Evolutionary Approach for Land cover classification using GA based
Fuzzy Clustering Techniques
Final Report for the Minor Research Project submitted to
Tamil Nadu State Council for Higher Education
Chennai-600 005.
by
Dr. N. Sujatha
Assistant professor
PG & Research Department of Computer Science
Raja Dorai Singam Govt Arts College
Sivagangai- 630561 ,Tamilnadu.
TAMIL NADU STATE COUNCIL FOR HIGHER EDUCATION
FORMAT FOR SUBMISSION OF FINAL REPORT FOR
MINOR RESEARCH PROJECT
1. Name of the Teacher
: Dr.N.Sujatha
2. Designation
: Assistant professor
3. Communication Address:
: Dr.N.Sujatha,
Assistant Professor,
PG & Research Department of Computer
Science,
Raja Dorai Singam Govt Arts College,
Sivagangai-630 561.
Contact No
: 9443093017
E-Mail-Id
:sujamurugan@gmail.com
4. Institutional Address
: Raja Dorai Singam Govt Arts College,
Sivagangai-630 561.
5. Project Titile
:” Evolutionary approach for Land cover
classification using GA based Fuzzy clustering
techniques”
6. Sector
:Computer Science (Science)
7. Grant approved and expenditure incurred during the project
Total amount approved (Rs.)
1,000,00.00
Total Expenditure (Rs.)
1,00,103.00
CONTENTS
Chapters
Title
Page No
1
INTRODUCTION
1.1
Data mining
1
1.2
Introduction to Image Processing
5
1.3
Introduction to Remote Sensing
7
1.4
Introduction to Genetic Algorithm
8
2
LITERATURE SURVEY
2.1
Unsupervised Classification
11
2.2
Supervised Classification
16
2.3
Classification using genetic algorithm
18
3
PREPROCESSING
3.1
Preprocessing Image
21
3.2
Median Filters
28
4
LAND COVER CLASSIFICATION
4.1
Introduction
31
4.2
Cluster Analysis
32
4.3
Median Filters
33
4.4
Fuzzy C Means
34
4.5
Genetic Algorithm
36
4.6
Proposed work
39
5
CONCLUSION
44
Chapter 1
Introduction
1.1 Data Mining
Progress in digital data acquisition and storage technology has resulted in the growth
of huge databases. This has occurred in all areas of human endeavor, from the mundane (such
as supermarket transaction data, credit card usage records, telephone call details, and
government statistics) to the more exotic (such as images of astronomical bodies, molecular
databases, and medical records). The interest has grown in the possibility of tapping these
data, of extracting from them information that might be of value to the owner of the database.
The discipline concerned with this task has become known as data mining.
Data mining [1] refers to extracting or “mining” knowledge from large amounts of
data. Many people treat data mining as a synonym for another popularly used term,
Knowledge Discovery from Data, or KDD.
Knowledge discovery as a process consists of an iterative sequence of the following
steps:

Data cleaning (to remove noise and inconsistent data)

Data integration (where multiple data sources may be combined)

Data selection (where data relevant to the analysis task are retrieved from the
database)

Data transformation (where data are transformed or consolidated into forms
appropriate for mining by performing summary or aggregation operations)

Data mining (an essential process where intelligent methods are applied in order to
extract data patterns)

Pattern evaluation (to identify the truly interesting patterns representing knowledge
based on some interestingness measures)

Knowledge presentation (where visualization and knowledge representation
techniques are used to present the mined knowledge to the user).
Frequently, the data to be mined is first extracted from an enterprise data warehouse
[2] into a data mining database or data mart. OLAP (On-Line Analytical Processing) is
part of the spectrum of decision support tools. Traditional query and report tools describe
what is in a database. OLAP goes further; it’s used to answer why certain things are true.
The user forms a hypothesis about a relationship and verifies it with a series of queries
against the data. The OLAP analyst generates a series of hypothetical patterns and
relationships and uses queries against the database to verify them or disprove them.
OLAP analysis is essentially a deductive process. Data mining is different from OLAP
because rather than verify hypothetical patterns, it uses the data itself to uncover such
patterns. It is essentially an inductive process.
Task of Data Mining
Data mining as a term used for the specific set of six activities or tasks [3] as follows:

Classification

Estimation

Prediction

Affinity grouping or Association Rules

Clustering

Description and Visualization
The first three tasks - classification, estimation and prediction are all examples of
directed data mining or supervised learning. In directed data mining, the goal is to use the
available data to build a model that describes one or more particular attribute(s) of interest
(target attributes or class attributes) in terms of the rest of the available attributes. The next
three tasks – association rules, clustering and description are examples of undirected data
mining i.e. no attribute is singled out as the target; the goal is to establish some relationship
among all the attributes.
Clustering
During a cholera outbreak in London in 1854, John Snow used a special map to plot
the cases of the disease that were reported. A key observation, after the creation of the map,
was the close association between the density of disease cases and a single well located at a
central street. After this, the well pump was removed putting an end to the epidemic.
Associations between phenomena are usually harder to detect, but the above is a very simple,
and for many researchers, the first known application of cluster analysis. Since then, cluster
analysis has been widely used in several disciplines, such as statistics, software engineering,
biology, psychology and other social sciences, in order to identify natural groups in large
amounts of data. These data sets are constantly becoming larger, and their dimensionality
prevents easy analysis and validation of the results. Clustering has also been widely adopted
by researchers within computer science and especially the database community, as indicated
by the increase in the number of publications involving this subject, in major conferences.
Clustering [4] [5] is a basic tool used in data analysis, pattern recognition and data
mining for finding unknown groups in data. It can be considered the most important
unsupervised learning problem; so, as every other problem of this kind, it deals with finding a
structure in a collection of unlabeled data. It is therefore a collection of objects which are
“similar” between them and are “dissimilar” to the objects belonging to other clusters.
Besides the term data clustering as synonyms like cluster analysis, automatic classification,
numerical taxonomy, botrology and typological analysis.
Cluster analysis [8] is a difficult problem because many factors i.e., effective
similarity measures, criterion functions, algorithms are come into play in devising a perfect
clustering technique for a given clustering problems. Also no clustering method can
effectively handle all sorts of cluster structures i.e. shape, size and density. Sometimes the
quality of the clusters that are found can be improved by preprocessing the given data. It is
common to try to find noisy values and eliminate them by a preprocessing step. The input for
a system of cluster analysis is a set of samples and a measure of similarity (or dissimilarity)
between two samples. The output from cluster analysis is a number of groups /clusters that
form a partition, or a structure of partitions, of the data set. The final goal of clustering can be
mathematically described as follows:
Where X denotes the original data set, Ci, Cj are clusters of X, and n is the number of
clusters. Data Pre-processing [9] describes any kind of processing performed on raw data to
prepare it for further processing method. This process includes Data cleaning, which fill in
missing values, smooth noisy data, identify or remove outliers and resolve in consistencies;
Data integration, which integration of multiple data bases, data cubes, or files; Data
transformation, which is normalization and aggregation.
Clustering has wide applications in

Image Processing

Document classification

Pattern Recognition

Spatial Data Analysis

Economic Science

Cluster Web log data to discover groups of similar access patterns
Types of Clustering
There are four types of Clustering. They are Partitional, Hierarchical, Density-Based
and Grid-Based Clustering.
Partitional type of Clustering
The Partition Clustering algorithm [10] splits the data points into k partition, where
each partition represents a cluster. Given D, a data set of n objects, and k, the number of
clusters to form, a partitioning algorithm organizes the objects into k partitions (k ≤ n), where
each partition represents a cluster. The clusters are formed to optimize an objective
partitioning criterion, such as a dissimilarity function based on distance, so that the objects
within a cluster are “similar,” whereas the objects of different clusters are “dissimilar” in
terms of the data set attributes. The cluster should exhibit two properties; they are (1) each
group must contain at least one object (2) each object must belong to exactly one group. In
this type of clustering, the familiar algorithms are K-Means, K-Medoids, CLARANS, Fuzzy
K-Means, K-Modes.
Hierarchical type of Clustering:
In hierarchical clustering [11], a treelike cluster structure (dendrogram) is created through
recursive partitioning (divisive methods) or combining (agglomerative) of existing clusters.
Agglomerative clustering methods initialize each observation to be a tiny cluster of its own.
Then, in succeeding steps, the two closest clusters are aggregated into a new combined
cluster. In this way, the number of clusters in the data set is reduced by one at each step.
Eventually, all records are combined into a single huge cluster. Divisive clustering methods
begin with all the records in one big cluster, with the most dissimilar records being split off
recursively, into a separate cluster, until each record represents its own cluster.
Hierarchical
Clustering,
the
familiar
algorithms
are
AGNES,
DIANA,
In
CURE,
CHAMELEON, BIRCH, ROCK.
Density-Based Clustering:
The Density-Based Clustering [12] method group objects according to specific
density objective functions. Density is usually defined as the number of objects in a particular
neighborhood of a data objects. In these approaches a given cluster continues growing as long
as the number of objects in the neighborhood exceeds some parameter. This is considered to
be different from the idea in partitional algorithms that use iterative relocation of points given
a certain number of clusters. The algorithms in this method include DBSCAN, DENCLUE
and OPTICS.
Grid-Based Clustering:
Grid based methods [13] quantize the object space into a finite number of cells (hyperrectangles) and then perform the required operations on the quantized space. It has fast
processing time which depends on number of cells in each dimension in quantized space. It
uses a multi resolution grid data structure. The grid-based clustering approach differs from
the conventional clustering algorithms in that it is concerned not with the data points but with
the value space that surrounds the data points. The important algorithms in this method
include STING, Wavelet and CLIQUE.
1.2 Introduction to Image Processing:
Origin of Image Processing:
The basic techniques [7] used to generate high-quality images from digital data were
originally developed to process spacecraft images of Mars. These images represented one
experiment that was included among others on the Mariners 4, 6, 7 and 9 missions. These
digital images have a dynamic range (sensitivity) from 10 to 50 times that of the eye. Thus in
raw format, only a small part of the data is available to the eye.
In order to derive the maximum information available from the digital images,
techniques were developed first at IPL/JPL and later at the U.S. Geological Survey to extract
these data and put them into an optimally interpretable format for the human eye (Levinthal
and others, 1973, Batson, 1973). One of the basic problems that arose in processing the
Mariner images was caused by coherent and random noise introduced by the detector, by
digital recording, and by data transmission, reception, and reduction systems. Most of the
processes used to enhance digital images will also enhance noise, thereby seriously degrading
the quality of the final image. Therefore, major efforts were undertaken to develop image
processing techniques to remove the noise (Rindfleisch and others, 1971; Chavez and
Soderblom, 1975). After these clean-up procedures were applied to create noise-free data
bases, other techniques were used to improve the images further, including techniques for
removing effects of variation in solar illumination angle and for correcting geometry. The
image was then processed to enhance fine detail (high-pass filtering) or albedo variations
(low-pass filtering) and to enhance contrast (stretching). Most of the image-processing
techniques developed for Mariner 9 images were easily modified for application to Landsat
(formerly ERTS, or Earth Resources Technology Satellite) data, although some alteration
was necessary because of the much larger image data sets. Landsat data were acquired in four
spectral bands (bands 4, 5, 6, and 7, respectively, 0.5 to 0.6, 0.6 to 0.7, 0.7 to 0.8, and 0.8 to
1.1/xm). Each Landsat image contains roughly 60 times more data than a Mariner image.
Although new problems were introduced because of this larger data set, the final product
contains much more information than that available from Mariner 9 data.
An image [6] may be defined as a two-dimensional function, f(x,y) where x and y are
spatial (plane) coordinates, and the amplitude of f at any pair of coordinates (x,y) is called the
intensity of a gray level of the image at that point. When x,y and the intensity values of f are
all finite, discrete quantities, we call the image a digital image. The field of digital image
processing refers to processing digital images by means of a digital computer.
Image Processing [14] is a technique to enhance raw images received from
cameras/sensors placed on satellites, space probes and aircrafts or pictures taken in normal
day-to-day life for various applications. Various techniques have been developed in Image
Processing during the last four to five decades. Most of the techniques are developed for
enhancing images obtained from unmanned spacecrafts, space probes and military
reconnaissance flights. Image Processing systems are becoming popular due to easy
availability of powerful personnel various applications such as:

Remote Sensing

Medical Imaging

Non-destructive Evaluation

Forensic Studies

Textiles

Material Science.

Military

Film industry

Document processing

Graphic arts

Printing Industry
1.3 Introduction to Remote Sensing and Image Processing
Of all the various data sources used in GIS [15], one of the most important is
undoubtedly that provided by remote sensing. Through the use of satellites, we now have a
continuing program of data acquisition for the entire world with time frames ranging from a
couple of weeks to a matter of hours. Very importantly, we also now have access to remotely
sensed images in digital form, allowing rapid integration of the results of remote sensing
analysis into a GIS. The development of digital techniques for the restoration, enhancement
and computer-assisted interpretation of remotely sensed images initially proceeded
independently and somewhat ahead of GIS. However, the raster data structure and many of
the procedures involved in these Image Processing Systems (IPS) were identical to those
involved in raster GIS. As a result, it has become common to see IPS software packages add
general capabilities for GIS, and GIS software systems add at least a fundamental suite of IPS
tools.
IDRISI is a combined GIS and image processing system that offers advanced
capabilities in both areas. Because of the extreme importance of remote sensing as a data
input to GIS, it has become necessary for GIS analysts (particularly those involved in natural
resource applications) to gain a strong familiarity with IPS.
1.4 Introduction to Genetic Algorithm:
Genetic algorithms (GAs) [16] were invented by John Holland in the 1960s and were
developed by Holland and his students and colleagues at the University of Michigan in the
1960s and the 1970s. In contrast with evolution strategies and evolutionary programming,
Holland's original goal was not to design algorithms to solve specific problems, but rather to
formally study the phenomenon of adaptation as it occurs in nature and to develop ways in
which the mechanisms of natural adaptation might be imported into computer systems.
Holland's 1975 book Adaptation in Natural and Artificial Systems presented the genetic
algorithm as an abstraction of biological evolution and gave a theoretical framework for
adaptation under the GA. Holland's GA is a method for moving from one population of
"chromosomes" (e.g., strings of ones and zeros, or "bits") to a new population by using a kind
of "natural selection" together with the genetics−inspired operators of crossover, mutation,
and inversion. Each chromosome consists of "genes" (e.g., bits), each gene being an instance
of a particular "allele" (e.g., 0 or 1).
The selection operator chooses those chromosomes in the population that will be
allowed to reproduce, and on average the fitter chromosomes produce more offspring than the
less fit ones. Crossover exchanges subparts of two chromosomes, roughly mimicking
biological recombination between two single−chromosome ("haploid") organisms; mutation
randomly changes the allele values of some locations in the chromosome; and inversion
reverses the order of a contiguous section of the chromosome, thus rearranging the order in
which genes are arrayed. (Here, as in most of the GA literature, "crossover" and
"recombination" will mean the same thing.)
Holland's introduction of a population−based algorithm with crossover, inversion, and
mutation was a major innovation. (Rechenberg's evolution strategies started with a
"population" of two individuals, one parent and one offspring, the offspring being a mutated
version of the parent; many−individual populations and crossover were not incorporated until
later. Fogel, Owens, and Walsh's evolutionary programming likewise used only mutation to
provide variation.) Moreover, Holland was the first to attempt to put computational evolution
on a firm theoretical footing (see Holland 1975). Until recently this theoretical foundation,
based on the notion of "schemas," was the basis of almost all subsequent theoretical work on
genetic algorithms.
In the last several years there has been widespread interaction among researchers
studying various evolutionary computation methods, and the boundaries between GAs,
evolution strategies, evolutionary programming, and other evolutionary approaches have
broken down to some extent. Today, researchers often use the term "genetic algorithm" to
describe something very far from Holland's original conception. In this book I adopt this
flexibility. Most of the projects I will describe here were referred to by their originators as
GAs; some were not, but they all have enough of a "family resemblance" that I include them
under the rubric of genetic algorithm.
In 1992 John Koza has used genetic algorithm to evolve programs to perform certain
tasks. He called his method "genetic programming" (GP). A genetic algorithm (or GA) is a
search technique used in computing to find true or approximate solutions to optimization and
search problems. GAs are categorized as global search heuristics. They are a particular class
of evolutionary algorithms that use techniques inspired by evolutionary biology such as
inheritance, mutation, selection, and crossover (also called recombination). The evolution
usually starts from a population of randomly generated individuals and happens in
generations. In each generation, the fitness of every individual in the population is evaluated,
multiple individuals are selected from the current population (based on their fitness), and
modified to form a new population. The new population is used in the next iteration of the
algorithm. The algorithm terminates when either a maximum number of generations has been
produced, or a satisfactory fitness level has been reached for the population.
Genetic algorithms [17] are a family of computational models belonging to the class
of evolutionary algorithms, part of artificial intelligence. These algorithms encode a potential
solution to a specific problem on a simple chromosome like data structure. Genetic
algorithms are a family of computational models belonging to the class of evolutionary
algorithms, part of artificial intelligence These algorithms encode a potential solution to a
specific problem on a simple chromosome like data structure Uses techniques inspired by
natural evolution such as inheritance, mutation, selection and crossover. Genetic algorithms
are a family of computational models belonging to the class of evolutionary algorithms, part
of artificial intelligence These algorithms encode a potential solution to a specific problem on
a simple chromosome like data structure Uses techniques inspired by natural evolution such
as inheritance, mutation, selection and crossover They are often viewed as function
optimizers
Chapter 2
Literature Survey
Remote sensing involves gathering information about the earth's surface remotely, and
generally encompasses acquiring this data from aircraft or satellites. Remote sensing is very much an
interdisciplinary area of scientific investigation, and relies in large part on knowledge of physics,
mathematics, Computer science and geography.
The land cover pattern of a region is an outcome of natural and socio – economic factors and
their utilization by man in time and space. Land is becoming a scarce resource due to immense
agricultural and demographic pressure. Hence, information on land cover is essential for the
selection, planning and implementation of land use schemes to meet the increasing demands for
basic human needs and welfare. This information also assists in monitoring the dynamics of changes
in land cover.
Land cover change has become a central component in current strategies for managing
natural resources and monitoring environmental changes. The advancement in the concept of satellite
image processing greatly influences the land cover mapping thus providing an accurate evaluation of
the spread and health of the world’s forest, grassland, and agricultural resources has become an
important priority.
Image classification in Remote Sensing
Digital image classification techniques group pixels to represent land cover features. Land cover
could be forested, urban, agricultural and other types of features. There are three main image
classification techniques. They are Unsupervised image classification, Supervised image
classification, Object-based image analysis. Pixels are the smallest unit represented in an image.
Image classification uses the reflectance statistics for individual pixels. Unsupervised and supervised
image classification techniques are the two most common approaches. However, object-based
classification has been breaking more ground as of late.
2.1 Unsupervised Classification
Pixels are grouped based on the reflectance properties of pixels. These groupings are
called “clusters”. The user identifies the number of clusters to generate and which bands to
use. With this information, the image classification software generates clusters. There are
different image clustering algorithms such as K-means and ISODATA.
The user manually identifies each cluster with land cover classes. It’s often the case that
multiple clusters represent a single land cover class. The user merges clusters into a land
cover type. The unsupervised classification image classification technique is commonly used
when no sample sites exist.
Unsupervised Classification Steps:

Generate clusters

Assign classes
Xingping Wen et al., [21] proposed an unsupervised classification method. Firstly, the hyperspectral
remote sensing image was atmospherically corrected. Accuracy atmospheric correction is the key to
the classification. Then, endmember spectra were extracted using PPI algorithm, and the image was
classified using SAM. Traditionally SAM algorithm used constant threshold. They improved and
used adjustable threshold, and the pixel belong to class which has the smallest spectral angle. Finally,
the endmember spectra were clustered based on K-means algorithm and classes were combined
according to the K-means algorithm result. The final classification map was projected and outputted.
It is an effective classification method especially for hyperspectral remote sensing image. Users also
can adjust the endmember and classes number according to their applications.
Gaussian mixture models (GMM) are widely used for unsupervised classification
applications in remote sensing. Expectation-Maximization (EM) is the standard algorithm
employed to estimate the parameters of these models. However, such iterative optimization
methods can easily get trapped into local maxima. Researchers use population based
stochastic search algorithms to obtain better estimates. Caglar Art et al., [22] presented a
novel particle swarm optimization-based algorithm for maximum likelihood estimation of
Gaussian mixture models. The proposed approach provides solutions for important problems
in effective application of population based algorithms to the clustering problem. They
presented a new parametrization for arbitrary covariance matrices that allows independent
updating of individual parameters during the search process. They also described an
optimization formulation for identifying the correspondence relations between different
parameter orderings of candidate solutions. Experiments on a hyperspectral image show
better clustering results compared to the commonly used EM algorithm for estimating
GMMs.
Mohd Hasmadi et al., [23] described a study that was carried out to perform supervised and
unsupervised techniques on remote sensing data for land cover classification and to evaluate
the accuracy result of both classification techniques. The study used SPOT 5 satellite image
taken on January 2007 for 270 / 343 (path / row) as a primary data and topographical map
and land cover maps as supporting data. The land cover classes for the study area were
classified into 5 themes namely vegetation, urban area, water body, grassland and barren
land. Ground verification was carried out to verify and assess the accuracy of classification.
A total of 72 sample points were collected using Systematic Random Sampling. The sample
point represented 25% of the total study area. The results showed that the overall accuracy for
the supervised classification was 90.28% where Kappa statistics was 0.86, while the
unsupervised classification result was 80.56% accurate with 0.73 Kappa statistics. In
conclusion, they found that the supervised classification technique appears more accurate
than the unsupervised classification.
Xiong Liu [24] used migrating means clustering unsupervised classification (MMC),
maximum likelihood classification (MLC) trained by picked training samples and trained by
the results of unsupervised classification (Hybrid Classification) to classify a 512 pixels by
512 lines NOAA-14 AVHRR Local Area Coverage (LAC) image. All the channels including
ch3 and ch3t are used in this project. The image is classified to six classes including water,
vegetation, thin partial clouds over ground, thin clouds, low/middle thick clouds and high
thick clouds plus unknown class for supervised classification. In total, the results using these
three methods are very consistent with the original three-band overlay color composite image
and the statistical mean vectors for each class are consistent using different methods and are
reasonable. He also noted that the ch3t temperature is usually much larger than the thermal
channel-measured temperature for clouds, the colder the thermal temperature, the larger their
difference. The ch3 reflectance is anti-correlated with the ch1 and ch2 reflectance, which is
due to that high reflectance ice clouds can absorb most of the energy in this channel. The
results of MMC and MLC trained by the results of MMC are better than that of the MMC
trained by picked samples. The MLC trained by picked samples produces more unknown
classes than that trained by MMC, which is probably due to that the standard deviation
(multivariate spreads) for each class generated by MMC is usually larger than that of picked
training samples. It takes more computation time to run MMC (5 iterations) than MLC if the
classes are the same, but take more time to pick samples over and over to get comparable
results. The results of MLC trained by picking samples are worse than the other two methods
due
to
the
difficulty of
picking
representative
training
samples.
The
hybrid
supervised/unsupervised classification combines the advantages of both supervised
classification and unsupervised classification. It doesn’t require the user have the
foreknowledge of each class, and can still consider the multivariate spreads and obtain
accurate mean vectors and covariance matrixes for each spectral class by using the entire
pixels image as training samples.
In Puerto Rico the land use has been changing, every day new developments (urban,
industrial, commercial and agricultural) are emerging. The purpose of Edwin Martínez
Martínez ‘s [25] work is to develop the land use of Río Jauca a sub-basin of the Rio Grande
de Arecibo watershed that is an important natural resource and supplies water to the
metropolitan area. Remote sensing techniques can be used to assess several water quality
parameters and also for land use classifications. For his work the ERDAS Imagine V8.5
computer software was used to develop a land use classification using IKONOS images. The
generated land use classification was compared with a land use generated using Arc View, to
decide which method provides better land use classification.
An unsupervised classification method provides the interpretation, feature extraction and
endmember estimation for the remote sensing image data without any prior knowledge of the
ground truth. Cheng-Yuan Liou [26] explored such method and constructed an algorithm
based on the non-negative matrix factorization (NMF). The use of the NMF is to match the
non-negative property in sensing spectrum data. The data dimensionality is estimated by
using the partitioned noise-adjusted principle component analysis (PNAPCA). The initial
matrix used to start the NMF is obtained by using the fuzzy c-mean (FCM). This algorithm is
capable to produce a region- or part-based representation of objects in images. Both
simulated and real sensing data are used to test the algorithm.
Traditionally, classification approaches have focused on per-pixel technologies. Pixels within
areas assumed to be automatically homogeneous are analyzed independently. These new
sources of high spatial resolution image will increase the amount of information attainable on
land cover. Significance is that the data can be acquired by our eyes and the energy can be
analyzed. But satellites are capable of collecting data beyond the visible band also. However,
the traditional methods of change detection are not suitable for high resolution remote
sensing images. To overcome the limitations of traditional pixel-level change detection of
high resolution remote sensing images, based on geo referencing and analysis method,
Babykalpana et al., [27] presented a unsullied way of multi-scale amalgamation for the high
resolution remote sensing images change detection. Their experiment showed that this
method has a stronger advantage than the traditional pixel-level method of high resolution
remote sensing image change detection. It is widely used for the crops classification in the
world and this classification method is used for land cover and land use because vegetation
components are important in the images. The basic axis is also to preserve the greenery of the
city for the healthy environment.
Ratika Pradhan et al., [28] developed the classification algorithm for remotely sensed satellite
data using Bayesian and hybrid classification approach. In their paper, a Bayesian classifier
and hybrid classifier algorithm for remotely sensed satellite data has been developed. To test
and validate the algorithm, the sample image taken into consideration is multi-spectral IRS1C/LISS III of East Sikkim, India. The proposed classifier can also be used for hyperspectral
remote sensing data considering the best bands as input for preparing spectral class
distribution. The sample image is classified by both Bayesian and Hybrid classifier and then
the overall accuracy were calculated. The overall accuracy for the sample test image was
found to be 90.53% using the Bayesian classifier and 91.59% using the Hybrid respectively.
The reason for high accuracy may be to some extent attributed for the reason that the part of
the training set is being considered as ground truths instead of actual data. Since the accuracy
of the results depends only upon the test set chosen, the efficiency of any algorithm shall not
be considered on the accuracy measure alone. The classified images shall also be compared
ground truth information physically. From the comparison, it is found that both the methods
are equally efficient.
2.1 Supervised Classification
The user selects representative samples for each land cover class in the digital image. These
sample land cover classes are called “training sites”. The image classification software uses
the training sites to identify the land cover classes in the entire image.
The classification of land cover is based on the spectral signature defined in the training set.
The digital image classification software determines each class on what it resembles most in
the training set. The common supervised classification algorithms are maximum likelihood
and minimum-distance classification.
Supervised Classification Steps:

Select training areas

Generate signature file

Classify
Nearest neighbor techniques are commonly used in remote sensing, pattern recognition and
statistics to classify objects into a predefined number of categories based on a given set of
predictors. These techniques are especially useful for highly nonlinear relationship between
the variables. In most studies the distance measure is adopted a priori. In contrast Luis
Samaniego et al., [29] proposed a general procedure to find an adaptive metric that combines
a local variance reducing technique and a linear embedding of the observation space into an
appropriate Euclidean space. To illustrate the application of this technique, two agricultural
land cover classifications using mono-temporal and multi-temporal Landsat scenes are
presented. The results of the study, compared with standard approaches used in remote
sensing such as maximum likelihood (ML) or k-Nearest Neighbor (k-NN) indicate substantial
improvement with regard to the overall accuracy and the cardinality of the calibration data
set. Also, using MNN in a soft/fuzzy classification framework demonstrated to be a very
useful tool in order to derive critical areas that need some further attention and investment
concerning additional calibration data.
A new approach to the classification of hyperspectral images is proposed. The main problem
with supervised methods is that the learning process heavily depends on the quality of the
training data set. In remote sensing, the training set is useful only for simultaneous images or
for images with the same classes taken under the same conditions; and, even worse, the
training set is frequently not available. On the other hand, unsupervised methods are not
sensitive to the number of labelled samples since they work on the whole image.
Nevertheless, relationship between clusters and classes is not ensured. In this context, L.
Gomez-Chova et al., [30] proposed a combined strategy of supervised and unsupervised
learning methods that avoids these drawbacks and automates the classification process. The
method is based on the general formulation of the expectation-maximization (EM) algorithm.
This method is applied to crop cover recognition of six hyperspectral images from the same
area acquired with the HyMap spectrometer during the DAISEX-99 campaign. For
classification purposes, six different classes are considered. Classification accuracy results
are compared to common methods: ISODATA, Learning Vector Quantization, Gaussian
Maximum Likelihood, Expectation-Maximization, and Neural Networks. The good
performance confirms the validity of the proposed approach in terms of accuracy and
robustness.
Benqin Song et al., [31] developed a new classification strategy that integrates sparse
representations and EMAPs for spatial–spectral classification of remote sensing data. Their
experiments reveal that the proposed approach, which combines the advantages of sparse
representation and the rich structural information provided by EMAPs, can appropriately
exploit the inherent sparsity present in EMAPs in order to provide state-of-the-art
classification results. This is mainly due to the fact that the samples in EMAP space can be
approximately represented by a few number of atoms in the training dictionary after solving
the optimization problem, whereas the same samples could not be represented in the original
spectral space with the same level of sparsity. The proposed strategy was tested on both
simulated and real multi/hyperspectral data sets. A comparison with state-of-the-art
classifiers shows very promising results for the proposed approach, particularly when a very
limited number of training samples are available. In this context, the SUnSAL algorithm
provided excellent classification performance as compared with other techniques.
Remote sensing image segmentation requires multi-category classification typically with
limited number of labeled training samples. While semi-supervised learning (SSL) has
emerged as a sub-field of machine learning to tackle the scarcity of labeled samples, most
SSL algorithms to date have had trade-offs in terms of scalability and/or applicability to
multi-categorical data. Ayse Naz Erkan et al., [32] evaluated semi-supervised logistic
regression (SLR), a recent information theoretic semi-supervised algorithm, for remote
sensing image classification problems. SLR is a probabilistic discriminative classifier and a
specific instance of the generalized maximum entropy framework with a convex loss
function. Moreover, the method is inherently multi-class and easy to implement. These
characteristics make SLR a strong alternative to the widely used semi-supervised variants of
SVM for the segmentation of remote sensing images. They demonstrated the competitiveness
of SLR in multispectral, hyperspectral and radar image classification.
Zhang Ming et al., [33] proposed to make a maximum use of remote sensing data and GIS
techniques to assess land use and soil classification in the rolling hilly area around Nanjing,
Eastern part of China (Sub – tropical region). Landsat MSS data perform a classification
using the ILWIS software. The supervised classification was based upon a multispectral
analysis and on “ground truth”. The minimum distance to mean classifier, the maximum
likelihood classifier and the box classification were used. According to the reflectance
characteristics of the surface material, the soil classification indicated 7 classes that belong to
4 units and 10 subunits. The land use map contains 5 classes. It has been proved that within
limitations, classification algorithms and threshold parameters have an important influence on
the classification result and should be selected carefully based on the training area.
2.2 Classification using Genetic Algorithm
Taking Jiading district of Shanghai as the study area, which is one of the most typical
urbanization fringe regions in Shanghai during the past decades, the urban sprawl in this
region from 1989 to 2006 was studied on the basis of multi-spectral remotely sensed images.
Multi-source data including four epochs of representative TM images (1989, 1995, 2001 and
2006) and the vector topographic map were used in our study. A genetic algorithm optimized
back propagation neutral network approach was first proposed by X. Zhang et al., [34] to
classify sorts of land use types from the four epochs of remotely sensed images. The accuracy
of the classification was assessed for the four remotely sensed images. The urban land use
type was thus extracted based on the land use classification result and three urban land use
change images were correspondingly derived from the extracted urban land use type. Based
on this, the study area was divided into four plates: the central plate, the southern plate, the
northern plate and the western plate, and the detailed temporal-spatial urban land use changes
in the study area were investigated in each plate by using overlapping pixel comparison based
change detection method during the three time intervals: 1989-1995, 1955-2001 and 2001-
2006. From the change detection analysis, the process and pattern of the urban land use
change in the Jiading district was finally revealed during nearly 20 years in our study.
The use of remote sensing images as a source of information in agribusiness applications is
very common. In those applications, it is fundamental to know how the space occupation is.
However, identification and recognition of crop regions in remote sensing images are not
trivial tasks yet. Although there are automatic methods proposed to that, users very often
prefer to identify regions manually. That happens because these methods are usually
developed to solve specific problems, or, when they are of general purpose, they do not yield
satisfying results. Jefersson Alex dos Santos [35] presented a new interactive approach based
on relevance feedback to recognize regions of remote sensing. Relevance feedback is a
technique used in content-based image retrieval (CBIR) tasks. Its objective is to aggregate
user preferences to the search process. The proposed solution combines the Optimum-Path
Forest (OPF) classifier with composite descriptors obtained by a Genetic Programming (GP)
framework. The new approach has presented good results with respect to the identification of
pasture and coffee crops, overcoming the results obtained by a recently proposed method and
the traditional Maximum Likelihood algorithm.
Support vector machine (SVM) is originally developed for linear two-class classification via
constructing an optimal separating hyper plane, where the margin is maximal. In case of not
linearly separable training data, SVM is by means of kernel trick to map the original input
space into a high dimensional feature space to enhance the classifier generalization ability.
Genetic Algorithm (GA) is a stochastic and heuristic searching algorithm that is inspired by
natural evolution. By using GA along with SVM here K. Moje Ravindra et al., [36] tried to
make classification of the objects such that it will be closer to the original image. Image
features are important to be considered since the images are retrieved based on these features.
The extraction of suitable features from the image is the basic step by which the query image
and the database images can be compared. The commonest features in an image are color,
shape and texture. Generally, in the classification of object-based high resolution remote
sensing images, numerous object features, such as spectral, texture, shape and contextual, are
calculated after segmentation. However, the determination of the most appropriate feature
subsets not only degrades the computational complexity, also can obtain a higher
classification rate. By using SVM classifier we can classify the objects more accurately when
we use a Non-parametric classifiers. A feature extraction from invisible bands for
hyperspectral image analysis, where there is no input to the human perception system at all.
The perception-centered approach has the obvious advantages in serving the “master” or
“user” of the image retrieval system human, given that matching human performance is the
ultimate goal of the system. They concluded that, whenever there is confusion occurs near the
boundary, genetic parameter for image classification is best.
CHAPTER 3
PREPROCESSING
According to Mather (1999) the term pre-processing is understood as the correction of
geometric and radiometric deficiencies and the removal of data errors. It seems natural that
errors within the data are removed, if possible, before image interpretations start. The choice
of methods to do so is always purpose dependent. If, for instance, a check of a certain land
cover or object with an satellite image is the purpose, visual interpretation might be sufficient
and even geometric correction not necessary (Jensen, 1996b). In our opinion the operator
should define precisely the demands on the data and chose the necessary processing steps to
achieve his specific task. The importance of pre-processing methods becomes obvious in
change detection or monitoring applications, where the operator must be able to distinguish
data noise, pre-processing and data handling errors from real changes.
3.1 Preprocessing in Image Processing
Preprocessing images commonly involves removing low-frequency background noise,
normalizing the intensity of the individual particles of images, removing, reflections, and
masking portions of images. Image preprocessing [18] is the technique of enhancing data
images prior to computational processing. It is a common name for operations with images at
the lowest level of abstraction -- both input and output are intensity images. The aim of preprocessing is an improvement of the image data that suppresses unwanted distortions or
enhances some image features important for further processing. Four categories of image preprocessing methods according to the size of the pixel neighborhood that is used for the
calculation of new pixel brightness:

Pixel brightness transformations.

Geometric transformations.

Pre-processing methods that use a local neighborhood of the processed pixel.

Image restoration that requires knowledge about the entire image.
Image pre-processing methods use the considerable redundancy in images.
Neighboring pixels corresponding to one object in real images have essentially the same or
similar brightness value. Thus, distorted pixel can often be restored as an average value of
neighboring pixels. One example of this is filtering impulse noise. If pre-processing aims to
correct some degradation in the image, the nature of a priori information is important:
knowledge about the nature of the degradation; knowledge about the properties of the image
acquisition device, and conditions under which the image was obtained. The nature of noise
(usually its spectral characteristics) is sometimes known. The Knowledge about the objects is
searched in the image. If knowledge about objects is not available in advance it can be
estimated during the processing.
Pixel brightness transformations
Brightness transformations modify pixel brightness -- the transformation depends on the
properties of a pixel itself. Two classes of brightness corrections: position dependent and
gray-scale transformations.

Position dependent brightness correction considers original brightness and the pixel
position in the image.

Gray scale transformations change brightness without regard to position in the image.
Geometric transformations
Geometric transforms permit the elimination of geometric distortion that occurs when
an image is captured. An example is an attempt to match remotely sensed images of the same
area taken after one year, when the more recent image was not taken from precisely the same
position. To inspect changes over the year, it is necessary first to execute a geometric
transformation, and then subtract one image from the other. A geometric transform is a vector
function T that maps the pixel (x,y) to a new position (x',y').
Figure 3.1 Geometric Transform on a Plane
The transformation equations are either known in advance or can be determined from
known original and transformed images. Several pixels in both images with known
correspondence are used to derive the unknown transformation.
A geometric transform consists of two basic steps

Determine the pixel co-ordinate transformation. Map the co-ordinates of the input
image pixel to the point in the output image. The output point co-ordinates should be
computed as continuous values (real numbers) as the position does not necessarily
match the digital grid after the transform.

Finding the point in the digital raster which matches the transformed point and
determining its brightness. Brightness is usually computed as an interpolation of the
brightness of several points in the neighborhood
Typical geometric distortions which have to be overcome in remote sensing:
distortion of the optical systems nonlinearities in row by row scanning and non constant
sampling period.
Pixel co-ordinate transformations
General case of finding the co-ordinates of a point in the output image after a
geometric transform, usually approximated by a polynomial equation
This transform is linear with respect to the coefficients ark, b rk. If enough pairs of
corresponding points (x,y), (x',y') in both images are known, it is possible to determine ark, b rk
by solving a set of linear equations. More points than coefficients are usually used to improve
the estimate of the coefficients. If the geometric transform does not change rapidly depending
on position in the image, low order approximating polynomials, m=2 or m=3, are used,
needing at least 6 or 10 pairs of corresponding points. The corresponding points should be
distributed in the image in a way that can express the geometric transformation -usually they
are spread uniformly. The higher the degree of the approximating polynomial, the more
sensitive to the distribution of the pairs of corresponding points the geometric transform. In
practice, the geometric transform is often approximated by the bilinear transformation 4 pairs
of corresponding points are sufficient to find transformation coefficients
Even simpler is the affine transformation for which three pairs of corresponding
points are sufficient to find the coefficients
Brightness interpolation
Assume that the planar transformation has been accomplished, and new point coordinates (x',y') were obtained. The position of the point does not in general fit the discrete
raster of the output image. Values on the integer grid are needed. Each pixel value in the
output image raster can be obtained by interpolating the some neighboring non-integer
samples. The brightness interpolation problem is usually expressed in a dual way (by
determining the brightness of the original point in the input image that corresponds to the
point in the output image lying on the discrete raster).
Local pre-processing
Pre-processing methods use a small neighborhood of a pixel in an input image to get a
new brightness value in the output image. Such pre-processing operations are also called
filtration. Local pre-processing methods can be divided into the two groups according to the
goal of the processing:

Smoothing suppresses noise or other small fluctuations in the image; equivalent to
the suppression of high frequencies in the frequency domain. Unfortunately,
smoothing also blurs all sharp edges that bear important information about the image.
 Gradient operators are based on local derivatives of the image function. Derivatives
are bigger at locations of the image where the image function undergoes rapid
changes. The aim of gradient operators is to indicate such locations in the image.
Gradient operators suppress low frequencies in the frequency domain (i.e. they act as
high-pass filters). Noise is often high frequency in nature; unfortunately, if a gradient
operator is applied to an image the noise level increases simultaneously.
Clearly, smoothing and gradient operators have conflicting aims. Some preprocessing algorithms solve this problem and permit smoothing and edge enhancement
simultaneously.
Averaging using a rotating mask
Avoids edge blurring by searching for the homogeneous part of the current pixel
neighborhood, the resulting image is in fact sharpened brightness average is calculated only
within the homogeneous region a brightness dispersion sigma^2 is used as the region
homogeneity measure. Let n be the number of pixels in a region R and g(i,j) be the input
image. Dispersion sigma^2 is calculated as
The computational complexity (number of multiplications) of the dispersion
calculation can be reduced if expressed as follows
Median smoothing
In a set of ordered values, the median is the central value. Median filtering reduces
blurring of edges. The idea is to replace the current point in the image by the median of the
brightness in its neighborhood. The advantages of median filtering are that it is not affected
by individual noise spikes , eliminates impulsive noise quite well, and it does not blur edges
much and can be applied iteratively. The main disadvantage of median filtering in a
rectangular neighborhood is its damaging of thin lines and sharp corners in the image -- this
can be avoided if another shape of neighborhood is used.
Edge detectors
Edges are pixels where brightness changes abruptly. Calculus describes changes of
continuous functions using derivatives; an image function depends on two variables - partial
derivatives. A change of the image function can be described by a gradient that points in the
direction of the largest growth of the image function. An edge is a property attached to an
individual pixel and is calculated from the image function behavior in a neighborhood of the
pixel. It is a vector variable with magnitude and direction:
The gradient magnitude and gradient direction are continuous image functions where arg(x,y)
is the angle (in radians) from the x-axis to the point (x,y). The gradient direction gives the
direction of maximal growth of the function, e.g., from black (f(i,j)=0) to white (f(i,j)=255).
There are three classes of gradient operators:

Operators which approximate derivatives of the image using differences such as the
Laplacian, Roberts or Prewitt operators.

Operators based on the zero-crossings of the image function second derivative such as
Marr-Hildreth or Canny edge detectors.
 Operators which match an image to a parametric model of the edges.
Laplace operator
The Laplace operator is a very popular operator approximating the second derivative
which gives the gradient magnitude only. The Laplacian is approximated in digital images by
a convolution sum. A 3 x 3 mask for 4-neighborhoods and 8-neighborhood
The Laplacian operator has a disadvantage -- it responds doubly to some edges in the image.
Canny edge detection
Optimal for step edges corrupted by white noise. Optimality related to three criteria.

Detection criterion ... important edges should not be missed; there should be no
spurious responses.

Localization criterion ... distances between the actual and located position of the edge
should be minimal.

One response criterion ... minimizes multiple responses to a single edge (also partly
covered by the first criterion since when there are two responses to a single edge one
of them should be considered as false)
Other local pre-processing operators
Several other local operations exist which are used for different purposes: Line
finding, line thinning, line filling and interest point operators are among them.
Adaptive neighboring pre-processing
The majority of pre-processing operators work in neighborhoods of fixed sizes in the
whole image, of which square windows (3x3, 5x5 or 7x7) are most common. Pre-processing
operators of variable sizes and shapes exist and bring improved pre-processing results. They
are based on detection of the most homogeneous neighborhood of each pixel. However they
are not widely used, mostly because of computational demands and the lack of a unifying
approach. A recent approach to image pre-processing introduces the concept of an adaptive
neighborhood which is determined for each image pixel. The neighborhood size and shape
are depending on the image data and on parameters which define measures of homogeneity of
a pixel neighborhood. A significant property of the neighborhood for each pixel is the ability
to self tune to contextual details in the image.
Image restoration
Image restoration - suppressing image degradation using knowledge about its nature.
Most image restoration methods are based on convolution applied globally to the whole
image.
Degradation causes:

Defects of optical lenses.

Nonlinearity of the electro-optical sensor.

Graininess of the film material.

Relative motion between an object and camera.

Wrong focus.

Atmospheric turbulence in remote sensing or astronomy.
The objective of image restoration is to reconstruct the original image from its degraded
version. Image restoration techniques - two groups:

Deterministic methods - applicable to images with little noise and a known
degradation function. The original image is obtained from the degraded one by a
transformation inverse to the degradation.

Stochastic techniques - the best restoration is sought according to some stochastic
criterion, e.g., a least squares method. In some cases the degradation transformation
must be estimated first.
3.2 Median Filters
There are many preprocessing techniques which are discussed above. In this paper,
the preprocessing is done by removing noise using median filters. The best-known orderstatistic filter is the median filter [6], which, as its name implies, replaces the value of a pixel
by the median of the intensity levels in the neighborhood of that pixel:
(g,t)} (s,t)
Sxy
The value of the pixel at (x, y) is included in the computation of the median. Median
filters are quite popular because, for certain type of random noise, they provide excellent
noise-reduction capabilities, with considerably less blurring than linear smoothing filters of
similar size. Median filters are particularly effective in the presence of both bipolar and
unipolar impulse noise. Median filtering [19] is a nonlinear process useful in reducing
impulsive, or salt-and-pepper noise. It is also useful in preserving edges in an image while
reducing random noise. Impulsive or salt-and pepper noise can occur due to a random bit
error in a communication channel. In a median filter, a window slides along the image, and
the median intensity value of the pixels within the window becomes the output intensity of
the pixel being processed. Like lowpass filtering, median filtering smoothes the image and is
thus useful in reducing noise. Unlike lowpass filtering, median filtering can preserve
discontinuities in a step function and can smooth a few pixels whose values differ
significantly from their surroundings without affecting the other pixels. An important
parameter in using a median filter is the size of the window. The choice of the window size
depends on the context. Because it is difficult to choose the optimum window size in
advance, it may be useful to try several median filters of different window sizes and choose
the best of the resulting images.
In median filtering [20], the neighboring pixels are ranked according to brightness
(intensity) and the median value becomes the new value for the central pixel. It can do an
excellent job of rejecting certain types of noise, in particular, “shot” or impulse noise in
which some individual pixels have extreme values. In its operation, the pixel values in the
neighborhood window are ranked according to intensity, and the middle value (the median)
becomes the output value for the pixel under evaluation.
Figure 3.2 Median Filtering Operations
In particular, compared to the smoothing filters examined thus far, median filters offer
three advantages:

No reduction in contrast across steps, since output values available consist only of
those present in the neighborhood (no averages).

Median filtering does not shift boundaries, as can happen with conventional
smoothing filters (a contrast dependent problem).

Since the median is less sensitive than the mean to extreme values (outliers), those
extreme values are more effectively removed.
The median is, in a sense, a more robust “average” than the mean, as it is not affected by
outliers (extreme values). Since the output pixel value is one of the neighboring values, new
“unrealistic” values are not created near edges. Since edges are minimally degraded, median
filters can be applied repeatedly, if necessary.
Advantages

It is simple to understand.

The median filter preserves brightness differences resulting in minimal blurring of
regional boundaries.

Preserves the positions of boundaries in an image, making this method useful for
visual examination and measurement.

Median computer algorithm can be customized.

We can repeat the median filter on the image until there are no further changes. In this
way the median filter works like a maximum expectation restoration.
Chapter 4
Land Cover Classification using GA based Fuzzy Clustering
Techniques for Remotely Sensed Data
4.1 Introduction
A digital remotely sensed image is typically composed of picture elements (pixels)
located at the intersection of each row i and column j in each K bands of imagery. Associated
with each pixel is a number known as Digital Number (DN) or Brightness Value (BV) that
depicts the average radiance of a relatively small area within a scene. A smaller number
indicates low average radiance from the area and the high number is an indicator of high
radiant properties of the area. The size of this area effects the reproduction of details within
the scene. As pixel size is reduced more scene detail is presented in digital representation.
While displaying the different bands of a multispectral data set, images obtained in different
bands is displayed in image planes (other than their own) the color composite is regarded as
False Color Composite (FCC). High spectral resolution is important when producing color
components. For a true color composite an image data used in red, green and blue spectral
region must be assigned bits of red, green and blue image processor frame buffer memory. A
color infrared composite ‘standard false color composite’ is displayed by placing the infrared,
red, green in the red, green and blue frame buffer memory (Fig. 2). In this healthy vegetation
shows up in shades of red because vegetation absorbs most of green and red energy but
reflects approximately half of incident Infrared energy. Urban areas reflect equal portions of
NIR, R & G, and therefore they appear as steel grey.
Geometric distortions manifest themselves as errors in the position of a pixel relative to
other pixels in the scene and with respect to their absolute position within some defined map
projection. If left uncorrected, these geometric distortions render any data extracted from the
image useless. This is particularly so if the information is to be compared to other data sets, is
it from another image or a GIS data set. Distortions occur for many reasons. Rectification is a
process of geometrically correcting an image so that it can be represented on a planar surface,
conform to other images or conform to a map. That is, it is the process by which geometry of
an image is made plan metric. It is necessary when accurate area, distance and direction
measurements are required to be made from the imagery. It is achieved by transforming the
data from one grid system into another grid system using a geometric transformation. Ground
Control Points (GCP) are the specific pixels in the input image for which the output map
coordinates are known. By using more points than necessary to solve the transformation
equations a least squares solution may be found that minimizes the sum of the squares of the
errors. Care should be exercised when selecting ground control points as their number,
quality and distribution affect the result of the rectification.
Remote sensing can be defined as any process whereby information is gathered about an
object, area or phenomenon without being in contact with it. The output of a remote sensing
system is usually an image representing the scene being observed. Image classification is an
important part of the remote sensing, image analysis and pattern recognition. In some
instances, the classification itself may be the object of the analysis. The image classification
therefore forms an important tool for examination of the digital images. Classification
strategies are basically divided into three. Supervised Classification techniques require
training areas to be defined by the analyst in order to determine the characteristics of each
category. Unsupervised Classification searches for natural groups of pixels, called clusters,
present within the data by means of assessing the relative locations of the pixels in the feature
space. Hybrid Classification takes the advantage of both the supervised classification and
unsupervised classification.
4.2 Cluster Analysis
Cluster analysis or clustering is the task of grouping a set of objects in such a way that
objects in the same group (called a cluster) are more similar (in some sense or another) to
each other than to those in other groups (clusters). It is a main task of exploratory data
mining, and a common technique for statistical data analysis, used in many fields,
including machine learning ,pattern recognition, image analysis, information retrieval,
and bioinformatics. Cluster analysis itself is not one specific algorithm, but the general task to
be solved. It can be achieved by various algorithms that differ significantly in their notion of
what constitutes a cluster and how to efficiently find them. Popular notions of clusters
include groups with small distances among the cluster members, dense areas of the data
space, intervals or particular statistical distributions. Clustering can therefore be formulated
as a multi-objective optimization problem. The appropriate clustering algorithm and
parameter settings (including values such as the distance function to use, a density threshold
or the number of expected clusters) depend on the individual data set and intended use of the
results. Cluster analysis as such is not an automatic task, but an iterative process of
knowledge discovery or interactive multi-objective optimization that involves trial and
failure. It will often be necessary to modify data preprocessing and model parameters until
the result achieves the desired properties.
Clustering analysis is classifying samples according to their similarity by means of
unsupervised training. It makes the samples, which have greater similarity, as a class, and
occupies the partial area of feature space. The clustering center of each partial area is
respectively acting as a representative of the corresponding type. There are various types of
clustering. One of its types is Partitional Clustering. Based on Partitional Clustering, many
algorithms are used. One of the familiar algorithms is the FuzzyCmeans clustering algorithm.
4.3 Median Filters
The median filter is a nonlinear digital filtering technique, often used to remove noise.
Such noise reduction is a typical pre-processing step to improve the results of later
processing. Median filtering is very widely used in digital image processing because, under
certain conditions, it preserves edges while removing noise. The main idea of the median
filter is to run through the signal entry by entry, replacing each entry with the median of
neighboring entries. The pattern of neighbors is called the "window", which slides, entry by
entry, over the entire signal. For 1D signal, the most obvious window is just the first few
preceding and following entries, whereas for 2D (or higher-dimensional) signals such as
images, more complex window patterns are possible (such as "box" or "cross" patterns). Note
that if the window has an odd number of entries, then the median is simple to define: it is just
the middle value after all the entries in the window are sorted numerically.
Given a set of random variables X=(X1,X2,…XN) , the order statistics X(1) ≤
X(2)≤…..X(N) are random variables, defined by sorting the values of Xi in an increasing order.
The median value is then given as
Where m=2K+1 is the median rank. The median is considered to be a robust estimator of the
location parameter of a distribution and has found numerous applications in smoothing and
denoising, especially for signals contaminated by impulsive noise. For a grayscale input
image with intensity values xi,j , the two-dimensional median filter is defined as
Where W is a window over which the filter is applied.
Median Filter Algorithm

The Median Filter is performed by taking the magnitude of all of the vectors within a
mask and sorted according to the magnitudes.

The pixel with the median magnitude is then used to replace the pixel studied.

The Simple Median Filter has an advantage over the Mean filter since median of the
data is taken instead of the mean of an image.

The pixel with the median magnitude is then used to replace the pixel studied.

The median of a set is more robust with respect to the presence of noise. The median
filter is given by
Median filter(x1….xN) =Median (||x1 ||2…….||xN||2).

When filtering using the Simple Median Filter, an original and the resulting filtered
pixel of the sample have the same pixel.

A pixel that does not change due to filtering is known as the root of the mask.
4.4 Fuzzy c means
The concept of fuzzy partition is essential for cluster analysis, and consequently also
for the identification techniques that are based on fuzzy clustering. Fuzzy and possibility
partitions can be seen as a generalization of hard partition which is formulated in terms of
classical subsets. Generalization of the hard partition to the fuzzy case follows directly by
allowing µ ik to attain real values in [0, 1]. Conditions for a fuzzy partition matrix are given by
(Ruspini, 1970):
The ith row of the fuzzy partition matrix U contains values of the ith membership function of
the fuzzy subset Ai of Z. The second Equation constrains the sum of each column to 1, and
thus the total membership of each z k in Z equals one. The fuzzy partitioning space for Z is
the set
Most analytical fuzzy clustering algorithms (and also all the algorithms presented in this
chapter) are based on optimization of the basic c-means objective function, or some
modification of it. Hence we start our discussion with presenting the fuzzy cmeans
functional. A large family of fuzzy clustering algorithms is based on minimization of the
fuzzy c-means functional formulated as (Dunn, 1974; Bezdek, 1981):
Where
is a fuzzy partition matrix of Z,
is a vector of cluster prototypes (centers), which have to be determined,
is a squared inner-product distance norm, and
is a parameter which determines the fuzziness of the resulting clusters. The value of the cost
function can be seen as a measure of the total variance of z k from vi.
The algorithm steps are as follows

Select m (m > 1); initialize the membership function values µij, i = 1, 2, ... , n;
j = 1, 2, . . . , c.

Compute the cluster centers zj,
j = 1, 2, ... , c

Compute Euclidian distance dij,
i = 1, 2, ... , n; j = 1, 2, ... , c.

Update the membership functionµij, i = 1, 2, ... , n; j = 1, 2, ... , c

If not converged, go to step 2.
Several stopping rules can be used. One is to terminate the algorithm when the relative
change in the centroid values becomes small or when the objective function cannot be
minimized more. The FCM algorithm is sensitive to initial values and it is likely to fall into
local optima.
4.5 Genetic Algorithm
American scholar, J. Holland, first raised the Genetic Algorithm (GA) concept in 1975. It
is based on “survival of the fittest” in Darwin’s theory of evolution. The basic genetic
operations, which are repetitively utilized for the groups possibly containing solution, make
the new groups generated then make them evolved constantly. At the same time, the
optimization individuals in optimized groups are searched based on the global parallel search
technique so as to obtain the global optimum solution fulfilled demands. GA generates
valuable solutions for hard optimization problems using techniques that are inspired by
natural evolutionary operators such as inheritance, mutation, selection, and crossover.
In a genetic algorithm, a population of strings (called chromosomes or the genotype of the
genome), which encode candidate solutions (called individuals, creatures, or phenotypes) to
an optimization problem, evolves toward better solutions. Traditionally, solutions are
represented in binary as strings of 0s and 1s, but other encodings are also possible. The
evolution usually starts from a population of randomly generated individuals and happens in
generations. In each generation, the fitness of every individual in the population is evaluated,
multiple individuals are stochastically selected from the current population (based on their
fitness), and modified (recombined and possibly randomly mutated) to form a new
population. The new population is then used in the next iteration of the algorithm.
Commonly, the algorithm terminates when either a maximum number of generations has
been produced, or a satisfactory fitness level has been reached for the population. If the
algorithm has terminated due to a maximum number of generations, a satisfactory solution
may or may not have been reached. A typical genetic algorithm requires:

All solutions should have a genetic representation (in a shape of chromosome).

There should be a fitness function to assess the solutions.
Chromosomes
The chromosomes represent set of genes, which code the independent variables. Every
chromosome represents a solution of the given problem. Individual and vector of variables
will be used as other words for chromosomes. A set of different chromosomes (individuals)
forms a generation. By means of evolutionary operators, like selection, recombination and
mutation an offspring population is created.
Selection
The selection of the best individuals is based on an evaluation of fitness function or
fitness functions. Fitness value is a quality measurement of each solution. Better fitness
values belong to better individuals in each population. When termination criteria are satisfied,
algorithm reaches to better fitness value. In the final generation, a solution with better fitness
value among others is found as the desired solution.
Crossover
The first step in the reproduction process is the recombination (crossover). In it the genes
of the parents are used to form an entirely new chromosome. The typical recombination for
the GA is an operation requiring two parents, but schemes with more parent area also
possible. Two of the most widely used algorithms are Conventional (Scattered) Crossover
and Blending (Intermediate) Crossover.
Mutation
The newly created by means of selection and crossover population can be further applied
to mutation. Mutation means random change of the value of a gene in the population.

Generate the initial population.

Calculate the values of the function that we want to minimize or maximize.

Check for termination of the algorithm.

Selection is done between all individuals in the current population are chose those,
who will continue and by means of crossover and mutation will produce offspring
population

Crossover – the individuals chosen by selection recombine with each other and new
individuals will be created. The aim is to get offspring individuals that inherit the best
possible combination of the characteristics (genes) of their parents.

Mutation – by means of random change of some of the genes, it is guaranteed that
even if none of the individuals contain the necessary gene value for the extreme, it is
still possible to reach the extreme.

New generation – the best individuals chosen from the selection are combined with
those who passed the crossover and mutation, and form the next generation.
Figure4.1. The GA Flowchart
4.6 Proposed Work
Land cover is an important component in understanding the interactions of the human
activities with the environment and thus it is necessary to be able to simulate changes. This
proposal aims at developing a novel land cover clustering method using GA based clustering
techniques.
The proposed method has two phases: the first step computes a refined starting condition
from a given initial one that is based on an efficient technique for estimating the modes of a
distribution. The refined initial starting condition allows the iterative algorithm to converge to
a “better” local minimum. And in the second step, a novel method has been proposed to
improve to cluster quality by GA based refinement algorithm.
Input Image
Pre Processing
Fuzzy C-means Clustering
GA based Tuning
Ground truth verification
Classified map
Figure 4.2: Methodology
Genetic algorithm (GA) is randomized search and optimization techniques guided by the
principles of evolution and natural genetics, having a large amount of implicit parallelism.
The basic reason for the improvement is, in any clustering algorithm the obtained clusters
will never give us 100% quality. There will be some errors known as misclustered. That is, a
data item can be wrongly clustered. These kinds of errors can be avoided by using the
improvement algorithm.
The genetic algorithms based clustering may not be able to handle large amounts of data.
The Fuzzy C-means algorithm does not lend itself well to adaptive clustering. And an
important point is, so far, the researchers are not contributed to improve the cluster quality
after grouping. In this proposed method, a new framework has been introduced to improve
the cluster quality from Fuzzy C-means algorithm. The proposed algorithm is applied to the
remotely sensed data (Survey of India toposheets and IRS-1C satellite imageries) of Theni
region.
Preprocessing
Satellite images cannot be given directly as the input for the proposed technique. Thus, it
is indispensable to perform pre-processing on the input image, so that the image gets
transformed to be relevant for the further processing. In proposed technique, A Median filter
which is a non linear filter is used in the R, G and B layers for filtering noise. It is used
because, under certain conditions, it preserves edges while removing noise.
GA Based Clustering
A major problem with Fuzzy C-means algorithm is that it is sensitive to the selection of
initial partition and may converge to a local minimum of variation if the initial partition is not
properly chosen. So in the proposed method, we estimate the mode value as an initial
partition.
Performance optimization using genetic algorithms is given by a sequence of steps, which
are:
1. Generate initial population.
2. Evaluate population
3. Selection.
4. Crossover.
5. Mutation.
6. Reinsertion of new individuals to the population.
From step 2 to step 6, it performs an iterative process until a stopping criterion is met, in
Fig. 4.3 we can see the Scheme of GA for optimization of the Fuzzy C-Means algorithm
(FCM). In this figure we can observe that population evaluation is done by FCM algorithm,
but for us to know how good some individuals need something that does not indicate the
fitness of these, to measure aptitude of individuals evaluated by FCM, we use the proposed
validation index mentioned in section III. Individuals evaluated by the FCM algorithm, are
formed only by two parameters which are the number of clusters and the exponent of weight.
Figure 4.3. FuzzyCMeans with GA
Results
The proposed algorithm is applied to the remotely sensed data (Survey of India
toposheets and IRS-1C satellite imageries) of Theni region. Two Theni region images are
used to implement the proposed algorithm. The first image is 1152x1152 size tiff image. The
second image is 1153x1153 size tiff image. Both images are color image. The Original and
the clustered images are shown.
Figure4. 4. Original & Clustered Theni Region Image1
Figure 4.5. Original & Clustered Theni Region Image2
Even though the visual comparison gives detailed information for Fuzzy C-means
clustering, to further evaluate the performance of proposed work the accuracy assessment has
been done. The confusion matrix in terms of pixels and percentage is given in Table 4.1 and
Table 4. 2. The overall classification accuracy is 96.04%.
Table 4.1. Confusion Matrix (Pixels)
Class
Urban
Vegetation
Hilly
Region
Total
Urban
Vegetation
2239
20
21
1680
4
100
22642
1800
7
90
1963
2060
2266
1791
2067
6124
Hilly Region
Total
From the Confusion Matrix, it is clear that urban yields a maximum classification accuracy of
98.81% when compared to Vegetation and Hilly Region.
Table 4.2. Confusion Matrix (Percentage)
Class
Urban
Vegetation
1.17
93.80
5.03
Hilly
Region
0.19
4.84
94.97
Urban
Vegetation
Hilly
Region
Total
98.81
0.88
0.31
100
Total
36.97
29.39
33.64
100
100
100
Table 4.3 gives the producer and user accuracy for individual classes. By reducing the
misclassification between the Vegetation and Hilly region the overall accuracy can be further
improved.
Table 4.3. Accuracy Assessment
Class
Producer
accuracy
(%)
User Accuracy
(%)
Urban
Vegetation
Hilly Region
98.81
93.80
94.97
98.90
93.33
95.29
Chapter 5
Conclusion
Remote sensing can be defined as any process whereby information is gathered about an
object, area or phenomenon without being in contact with it. The output of a remote sensing
system is usually an image representing the scene being observed. Image classification is an
important part of the remote sensing, image analysis and pattern recognition. In some
instances, the classification itself may be the object of the analysis. The image classification
therefore forms an important tool for examination of the digital images. Classification
strategies are basically divided into three. Supervised Classification techniques require
training areas to be defined by the analyst in order to determine the characteristics of each
category. Unsupervised Classification searches for natural groups of pixels, called clusters,
present within the data by means of assessing the relative locations of the pixels in the feature
space. Hybrid Classification takes the advantage of both the supervised classification and
unsupervised classification.
Land cover is an important component in understanding the interactions of the human
activities with the environment and thus it is necessary to be able to simulate changes. It is
essential component where in other parameters are integrated on the requirement basis to
derive various developmental index for earth resources. Digital classification methods of
remotely sensed images have acquired a growing importance in the automatic recognition of
the land cover patterns. Particularly, unsupervised classification methods have traditionally
been considered as an important approach for the interpretation of remotely sensed images.
Unsupervised classification is frequently performed through clustering methods. These
methods examine the unknown pixels in an image and incorporate them into a set of classes
defined through the natural clusters of the gray levels of the pixels.
Cluster analysis provides a practical method for organizing a large set of data so that the
retrieval of information may be made more efficiently. However, although there is a large
quantity of different clustering methods in the pattern recognition area, only a limited
quantity of them can be used in remote sensing applications. However there are two major
limitations that exist in these methods. The first is that a predefined number of clusters must
be given in advance. The second is that the FCM technique can get stuck in sub-optimal
solutions. Recent attempts have adapted the K-means clustering algorithm as well as genetic
algorithms based on rough sets to find interval sets of clusters. And an important point is, so
far, the researchers haven’t contributed to improve the cluster quality once it is clustered.
This method enables the clustering to be performed by taking the initial centroid using mode
function which allows the iterative algorithm to converge to a “better” local minimum. Then
the GA based refinement algorithm to improve the cluster quality. The expected outcome of
this work will be the Land use land cover map, of the study area namely Theni region,
TamilNadu. The proposed method has two phases: the first step computes a refined starting
condition from a given initial one that is based on an efficient technique for estimating the
modes of a distribution. The developed initial starting condition allows the iterative algorithm
to converge to a “better” local minimum. And in the second step, a novel method has been
proposed to improve to cluster quality by GA based improvement algorithm. Even though the
visual comparison gives detailed information for Fuzzy C-means clustering, to further
evaluate the performance of proposed work the accuracy assessment has been done.
REFERENCES
[1].
Jiawei Han and M Kamber , “Data Mining: Concepts and Techniques”, Second
Edition.
[2].
Two Crows Corporation, “Introduction to Data Mining and Knowledge Discovery”,
Third Edition.
[3].
Dr. Rajni Jain, “Introduction to DataMining Techniques”.
[4].
Pasi Franti, “Clustering Methods”.
[5].
T.Soni Madhulatha, “An Overview on Clustering Methods”, IOSR Journal of
Engineering Apr. 2012, Vol. 2(4) pp: 719-725.
[6].
Rafael C.Gonzalez & Richard E.Woods,‘Digital Image Processing’, Third Edition.
[7].
http://en.wikipedia.org/wiki/Image_processing
[8].
B.G.Obula Reddy, Dr. Maligela Ussenaiah, “Literature Survey on Clustering
Techniques”, IOSR Journal of Computer Engineering.
[9].
M. Kuchaki Rafsanjani, Z. Asghari Varzaneh, N. Emami Chukanlo , “A survey of
hierarchical clustering algorithms”, TJMCS Vol .5 No.3 (2012) 229-240 .
[10].
P. IndiraPriya & Dr. D.K. Ghosh, “A Survey on Different Clustering Algorithms in
DataMining Technique”, International Journal of Modern Engineering Research,
Vol.3, Issue.1, Jan-Feb. 2013 pp-267-274.
[11].
Daniel T. Larose, “Discovering Knowledge in Data”.
[12].
Periklis Andritsos, “Data Clustering Techniques”.
[13].
Ilango & Dr. V. Mohan, “A Survey of Grid Based Clustering Algorithm”,
International Journal of Engineering Science & Technology, Vol. 2(8), 2010, 34413446.
[14].
B. Eshwar Reddy, M. Veeresha & Nagaraja Ra, “Image Processing: A Survey”.
[15].
[16].
[17].
[18].
https://www.mtholyoke.edu/courses/tmillett/course/geog205/files/remote_sensing.pdf
Melanie Mitchell, “ An Introduction to Genetic Algorithm”.
http://www.cs.cmu.edu/~02317/slides/lec_9.pdf
http://www.eng.iastate.edu/ee528/sonkamaterial/chapter_4.htm#4.1
[19].
http://nptel.iitk.ac.in/courses/Webcourse-contents/IITKANPUR/Digi_Img_Pro/chapter_8/8_16.html
[20].
http://medim.sth.kth.se/6l2872/F/F7-1.pdf
[21].
Xingping Wen and Xiaofeng Yang, “An Unsupervised Classification Method for
Hyperspectral Remote Sensing Image Based on Spectral Data Mining”, INTECH.
[22].
Caglar Art et al., “Unsupervised Classification of Remotely Sensed Images Using
Gaussian Mixture Models And Particle Swarm Optimization”.
[23].
Mohd Hasmadi et al., “Evaluating supervised and unsupervised techniques for land
cover mapping using remote sensing data”, GEOGRAFIA , Malaysian Journal of
Society and Space 5 issue 1 (1 - 10)2009, ISSN 2180-2491.
[24].
Xiong Liu, “Supervised Classification and Unsupervised Classification”, ATS 670
Class Project.
[25].
Edwin Martínez Martínez, “Remote Sensing Techniques For Land Use Classification
Of Rio Jauca Watershed Using Ikonos Images”.
[26].
Cheng-Yuan Liou et al., “Unsupervised Classification of Remote Sensing Imagery
with Non-negative Matrix Factorization”, National Science Council under project
NSC 93-2213-E-002-008.
[27].
Y. Babykalpana et al., “Supervised/ Unsupervised Classification of LULC using
remotely Sensed Data for Coimbatore city, India”, International Journal of Computer
Applications (0975 – 8887) Volume 2 – No.7, June 2010.
[28].
Ratika Pradhan et al., “Land Cover Classification of Remotely Sensed Satellite Data
using Bayesian and Hybrid classifier”, International Journal of Computer
Applications (0975 – 8887) Volume 7– No.11, October 2010.
[29].
Luis Samaniego et al., “Supervised Classification of Agricultural Land Cover Using a
Modified k-NN Technique (MNN) and Landsat Remote Sensing Imagery”, Remote
Sens. 2009, 1, 875-895, ISSN 2072-4292.
[30].
L. Gomez-Chova et al., “Semi-Supervised Classification Method For Hyperspectral
Remote Sensing Images”, The Information Society Technologies (IST) programme of
the European Community.
[31].
Benqin Song et al., “Remotely Sensed Image Classification Using Sparse
Representations of APs”, IEEE Transactions on GeoScience And Remote Sensing,
Vol. 52, No. 8, August 2014.
[32].
Ays¸e Naz Erkan et al., “Semi-Supervised Remote Sensing Image Classification Via
Maximum Entropy”, The Spanish Ministry of Innovation and Science under projects.
[33].
Zhang Ming et al., “Application to Satellite Remote Sensing to Soil and Land use
Mapping in the rolling hilly areas of Nanjing, Eastern China”, EARSEL Advances in
Remote Sensing, Vol 2, No. 3, XI 1993.
[34].
X. Zhang et al., “Genetic Algorithm Optimized Neutral Network Classification And
Overlapping Pixel Change Detection Based On Remote Sensing For Urban Sprawl: A
Case Study In Jiading District Of Shanghai, China”, The International Archives of the
Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII.
Part B7. Beijing 2008.
[35].
Jefersson Alex dos Santos et al., “Interactive Classification of Remote Sensing
Images by using Optimum-Path Forest and Genetic Programming”.
[36].
K. Moje Ravindra et al., “Classification of Satellite images based on SVM classifier
Using Genetic Algorithm”, International Journal of Innovative Research in Electrical,
Electronics, Instrumentation and Control Engineering Vol. 2, Issue 5, May 2014.
IJCSIET--International
Journal
of
Computer
Science
information and Engg., Technologies ISSN 2277-4408 ||
01122014-005
Land Cover Classification
using GA based Fuzzy
Clustering Techniques for
Remotely Sensed Data
Dr.N.Sujatha
Assistant Professor, Department of Computer
Science,
Raja Doraisingham Govt. Arts College,
Sivagangai,Tamil Nadu, India
sujamurugan@gmail.com
ABSTRACT:
Remote Sensing Imagery is used by the
Government and private agencies for the
wide range of applications from military to
farm development. Fuzzy c-means
clustering is an effective algorithm, but the
random selection in center points makes
iterative process falling into the local
optimal solution easily. In this Paper, a
novel clustering method is developed
using GA based clustering techniques.
This technique enables the clustering to be
performed by taking the initial centroid
using mode function which allows the
iterative algorithm to meet to a “better”
local minimum. Then the GA based
improvement algorithm to get better
cluster quality. The study area taken here
is the Theni region, Tamil Nadu.
Keywords: Fuzzy C-means, Membership, Mode,
Euclidean Distance, Population, Chromosomes,
Mutation, Crossover, Selection.
being observed. Image classification [2] is
an important part of the remote sensing,
image analysis and pattern recognition. In
some instances, the classification itself
may be the object of the analysis. The
image classification therefore forms an
important tool for examination of the
digital images.
Classification strategies are basically
divided
into
three.
Supervised
Classification [3] techniques require
training areas to be defined by the analyst
in order to determine the characteristics of
each
category.
Unsupervised
Classification searches for natural groups
of pixels, called clusters, present within
the data by means of assessing the relative
locations of the pixels in the feature space.
Hybrid Classification takes the advantage
of both the supervised classification and
unsupervised classification.
2. Clustering Analysis
Clustering analysis [5] is classifying
samples according to their similarity by
means of unsupervised training. It makes
the samples, which have greater similarity,
as a class, and occupies the partial area of
feature space. The clustering center of
each partial area is respectively acting as a
representative of the corresponding
type.There are various types of clustering.
One of its type is Partitional Clustering.
Based on Partitional Clustering, many
algorithms are used. One of the familiar
algorithms is the FuzzyCmeans clustering
algorithm.
3. Fuzzy C-means Algorithm
1. INTRODUCTION
Remote sensing [1] can be defined as
any process whereby information is
gathered about an object, area or
phenomenon without being in contact with
it. The output of a remote sensing system
is usually an image representing the scene
It is an extension of k-means. Fuzzy CMeans [6] allows data points to be
assigned into more than one cluster. Each
data point has a degree of membership (or
probability) of belonging to each cluster.
In fuzzy clustering, each point has a
degree of belonging to clusters, as in fuzzy
logic, rather than belonging completely too
just one cluster. Thus, points on the edge
of a cluster may be in the cluster to a lesser
degree than points in the center of cluster.
For each point x we have a coefficient
giving the degree of being in the kth
cluster u k(x). Usually, the sum of those
coefficients is defined to be





With fuzzy k-means, the centroid
of a cluster is the mean of all points,
weighted by their degree of belonging to
the cluster:
Select m (m > 1); initialize the
membership function values µij,
i = 1, 2, ... , n; j = 1, 2, . . . , c.
Compute the cluster centers zj,
j = 1, 2, ... , c
Compute Euclidian distance dij,
i = 1, 2, ... , n; j = 1, 2, ... , c.
Update the membership functionµij,
i = 1, 2, ... , n; j = 1, 2, ... , c
If not converged, go to step 2.
Several stopping rules can be used.
One is to terminate the algorithm when the
relative change in the centroid values
becomes small or when the objective
function cannot be minimized more. The
FCM algorithm is sensitive to initial
values and it is likely to fall into local
optima.
4. Genetic Algorithm
The degree of belonging is related
to the inverse of the distance to the cluster
center:
then the coefficients are normalized and
fuzzyfied with a real parameter m > 1 so
that their sum is 1. So
For m equal to 2, this is equivalent
to normalizing the coefficient linearly to
make their sum 1. When m is close to 1,
then cluster center closest to the point is
given much more weight than the others,
and the algorithm is similar to k-means.
The algorithm [7] steps are as follows
American scholar, J. Holland, first
raised the Genetic Algorithm (GA)
concept in 1975. It is based on “survival of
the fittest” in Darwin’s theory of
evolution. The basic genetic operations,
which are repetitively utilized for the
groups possibly containing solution, make
the new groups generated then make them
evolved constantly. At the same time, the
optimization individuals in optimized
groups are searched based on the global
parallel search technique so as to obtain
the global optimum solution fulfilled
demands. GA [4] generates valuable
solutions for hard optimization problems
using techniques that are inspired by
natural evolutionary operators such as
inheritance, mutation, selection, and
crossover. A common genetic algorithm
involves two main parts:

All solutions should have a genetic
representation (in a shape of
chromosome).

There should be a fitness function
to assess the solutions.


4.1 .Chromosomes

The chromosomes represent set of
genes, which code the independent
variables. Every chromosome represents a
solution of the given problem. Individual
and vector of variables will be used as
other words for chromosomes. A set of
different chromosomes (individuals) forms
a generation. By means of evolutionary
operators, like selection, recombination
and mutation an offspring population is
created.


4.2 . Selection
The selection of the best individuals is
based on an evaluation of fitness function
or fitness functions. Fitness value is a
quality measurement of each solution.
Better fitness values belong to better
individuals in each population. When
termination criteria are satisfied, algorithm
reaches to better fitness value. In the final
generation, a solution with better fitness
value among others is found as the desired
solution.
4.3. Crossover
The first step in the reproduction
process is the recombination (crossover).
In it the genes of the parents are used to
form an entirely new chromosome. The
typical recombination for the GA is an
operation requiring two parents, but
schemes with more parent area also
possible. Two of the most widely used
algorithms are Conventional (Scattered)
Crossover and Blending (Intermediate)
Crossover.
4.4 Mutation
The newly created by means of
selection and crossover population can be
further applied to mutation. Mutation
means random change of the value of a
gene in the population.


Generate the initial population.
Calculate the values of the function
that we want to minimize or
maximize.
Check for termination of the
algorithm.
Selection is done between all
individuals
in
the
current
population are chose those, who
will continue and by means of
crossover and mutation will
produce offspring population
Crossover – the individuals chosen
by selection recombine with each
other and new individuals will be
created. The aim is to get offspring
individuals that inherit the best
possible combination of the
characteristics (genes) of their
parents.
Mutation – by means of random
change of some of the genes, it is
guaranteed that even if none of the
individuals contain the necessary
gene value for the extreme, it is
still possible to reach the extreme.
New generation – the best
individuals chosen from the
selection are combined with those
who passed the crossover and
mutation, and form the next
generation.
Figure1. The GA Flowchart
5. Methodology
Land cover is an important component
in understanding the interactions of the
human activities with the environment and
thus it is necessary to be able to simulate
changes. This proposal aims at developing
a novel land cover clustering method using
GA based clustering techniques.
The proposed method has two phases:
the first step computes a refined starting
condition from a given initial one that is
based on an efficient technique for
estimating the modes of a distribution. The
refined initial starting condition allows the
iterative algorithm to converge to a
“better” local minimum. And in the second
step, a novel method has been proposed to
improve to cluster quality by GA based
refinement algorithm.
improvement is, in any clustering
algorithm the obtained clusters will never
give us 100% quality. There will be some
errors known as misclustered. That is, a
data item can be wrongly clustered. These
kinds of errors can be avoided by using the
improvement algorithm.
The
genetic
algorithms
based
clustering may not be able to handle large
amounts of data. The Fuzzy C-means
algorithm does not lend itself well to
adaptive clustering. And an important
point is, so far, the researchers are not
contributed to improve the cluster quality
after grouping. In this proposed method, a
new framework has been introduced to
improve the cluster quality from Fuzzy Cmeans algorithm. The proposed algorithm
is applied to the remotely sensed data
(Survey of India toposheets and IRS-1C
satellite imageries) of Theni region.
5.1 . Preprocessing
Input Image
Pre Processing
Fuzzy C-means Clustering
GA based Tuning
Ground truth verification
Classified map
Figure 2: Methodology
Genetic algorithm (GA) is randomized
search and optimization techniques guided
by the principles of evolution and natural
genetics, having a large amount of implicit
parallelism. The basic reason for the
Satellite images cannot be given
directly as the input for the proposed
technique. Thus, it is indispensable to
perform pre-processing on the input
image, so that the image gets transformed
to be relevant for the further processing. In
proposed technique, A Median filter which
is a non linear filter is used in the R, G and
B layers for filtering noise. It is used
because, under certain conditions, it
preserves edges while removing noise.
5.2 . GA Based Clustering
A major problem with Fuzzy C-means
algorithm is that it is sensitive to the
selection of initial partition and may
converge to a local minimum of variation
if the initial partition is not properly
chosen. So in the proposed method, we
estimate the mode value as an initial
partition.
Performance
optimization
using
genetic algorithms is given by a sequence
of steps, which are:
7. Generate initial population.
8. Evaluate population
9. Selection.
10. Crossover.
11. Mutation.
12. Reinsertion of new individuals to
the population.
From step 2 to step 6, it performs an
iterative process until a stopping criterion
is met, in Fig. 3 we can see the Scheme of
GA for optimization of the Fuzzy CMeans algorithm (FCM). In this figure we
can observe that population evaluation is
done by FCM algorithm, but for us to
know how good some individuals need
something that does not indicate the fitness
of these, to measure aptitude of individuals
evaluated by FCM, we use the proposed
validation index mentioned in section III.
Individuals evaluated by the FCM
algorithm, are formed only by two
parameters which are the number of
clusters and the exponent of weight.
of Theni region. Two Theni region images
are used to implement the proposed
algorithm. The first image is 1152x1152
size tiff image. The second image is
1153x1153 size tiff image. Both images
are color image. The Original and the
clustered images are shown.
Figure 4. Original & Clustered Theni Region
Image1
Figure 5. Original & Clustered Theni Region
Image2
Even though the visual comparison gives
detailed information for Fuzzy C-means
clustering, to further evaluate the
performance of proposed work the
accuracy assessment has been done. The
confusion matrix in terms of pixels and
percentage is given in Table 1 and Table 2.
The overall classification accuracy is
96.04%.
Table 1. Confusion Matrix (Pixels)
Figure 3. FuzzyCMeans with GA
6. Results
Class
Urban
Vegetation
Hilly
Region
Total
Urban
Vegetation
2239
20
21
1680
4
100
22642
1800
7
90
1963
2060
2266
1791
2067
6124
Hilly Region
Total
The proposed algorithm is applied to
the remotely sensed data (Survey of India
toposheets and IRS-1C satellite imageries)
From the Confusion Matrix, it is clear that
urban yields a maximum classification
accuracy of 98.81% when compared to
Vegetation and Hilly Region.
[2].Aykut Akgun et al., “Comparing
Different Satellite Image Classification
Methods: An Application In Ayvalik
District,Western Turkey”.
Table 2. Confusion Matrix (Percentage)
[3].Ratika Pradhan et al., “Land Cover
Classification of Remotely Sensed
Satellite Data using Bayesian and Hybrid
classifier”, International Journal of
Computer Applications (0975 – 8887)
Volume 7– No.11, October 2010.
Class
Urban
Vegetation
1.17
93.80
5.03
Hilly
Region
0.19
4.84
94.97
Urban
Vegetation
Hilly
Region
Total
98.81
0.88
0.31
100
Total
36.97
29.39
33.64
100
100
100
Table 3 gives the producer and user
accuracy for individual classes. By
reducing the misclassification between the
Vegetation and Hilly region the overall
accuracy can be further improved.
Table 3. Accuracy Assesment
Class
Urban
Vegetation
Hilly Region
Producer
accuracy
(%)
98.81
93.80
94.97
User Accuracy
(%)
98.90
93.33
95.29
7. Conclusion
The proposed method has two phases:
the first step computes a refined starting
condition from a given initial one that is
based on an efficient technique for
estimating the modes of a distribution. The
developed initial starting condition allows
the iterative algorithm to converge to a
“better” local minimum. And in the second
step, a novel method has been proposed to
improve to cluster quality by GA based
improvement algorithm.
8. References
[1].“Introduction to Remote Sensing and
Image Processing” IDRISI Guide to GIS
and Image Processing Volume 1.
[4].K. Moje Ravindra et al., “Classification
of Satellite images based on SVM
classifier using Genetic Algorithm”,
IJIREEICE, Vol 2, issue 5, May 2014.
[5].Yingjie Wang, “Fuzzy Clustering
Analysis by Using Genetic Algorithm”,
ICIC Express Letters, Volume 2, Number
4, December 2008, pp. 331—337
[6].Ramandeep Kaur & Gurjith Singh
Bhathal, “A Survey of Clustering
Techniques”, International Journal on
Computer Science and Engineering Vol.
02, No. 09, 2010, 2976-2980
[7].Hesam Izakian & Ajith Abraham,
“Fuzzy C-means and fuzzy swarm for
fuzzy clustering problem”, Expert Systems
with Applications 38 (2011) 1835–1838,
www.elsevier.com/locate/eswa.