A Survey on: Hyper Spectral Image Segmentation and Classification Using FODPSO

International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 11 100 – 103
_______________________________________________________________________________________________
100
IJRITCC | November 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
A Survey on: Hyper Spectral Image Segmentation and Classification using
FODPSO
Jha Soni Vinit
Department of Computer
Engineering, Mumbai University,
Shree L.R. Tiwari College of
Engineering and Technology,
Mira Road,
India sonijha.1911@gmail.com
Prof Sachin Bojewar
Department of Information
Technology, Mumbai University,
Vidyalankar Institute of
Engineering ,Mumbai, India
sachin.bojewar@vit.edu.in
Prof. Deepali Patil
Department of Computer
Engineering, Mumbai University,
Shree L.R. Tiwari College of
Engineering and Technology,
Mira Road, India
deep.patil1987@gmail.com
Abstract- The Spatial analysis of image sensed and captured from a satellite provides less accurate information about a remote
location. Hence analyzing spectral becomes essential. Hyper spectral images are one of the remotely sensed images, they are
superior to multispectral images in providing spectral information. Detection of target is one of the significant requirements in
many are assuc has military, agriculture etc. This paper gives the analysis of hyper spectral image segmentation using fuzzy C-
Mean (FCM)clustering technique with FODPSO classifier algorithm. The 2D adaptive log filter is proposed to denoise the
sensed and captured hyper spectral image in order to remove the speckle noise.
Keywords: Hyper spectral image segmentation, FODPSO-FCM
*****
I. INTRODUCTION
The objective of this research is to provide a better spectral
information about remotely sensed images which are the
most significant requirement of various areas such as
military and agriculture. The proposed algorithm is also
used to segment the mineral information from the hyper
spectral satellite images. However, the data so-obtained
consist of high dimension (hundreds of bands) spectrum
that needs to be processed to obtain classification data
through which we can identify the existing minerals .This
paper provides a better methodology to get image in good
quality, provide low computational complexity, high visual
quality and can achieve good performance.
In recent times a lot of work has been carried out in the field
of image segmentation and classification. And all the
techniques proposed have, in their unique way, proven to be
very useful. But the main hindrance that remains is that
degradation in the image quality, accuracy of the
segmentation is not proper, the algorithms depends only on
the intensities not on the shape and texture. Further research
needs to be undertaken to tackle these drawbacks.
Even though lot of research has been carried out for
hyper spectral image segmentation, unfortunately no
credible techniques have been proposed for providing
an efficient result which can give the spatial and
contextual information. The automatic seed select is
possible for image segmentation. So the accuracy of
the segmentation is high. The algorithm depend not
only the intensity but also depend on the shape and
texture. So the segmentation of the object and pixel is
proper.
II. Hidden Markov Random Field [HMRF]
MRFs can be described as 2-D statistic processes over
discrete pixel lattices [5] as they are a family of random
probabilistic models. MRFs can be considered as a powerful
tool for incorporating spatial and contextual information into
the classification framework [6]. Lately the Hidden Markov
Random Field (HMRF) was introduced as a special case of
the hidden Markov model (HMM). In HMRF, instead of
Markov chains in HMM, the MRF is used as a statistic
process. Hence , HMRF is not restricted to 1-D so it can be
used in order to extract spatial information from 2-D and 3-
D images also. There is extensive literature on the use of
MRFs for increasing the accuracy of classification. For
instance, in [7], also the result of the probabilistic SVM was
regularized by an MRF. The mean field-based SVM
regression was used for image classification In [5]. Also, in
[6] and [7] for modeling spatial and contextual information
to improve the accuracy of the classification the MRFs were
taken into consideration. In addition, a generalized MRF,
called conditional MRF, was developed for the spectral and

Volume: 5 Issue: 11 100 – 103
_______________________________________________________________________________________________
101
_______________________________________________________________________________________
spatial classification of sensed remote sensing images. For
incorporating spectral and contextual information into a
framework [7] the concept of HMM was used for
performing unsupervised classification of sensed remote
sensing multispectral images.
In addition, Gaussian MRF was employed in for the purpose
of segmentation and anomaly detection. Based on the
previous discussion, the integration of SVM classifiers and
MRFs for the accurate classification of remote sensing
images by considering both spectral information and spatial
information into the same framework is completely obvious.
A new fully automatic spectral and spatial approach is
introduced for the classification of hyperspectral images.
This approach was based on the SVM and HMRF. In order
to preserve the edges in the classification map (CM), a
gradient step based on the Sobel edge detector was taken
into account.
III. Fuzzy Clustering Means[FCM]
In fuzzy clustering method each pixel can belong to
several land cover classes partially. It gives membership
vectors for each sample for each class with the ranges
between 0 and 1. Thus a pixel can belong to a class to a
certain degree and may belong to another class to another
degree and the degree of belongingness is indicated by
fuzzy membership values. In the feature space if any
point lies closer to the centre of a cluster, then its
membership grade is also higher (closerto1)for that
cluster .In case of fuzzy membership grades feature space
is not sharply partitioned into clusters, the main
advantage of this approach is that no spectral information
is lost like in the case of hard partitioning of feature
space[1]. In hard partitioning one pixel is assigned to a
single land cover class, thus there is some loss of
information, but in case of fuzzy partitioning for a single
pixel, there is partial belongingness to several land cover
classes, it preserves information for a class.
Disadvantage: FCM is sensitive to initialization and is easily
trapped in local optima.
IV. Particle Swarm Optimization[PSO]
Particle Swarm Optimization (PSO) is a population-based
statistic optimization technique developed by Dr. Eberhart
and Dott. in 1995, inspired by the social behavior of the
neighborhood or aquaculture. PSO shares many analogies
with evolutionary computing techniques, such as genetic
algorithms (GAs). Partial swarm optimization is similar to a
genetic algorithm [8] where the system is initialized with a
random solution population. However, it is not a genetic
algorithm, since every potential solution is also assigned at
random speed and potential solutions, called particles, are
"piloted" through hyperspace.
Each particle traces its coordinates in the hyperspace
associated with the best solution (aptitude) it has
achieved so far. The value of this aptitude is also stored and
this value is called pbest. Another "best" value is also
monitored. The "global" version of the particle swarm
optimizer tracks the best overall value and position obtained
so far by any particle in the population; the concept of
optimizing particle transplantation is at each stage of time
changing the speed (acceleration) of each particle to its
pbest and gbest (global version). Acceleration is weighted
by a random term, with separate random numbers generated
for pbest and gbest acceleration.
Disadvantage:The disadvantages of particle swarm
optimization (PSO) algorithm are that it is easy to fall into
local optimum in high-dimensional space and have a low
convergence rate in the iterative process.
V. Clustering with FCM-PSO
The Fuzzy clustering is an important issue that is the subject
of active research in different real-world applications. The c-
means Fuzzy (FCM) algorithm is one of the most popular
fuzzy clustering techniques, because it is efficient, simple
and easy to deploy. Particle Swarm Optimization (PSO) is a
global stochastic optimization tool that is used in many
optimization issues. In many areas, a hybrid diffused cluster
based on FCM and diffused PSO (FPSO) has been proposed
that uses the merits of both algorithms and this method is
efficient and can reveal encouraging results. The diffusion
of diffused particle swirls can be more effective in some
cases. Fuzzy hybrid means and widespread particle swirling
optimization can get the best results.
To solve the issues of the fuzzy clustering algorithm (FCM),
when applied to segmentation of images, such as easy to
cheat on optimum and huge local calculations, a
segmentation algorithm based on modified particle swarm
optimization (MPSO) and the FCM algorithm was proposed.
The simulation results and comparison between the new
algorithm and the FCM algorithm indicate that the new
algorithm can achieve better segmentation effects and
overcome the existing issues of FCM algorithm in different
performance such as average dispersion, maximum distance
between pixels and its cluster center and the minimum
distance between any pair of clusters.
i ) Clustering
Clustering is the process of assigning data sample objects to
a set of disjoint groups called clusters, so that objects in
each cluster are more similar to objects in different clusters.
Clustering techniques are applied in many application areas,


Volume: 5 Issue: 11 100 – 103
_______________________________________________________________________________________________
102
_______________________________________________________________________________________
such as pattern recognition [1], data extraction [2],
automatic learning [3], and so on. Grouping algorithms can
typically be categorized as Hard, Fuzzy, Possible, and
Probabilistic algorithms.
ii ) K-means
K-means is one of the most widely used hard cluster
algorithms that divides data objects into cluster k where the
number of clusters k is decided in advance depending on the
purpose of the application. This template is not appropriate
for real data sets where there are no boundaries defined
between clusters. After the widespread theory presented by
LotfiZadeh, the researchers placed diffuse theory in a
grouping. Fuzzy algorithms can partially assign to multi-
cluster data objects. The degree of fuzzy cluster membership
depends on the proximity of the data object to the cluster
centers.
Disadvantage: K-means is not appropriate for real data sets
where no boundaries exist between clusters.
iii) FCM-FPSO
The most popular fuzzy clustering algorithm is fuzzy c mean
(FCM) clustering algorithm, introduced by [9] Bezdek in
1974, is now widely used. The Fuzzy c-mean cluster
algorithm is effective, but random selection at the center
points makes the iterative process easily fall off the local
optimum solution. To solve this problem, evolutionary
algorithms, such as genetic algorithms (GAs), simulated
annealing (SA), ant colony optimization (ACO) and particle
swarm optimization (PSO), have recently been successfully
applied. Particle Swarm Optimization (PSO) is a population-
based optimization technique that can be easily implemented
and applied to solve the various problems of optimization
features, or problems that can turn into optimization feature
problems [8]. A version of particle swarm optimization for
TSP called FPSO (Fuzzy Particle Swarm Optimization) was
proposed by Pang et al. [10] In 2004.
An FCM and FPSO-based hybrid clustering algorithm called
FCM-FPSO was also proposed. Experimental results over
six real-life data series indicate that the FCM-FPSO
algorithm is superior to the FCM algorithm and the FPSO
algorithm for the problem. The rest of the document was
organized as follows to show The FCM algorithm is faster
than the FPSO algorithm because it requires less function
evaluations, but generally falls into local optima. The FCM
algorithm was integrated with the FPSO algorithm to form a
hybrid clustering algorithm called FCM-FPSO that retains
the merits of FCM and PSO algorithms. The FCM-FPSO
algorithm applies FCM to particles in the swarm every
number of iterations / generations so that the aptitude value
of each particle is improved.
Disadvantage:A general problem with these techniques,
such as the PSO and an FCM-PSO algorithm, is that they
can be trapped in local optimum locations so that they can
be successful in some problems, but not in others.
VI. Clustering and segmentation with FODPSO-FCM
Conventional hard grading techniques do not consider the
continuous changes of the different soil cover classes from
one to the other. For example, the crisp K-means standard
can be considered as the most widely used clustering
technique in the pattern recognition field [1]. However, this
technique uses hard partitions, in which each data point
belongs exactly to a cluster. To model the gradual changes
of the limits, "soft" classifiers were used. To further improve
the existing techniques, we combine the FCM with the
fractional Order Darwin PSO (FODPSO) previously
proposed by Couceiro et al. and Ghamisi et al. and applied
for different applications [14]. The FODPSO algorithm
takes advantage of a cooperative paradigm where particles
within each swarm cooperate, while more swarm compete to
find the most suitable solution, that is, the optimal solution.
By combining FODPSO with the FCM technique, named in
this paper as FODPSO-FCM, each particle will be
represented by a particular cluster configuration and the
objective FCM function. FODPSO's emerging collective
properties, along with fractional order velocity and a set of
penalty award rules designed to simulate Darwin's natural
selection mechanism, converge to optimum cluster
configuration. In addition, in order to speed up the grouping
process, you intend to use the image intensity histogram as
input instead of raw image data. In this paper we propose a
new widespread grouping approach based on evolution. This
approach benefits from a fractional calculation approach to
improve the convergence speed of traditional FCM, while at
the same time benefiting from the same natural selection
mechanism of the original PSO to avoid local optima
stagnation[4] . Therefore, the proposed widespread
clustering approach is used to improve the classification of
hyperspectral images with FODPSO.
However, it is a difficult classifier and can not model
temporary gradual changes between different classes.
Conversely, the new Fuzzy-based clustering technique can
model gradual changes between different classes. In the
proposed method, the Combination of FODPSO and the new
fuzzy- clustering strategy can be considered as a desirable
strategy to classify hyperspectral imaging.The experimental
result for the hyperspectral image demonstrates that the
performance of the proposed method, which uses a
biologically inspired behavior based on natural selection and
noninteger convergence, results in a statistically significant
improvement in terms of overall classification accuracy . It

Volume: 5 Issue: 11 100 – 103
_______________________________________________________________________________________________
103
_______________________________________________________________________________________
should be noted that because of the effective
implementation of all clustering approaches in this proposed
method based on an image histogram, clustering approaches
here are very fast and can lead to a conclusion within a few
seconds.
VII. CONCLUSION
This paper presents different clustering and segmentation
methods and also included FCM with FODPSO as new
method for clustering and segmentation which gives better
performance than existing methods of clustering and
segmentation. So the FCM-FODPSO can be used for
clustering and segmentation with better performance.
REFERENCES
[1] A. Dutta, “Fuzzy c-means classification of multispectral
data incorporating spatial contextual information by
using Markov random field,” M.S. thesis,
Geoinformatics, ITC, 2009.
[2] W. Wang, Y. Zhang, Y. Li, and X. Zhang, “The global
fuzzy c-means clustering algorithm,” in Intelligent
Control and Automation, vol. 1, pp. 3604–3607, 2006.
[3] J. Kennedy and R. Eberhart, “A new optimizer using
particle swarm theory,” in Proc. IEEE 6th Int. Symp.
Micro Mach. Hum. Sci., 1995, vol. 34, no. 2008, pp. 39–
43.
[4] M. S. Couceiro, R. P. Rocha, N. M. F. Ferreira, and J. A.
T. Machado, “Introducing the fractional order darwinian
PSO,” Signal Image VideoProcess., vol. 102, no. 1, pp.
8–16, 2007.
[5] H. Derin and P. A. Kelly, “Discrete-index Markov-type
random processes,” Proc. IEEE, vol. 77, no. 10, pp.
1485–1510, Oct. 1989.
[6] G. Moser, S. B. Serpico, and J. A. Benediktsson, “Land-
cover mapping by Markov modeling of spatial–
contextual information in very-high resolution remote
sensing images,” Proc. IEEE, vol. 101, no. 3, pp. 631–
651, Mar. 2013.
[7] Y. Tarabalka, M. Fauvel, J. Chanussot, and J. A.
Benediktsson, “SVM- and MRF-based method for
accurate classification of hyperspectral images,” IEEE
Geosci. Remote Sens. Lett., vol. 7, no. 4, pp. 736–740,
Oct. 2010
[8] J. Kennedy and R. Eberhart, “A new optimizer using
particle swarm theory,” in Proc. IEEE 6th Int. Symp.
Micro Mach. Hum. Sci., 1995, vol. 34, no. 2008, pp. 39–
43.
[9] Z. Xian-cheng, “Image segmentation based on modified
particle swarm optimization and fuzzy c-means
clustering algorithm,” in Proc. 2nd Int.Conf. Intell.
Comput. Technol. Autom., 2009, pp. 611–616.
[10] H. Izakian and A. Abraham, “Fuzzy c-means and fuzzy
swarm for fuzzy clustering problem,” Expert Syst. Appl.,
vol. 38, no. 3, pp. 1835–1838, 2011.
[11] Y.del Valle et al., “Particle swarm optimization: Basic
concepts, variants and applications in power systems,”
IEEE Trans. Evol. Comput., vol. 12, no. 2, pp. 171–195,
Apr. 2008.
[12] P. Ghamisi and J. A. Benediktsson, “Feature selection
based on hybridization of genetic algorithm and particle
swarm optimization,” IEEE Geosci.Remote Sens. Lett.,
vol. 12, no. 2, pp. 309–313, Feb. 2015.
[13] D. Arthur and S. Vassilvitskii, “How slow is the k-means
method?” in Proc. 22nd Annu. Symp. Comput. Geom.
(SCG’06), 2006, pp. 144–153.
[14] S. P. Lloyd, “Least squares quantization in PCM,” IEEE
Trans. Inf., vol. 28, no. 2, pp. 129–136, Mar. 1982.
[15] P. Agarwal and N. Mustafa, “K-means projective
clustering,” in Proc. 23rd ACM SIGMOD-SIGACT-
SIGAR Symp. Principles Database Syst. (PODS ’04),
2004, pp. 155–165.

A Survey on: Hyper Spectral Image Segmentation and Classification Using FODPSO

Recommended

More Related Content

What's hot (18)

Similar to A Survey on: Hyper Spectral Image Segmentation and Classification Using FODPSO (20)

More from rahulmonikasharma (20)

Recently uploaded (20)

A Survey on: Hyper Spectral Image Segmentation and Classification Using FODPSO