International Journal of Computer Trends and Technology (IJCTT) volume 5 number 5 Nov 2013

ISSN: 2231-2803 http://www.ijcttjournal.org Page252

Automatic Generation of Standard Deviation
Attribute Proles for SpectralSpatial Classication of
Remote Sensing Data

Ch. Bhavani

Y. Jaya babu

Abstract: In recent years, the development of high-
resolution remote sensing image extends the visual field
of the terrain features. Back Propagation neural network
is widely used in remote sensing image classification in
recent years, it is a self-adaptive dynamical systemwhich
is widely connected by large amount of neural units, and
it bases on distributing store and parallel processing. It
study by exercise and had the capacity of integrating the
information, synthesis reasoning, and rapid overall
processing capacity. It can solve the regular problem
arise fromremote sensing image processing; therefore, it
is widely used in the application of remote sensing.
Utilizing data mining tasks such as classification on
spatial data is more complex than those on non-spatial
data. It is because spatial data mining algorithms have to
consider not only objects of interest itself but also
neighbors of the objects in order to extract useful and
interesting patterns. One of classification algorithms
namely the ID3 algorithmwhich originally designed for a
non-spatial dataset has been improved by other
researchers in the previous work to construct a spatial
decision tree from a spatial dataset containing polygon
features only. The objective of this paper is to propose a
new spatial decision tree algorithm based on the ID3
algorithmfor discrete features represented in points, lines
and polygons. As in the Improved ID3 algorithmthat use
information gain in the attribute selection, the proposed
algorithmuses the spatial information gain to choose the
best splitting layer froma set of explanatory layers. The
new formula for spatial information gain is proposed
using spatial measures for point, line and polygon
features. In this proposed work, the use of Back
Propagation neural network classifier and decision tree
approach will be implemented on the high resolution
remote sensing images of their pattern recognition
Proposed systemwill shows better performance compare
to existing RF and SVM based classification techniques.


Keywords: Decision Tree, ID3, Support Vector
Machine, and Spatial Information.

I. INTRODUCTION

Prerequisite for many environmental and socio-economic
Applications such as urban change detection, urban heat
islands, etc. [1]. Satisfactory classification of RS data
depends on several factors including (a) the
characteristics of study area,

(b) Availability of suitable RS data, (c) ancillary and
ground reference data, (d) proper use of variables and
classification algorithms, (e) users experience with
reference to the application and (f) time constraints [2].
Furthermore, diverse landscapes and terrain types have a
mixture of both homogeneous and heterogeneous land
cover (LC) classes and require supplemental
environmental or geographical layers for improved
classification accuracies. Increased spectral variation is
common with high degree of spectral heterogeneity in
complex landscapes [3]. For example, urban landscapes
are composed of features having a complex mix of
buildings, roads, flyovers, pavements, trees and lakes
which are sometimes smaller than the mediumspatial
resolution sensors [4].
Remote sensing image classification is regarded as an
important mean for quantified remote sensing image
analysis. Classification technology as an important
branch of remote sensing technology has been the
widespread attention of researchers in remote sensing
community for many years. In general, the remote
sensing image classification has mainly two ways. One is
the visual interpretation; the other is computer automatic
classification. The computer automatic classification uses
the pattern recognition technology and the artificial
intelligence technology to carry on the analysis and the
deduction, understand the remote sensing image, and
complete the interpretation of the remote sensing image
finally, according to the land targets characteristic in the
remote sensing image and the goal land targets
interpretation experience and image formation rules in
the expert knowledge library, based on the computer
system. Minimum distance from means, maximum-
likelihood, cluster analysis and
Bayesian classification is a simple and conventional
classification method [1-2] based on statistical principles.
Some new methods for remote sensing classification also
have been developed, including machine learning,
support vector machine, neural network, and fuzzy set
and genetic algorithm [3]. These methods may have
higher accuracies than conventional classifiers.





International Journal of Computer Trends and Technology (IJCTT) volume 5 number 5 Nov 2013
ISSN: 2231-2803 http://www.ijcttjournal.org Page253

II. LITERATURE SURVEY

K Nearest Neighbor (KNN) queries is frequently used in
Geo Spatial applications. The k Nearest Neighbor (KNN)
query is an important class of queries in geospatial
databases. The KNN query searches for sites (e.g., hotels,
gas stations or other points of interest) with minimum
distance with reference to one or more query points. The
most common type is the point KNN query, which is
defined as: given a set of spatial objects (or points of
interest), and an input query point, retrieve the (K)
nearest neighbors to that query point [2,3].The efficient
implementation of KNN query is of a particular interest
in geographical information systems (GIS).
However, although the exploitation of the features
extracted by the profiles already proved to improve the
accuracies in a classification task [1], the profiles in
general increase significantly the dimensionality of the
data being analyzed. This effect is further increased when
considering multiple APs/EAPs built on different
attributes in order to provide richer descriptions of the
spatial information contained in the scene. Furthermore,
the information contained in a profile is, in general,
intrinsically redundant (e.g., adjacent levels in a profile
can differ one to the other only by few regions in the
whole image). The increase in the dimensionality and the
high redundancy can be an issue, especially if a classifier
non robust to the Hughes phenomenon is considered. For
This reason, in [5, 6] the use of Feature Extraction (FE)
techniques applied to the MP and EMP was proposed
prior to the classification stage. The reduction of the
dimensionality of the data performed with FE techniques
suitable for classification proved to increase the
classification accuracies when considering MPs and
EMPs. Thus, this approach becomes very interesting
when applied to EAPs especially when multiple attributes
are considered.

PROBLEMS IN EXISTING WORK:

Does not handle noisy remote sensing data.
Loading time is more.
Does not predict the geographical areas and
shapes.
Accuracy rate is between 80 to 88% i.e. high
noisy error rate.


III. PROPOSED SYSTEM

ALGORITHM-1: Neural Network Approach:

Back Propagation neural network is widely used in
remote sensing image classification in recent year.
Researches show that the precision of the image
classification has been greatly improved by neural
network model for supervised classification of remote
sensing images. The reason can be listed as follows:
Neural network classifier can study discontinuous, no-
linear classification model; Neural network model has
good robustness and self-adaptability, it is able to end the
question in the specific conditions; Neural networks able
to combine the parameters of the remote sensing image
such as shape, spectral, texture and so on, then it can
extract the potential information




The networks learning process consist of forward
propagation and back propagation. In the forward
process, input signals frominput layer pass hidden layer
and transmit to the output layer. If there is a difference
which is defined as error value between the actual output
values and known-correct output values, then BP network
turns to the process of error back propagation. The error
value is then propagated backwards through the network,
and small changes are made to the weights in each layer.
The activation function of neurons in the network is S
Type Function (tangent function). Determine the number
of neurons of the each layer is the key point of network
architecture. First we determine the number of neurons of
the input layer. Because of the image to be classified has
four bands, we determine the number of input neurons is
four, and then we fix on the number of neurons of the
input layer. We classify the image into five ground
objects, so the output neurons are five. We determine the
number of neurons in the hidden layer finally, through
some experiments, we increase or decrease the number of
hidden layer neurons in a suitable way. We will use 24
hidden layer neurons to obtain more precise classified
results.

Normalized training samples and expect output
vectors:

Before we use BP Neuron Network to train or classify the
remote sensing image, in order to avoid the saturation in
the process of network broadcasting, input vectors must
be normalized.


Normalized formula:



International Journal of Computer Trends and Technology (IJCTT) volume 5 number 5 Nov 2013
ISSN: 2231-2803 http://www.ijcttjournal.org Page254





ALGORITHM-2: SPATIAL DECISION TREE:

The Improved Spatial ID3 calculates information gain to
define the best splitting layer for the dataset. In spatial
decision tree algorithmwe define the spatial information
gain to select an explanatory layer L that gives best
splitting the spatial dataset according to values of
predictive attribute in the layer L.
Let a dataset D be a training set of class-labeled tuples. In
the non-spatial decision tree algorithm we calculate
probability that an arbitrary tuples in D belong to class C
i

and it is estimated by |C
i
, D|/|D| where |D| is number of
tuples in D and |C
i
, D| is number of tuples of class C
i
in D
[2]. In this study, a dataset contains some layers
including a target layer that store class labels. Number of
tuples in the dataset is the same as number of objects in
the target layer because each tuple will be created by
relating features in the target layer to features in
explanatory layers. One feature in the target layer will
exactly associate with one tuple in the dataset. For
simplicity we will use number of objects in the target
layer instead of using number of tuples in the spatial
dataset. Furthermore in a non-spatial dataset, target
classes are discrete-valued and unordered (categorical)
and explanatory attributes are categorical or numerical. In
spatial dataset, features in layers are represented by
geometric type (polygons, lines or points) that have
quantitative measurements such as area and distance. For
that we calculate spatial measures of layers (1) to replace
number of tuples in a non-spatial data partition.


V. EXPERIMENTAL RESULTS


Home page

International Journal of Computer Trends and Technology (IJCTT) volume 5 number 5 Nov 2013
ISSN: 2231-2803 http://www.ijcttjournal.org Page255


Loading hyper spectral image database


Applying the proposed algorithm for
classification


Image class color mapping to classification


Final image classification region approach


V. CONCLUSION AND FUTURE SCOPE

This work has shown that use of spatial information
along with ancillary and derived geographical layers is an
effective way to improve classification performance,
which was demonstrated through implementation in three
different terrains. Compared with the classification of
only optical remotely sensed data, the integration of
optical-based data can effectively improved the accuracy
of classification. Experimental results give better results
compare to exiting approach in terms of accuracy and
error rate.
The further improvement in terms of efficiency and
accuracy, which can be achieved by improving the image

International Journal of Computer Trends and Technology (IJCTT) volume 5 number 5 Nov 2013
ISSN: 2231-2803 http://www.ijcttjournal.org Page256

enhancement techniques or by improving the hardware to
capture the image. So that the input image to the
thinning stage could be made, better this could improve
the future stages and the outcome.

REFERENCES

[1] Chang- Tien, Jing Dai, Ying Jin and Jank Mathurai
GLIP: A Concurrency Control Protocol for Clipping
Indexing, vol. 21 no. 5. pp 714-728: May 2009.
[2] J. Dai, C.-T. Lu, and L.-F. Lai. A Concurrency
Control Protocol for Continuously Monitoring Moving
Objects, in Proc. International Conference on Mobile
Data Management. pp 132-141: 2009.
[3] B. Yao, F. Li, and P. Kumar, K nearest neighbor
queries and KNN-joins in large relational databases
(almost) for free, Technical report: 2009.
[4] R. Laurini, and D. Thomson, Fundamentals of Spatial
Information Systems, London: Academic Press, 1992.
[5] A. Corral, Y. Manolopoulos, Y. Theodoridis, and M.
Vassilakopoulos, Closest pair queries in spatial
databases, in Proc. ACM SIGMOD. Pp 189-200: 2000.
[6] A. Corral, Y. Manolopoulos, Y. Theodoridis, and
M.Vassilakopoulos, Algorithms for processing K-closest-
pair queries in spatial databases, vol 49 pp 67-104: 2004.
[7] Orenstein J. A, Spatiul Query Processing in an
Object-Oriented Database System, Proc. ACM SIGMOD
Int. Conf. on Management of Data Washington D.C.: pp
326-333, 1986.
[8] Borrough P. A, Principles of Geographical
Information Systems for Land Resources Assessment:
Oxford University Press, 1986.
[9] Y. W. Hung, N. Jing and E. A. Rundesteiner, BFRJ
global optimization of spatial joins using R-trees, Dept.
of Computer Science, Worcester Polytechnic Institute,
Tech. Report. WPI-CS-TR-97-5: January, 1997.
[10] Y. W. Hung, N. Jing, and E. A. Rundesteiner,
Spatial joins using R-trees breadth-first traversal with
global optimizations, Proc. 23rd Int. Conf. on VLDB. pp
396-405: 1997.

Author Profile:


Name: Y.Jaya Babu professor in cse dept at pragati
engineering college surampalem with total 18years
experience. Intrested areas are database management
systems and operating system with research
specialization data mining and data warehousing.