rest of the paper describes about each
section as follows: In section II, Principal Component
Analysis (PCA), as a feature extraction technique, is
All Rights Reserved 2016 IJORAT

explained. In section III, FIREFLY based band selection

algorithm has been explained. In section IV, the work
flow of our proposed methodology is described. In
section V, experimental results are discussed. Finally, in
section VI, conclusion is given.

II.PRINCIPAL
(PCA)

COMPONENT

ANALYSIS

Principal component analysis (PCA) is a

statistical procedure that uses an orthogonal
transformation to convert a set of observations of
possibly correlated variables into a set of values of
linearly uncorrelated variables called principal
components. The number of principal components is
less than or equal to the number of original variables.
This transformation is defined in such a way that the
first principal component has the largest possible
variance and each succeeding component in turn has the
highest variance possible under the constraint that it is
orthogonal to the preceding components. The principal
components are orthogonal because they are the
eigenvectors of the covariance matrix, which is
symmetric. PCA is sensitive to the relative scaling of the
original variables.
PCA finds a smaller-dimensional linear representation
of data vectors such that the original data could be
reconstructed from the compressed representation with
the minimum square error. Assume that we have n d 1
data vectors y1, y2..., yn that are modeled as
yj Wxj + m

(4)
1

International Journal of Research in Advanced Technology - IJORAT

Vol. 1, Issue 7, JULY 2016

where W is a d c matrix, xj are c 1 vectors of

principal components, and m is a d 1 bias vector. We
assume that c d n. Principal subspace methods find
W, xj and m such that the reconstruction error
C=

Xt+1 =xt + +

(9)

- attraction
randomization

=1 ||Yj-Wxj m ||

(5)
3) Determine the light intensity Ii at xi via f(xi)

is minimized. In matrix notation, the data vectors and

principal components can be compiled into d n and c
n matrices Y = [y1 y2 ... yn ] and X = [x1 x2 ... xn ],
and Yi j, Wi k,Xk j denote the elements of the matrices Y,
W and X, respectively. The bias matrix M contains n
copies of the bias vector m as its columns. Principal
subspace methods find W and X such that Y WX + M
and the minimized cost function is the sum of the
squared elements (or Frobenius norm) of matrix
YWXM:
2

C = F

(6)

III. FIREFLY BASED BAND SELECTION

The Firefly algorithm was introduced by Dr. Xin She
yang at Cambridge University which was inspired by
the mating or flashing behavior of fireflies. Although
the algorithm has many similarities with other swarm
based algorithms such as Particle Swarm Optimization
[9], Artificial Bee Colony Optimization [12][5] and Ant
Colony Optimization [15], the Firefly algorithm has
proved to be much simpler both in concept and
implementation. Fireflies or lightning bugs belong to a
family of insects [1] that are capable to produce natural
light to attract a mate or prey. There are about two
thousand firefly species which produce short and
rhythmic flashes. These flashes often appear to be in a
unique pattern and produce an amazing sight in the
tropical areas during summer. The intensity (I) of
flashes decreases as the distance (r) increases and thus
most fireflies can communicate only up to several
hundred meters. In the implementation of the algorithm,
the flashing light is formulated in such a way that it gets
associated with the objective function to be optimized.
The steps are to be followed.
1) objective function
The light intensity I(r) varies according to the inverse
square law.
=

(7)

I(r)-intensity at the source.

r-observers distance from source.
If we take absorption coefficient into account, the light
intensity I varies with the square of distance r.
2

I = I0 er
(8)
2) generate initial population of fireflies
Initial population of fireflies (say n)
All Rights Reserved 2016 IJORAT

determine the light intensities of each of the

fireflies to find the brightness of each fireflies.
=

(10)

4) calculate the attractiveness of fireflies

=

(11)

5) Movement of less brightness fireflies towards n

brighter
one
The movement of the firefly I is attracted to
another more attracted firefly j is determined by,
= + , ( )+

(12)

6) Find the current best

Update light intensities of the fireflies and rank
the fireflies, after ranking of the fireflies, find the
current best solution.

IV.PROPOSED METHODOLOGY
In this proposed method, the hyperspectral image is first
read and the dimension is roughly reduced by removing
low SNR bands.Salinas is a popular hyperspectral
image. It has 224 bands in original, but after removing
noisy bands due to dense of water and atmospheric
condition, only 198 bands remain for further processing.
Virtual Dimensionality (VD) estimates the number of
endmembers present in the HSI image. Two familiar
methods widely used for estimating VD are HarsanyiFarrand-Chang (HFC) method as well as a noise
whitened version (NWHFC). HFC method does not
have a noise-whitening process, an alternative is to
modify the HFC method by including a noise-whitening
process as preprocessing to remove the second-order
statistical correlation such that the noise variance in the
corresponding correlation eigen value and covariance
eigen value will be the same. As a result, the VD
estimate can be more accurate due to the fact that the
noise variances have been decorrelated and do not have
effects on the eigen value comparison. The resulting
HFC method will be referred to as noise-whitened HFC
method.PCA is a mathematical procedure that
transforms a number of possibly correlated variables
into a smaller number of uncorrelated variables called
principal components.It aims to find the project
directions along which the reconstructing error to the
original data is minimum, and projects the original data
into a lower dimensional space scanned by those
directions corresponding to the top eigenvalues. In face
recognition, those directions which are the eigenvectors
of the covariance matrix of face images are orthogonal
2

International Journal of Research in Advanced Technology - IJORAT

Vol. 1, Issue 7, JULY 2016

basis vectors. PCA is applied for feature extraction. And

hence PCA is used for dimensionality reduction.
FIREFLY algorithm is adopted for band selection
because it is not only faster than Particle Swarm
Optimization, but also it provides better and more
consistent results. Classification process is done by
SVM. After getting best bands then the image is
classified using the selected bands with SVM classifier
The work flow of our proposed work has been depicted
in Fig.1.
HYPERSPECTRAL DATA

Fig.2 Salinas Image

VD ESTIMATION

PCA BASED FEATURE

EXTRACTION

FIREFLY BASED BAND

SELECTION

Fig.3 Groundtruth of Salinas

SVM CLASSIFICATION

Using the VD estimation, the number of bands to be

selected has been determined and it is given in table1.
Table 1. VD Estimation

CLASSIFIED OUTPUT

pf=10^1

pf=10^2

pf=10^3

pf=10^4

pf=10^5

15

14

12

11

Fig.1 Block Diagram for Proposed Methodology

V. EXPERIMENTAL RESULTS
Salinas is a popular hyper spectral data has been
considered for our work. This image is of size 217x 512
pixels. Each pixel is recorded at 224 bands covering the
wavelengths from 0.4 m to 2.5 m. The spectral
resolution is 10nm. As the original image is too large,
which is very expensive in terms of computational cost,
a region of Salinas comprising 83 x 86 pixels is used.
Specifically, there are six classes in this image,
i.e.("#1Brocoli_green_weeds","#2Corn_senesced_green
_weed","#3Lettuce_romaine_4wk","#4Lettuce_romaine
_5wk","#5Lettuce_romaine_6wk","#6Lettuce_romaine_
7wk) respectively. The Salinas image has been shown
in Fig.2 and its ground truth has been shown in Fig.3

All Rights Reserved 2016 IJORAT

VD

The most informative bands are selected using the

proposed FIREFLY algorithm and the result has been
tabulated in table 2.The selected bands are given as
input to the SVM classifier and the classified output is
shown in Fig.4
Table2. Selected bands from Salinas data

Method

Selected band indices (9 from198 bands)

FIREFLY

136 144 82 17 44 177 30 167 102

International Journal of Research in Advanced Technology - IJORAT

Vol. 1, Issue 7, JULY 2016

[6]

[7]

[8]

[9]

Fig.4. Classified map of Salinas image

The Classification Accuracy of SVM [17] have been
calculated using this formula below
ACC = TP+TN/TP+TN+FP+FN (13)
TP-number of true-positive decisions
TN-number of true-negative decisions
FP-number of false-positive decisions
FN-number of false-negative decisions

VI. CONCLUSION
Obtaining of correct spectral features is the major
task of Band Selection. This paper provides originality of
bands using Firefly based selection (FA) method. The FA is
one of the latest artificial intelligence algorithms developed.
This algorithm is intriguing because it is not only faster
than Particle Swarm Optimization (PSO), but also it
provides better, more consistent results in existing method
(PSO), has classification accuracy of Salinas image with
SVM classifier is 94.4%. Using the proposed Firefly based
band selection method, the classification accuracy obtained
for Salinas image with SVM classifier is 96.3%.So FA is a
one of the best evolutionary algorithm compared with other
existing methods.

REFERENCES
[1]

[2]

[3]

[4]

[5]

Adil Hashmi, NishantGoel, ShurtiGoel, Divya

Gupta, Firefly Algorithm for Unconstrained
Optimization. IOSR Journal of Computer
Engineering (IOSR-JCE) Jun. 2013,
C. Sheffield, Selecting band combinations from
multispectral data, Photogramm. Eng. Remote
Sens., vol. 51, no. 6, pp. 681687,1985.
C.-I Chang and Q. Du, Estimation of the number of
spectrally distinct signal sources in hyperspectral
imagery, IEEE Trans. Geosci. Remote Sens. 42 (3),
pp. 608619, 2004.
C.-I. Chang and S. Wang, Constrained band
selection for hyperspectral imagery, IEEE Trans.
Geosci. Remote Sens., vol. 44, no. 6, pp. 15751585,
Jun. 2006.
HeYang,QianDu,andGenshe Chen, Particle Swarm
Optimization-Based Hyperspectral Dimensionality
Reduction for Urban Land Cover Classification,
IEEE journal of selected topics in applied earth

All Rights Reserved 2016 IJORAT

[10]

[11]

[12]

[13]

[14]

[15]
[16]

[17]

observations and remote sensing, vol. 5, no. 2, april

2012
H. Yang, Q. Du, H. Su, and Y. Sheng, An efficient
method for supervised hyperspectral band
selection, IEEE Geosci. Remote Sens. Lett.,vol. 8,
no. 1, pp. 138142, Jan. 2011.
K.Chandrasekar
and
N.V.RamanaPerformance
Comparison of GA, DE, PSO and SA Approaches in
Enhancement of Total Transfer Capability using
FACTS Devices Journal of Electrical Engineering
& Technology 2012.
Kang
Sun,XiuruiGeng,andLuyanJi,Exemplar
Component Analysis: A Fast Band Selection
Method for Hyperspectral Imagery, IEEE geo
science and remote sensing letters, vol. 12, no. 5,
may 2015,
Pei-Wei TSai, Jeng-Shyang Pan, Bin-Yih Liao, ShuChuan Chu, Enhanced Artificial Bee Colony
Optimization, International Journal of Innovative
Computing, Information and Control, Volume 5,
Number 12, December 2009.
Q. Du and H. Yang, Similarity-based unsupervised
band selection for hyperspectral image analysis,
IEEE Geosci. Remote Sens. Lett., vol. 5,no. 4, pp.
564568, Oct. 2008
R. Huang and M. He, Band selection based on
feature weighting for classification of hyperspectral
imagery, IEEE Geosci. Remote Sens.Lett., vol. 2,
no. 2, pp. 156159, Apr. 2005
R. Poli, J. Kennedy, T. Blackwell, Particle swarm
optimization: An overview (Springer Science and
Business Media, LLC 2007).
S. Ding and L. Chen, Classification of
hyperspectral remote sensing images with support
vector
machines
and
particle
swarm
optimization,IEEE in Proc. Int. Conf. Inf. Eng.
Comput. Sci., 2009, no. 19-20, pp.15.
S. Jia, Z. Ji, Y. Qian, and L. Shen, Unsupervised
band selection for hyperspectral imagery
classification without manual band removal, IEEE
J. Sel. Topics Appl. Earth Observ. Remote Sens.,
vol. 5, no. 2, pp. 531543, Apr. 2012.
Yang, X. S. (2008), Nature-Inspired Metaheuristic
Algorithms (Luniver Press).
Y. Qian, F. Yao, and S. Jia, Band selection for
hyperspectral imagery using affinity propagation,
IET Comput. Vis., vol. 3, no. 4, pp. 213222, Dec.
2009.
Felipe Alonso-Atienza, Eduardo Morgado, Lorena
Fernandez-Martnez,
ArcadiGarca-Alberola,
Jose Luis Rojo-A lvarezDetection of LifeThreatening Arrhythmias Using Feature Selection
and Support Vector Machines, IEEE Transactions
on Biomedical Engineering, vol. 61, no. 3, March
2014.