Babawuro Usman/ Elixir Comp. Sci. & Engg. 63 (2013) 18671-18675 18671
Introduction
Satellite imagery is an indispensable tool in scientific
research and environmental planning, with applications in
numerous fields. One of the applications is in Image
classification. Image classification approaches have evolved
over the years. This development has been driven by the need
for higher accuracies in the classified results coupled with the
emergence of high resolution satellite imageries, such as Quick
Bird and IKONOS which pose a greater challenge to many
image classification methods [1]. The purpose of image
classification is to label the pixels in the image with meaningful
information of the real world for better and useful information
extraction. Through classification of satellite imagery, thematic
maps bearing the information such as cadastral information, land
cover type, vegetation type, etc. could be obtained [2]. Image,
classification methods may be grouped into two main categories
depending on the image primitive used, viz pixel based or object
based methods. Pixel based methods classify individual pixels
without taking into account any spatial information of the pixel.
Only the spectral patterns are used. On the other hand, object
based methods attempt to group pixels into objects by an image
segmentation process based on a chosen similarity, e.g., texture,
color, intensity and then use the spectral, spatial and contextual
information inherent in these objects to classify the whole image
[1]. It has emerged as a superior way of doing image
classification. One of its strength is the ability to extract real
world objects, proper in shape and accurate in classification. It
eliminates the mixed pixel problem suffered by most pixel based
methods. This is because the image is classified on an object
level and usually more information is used. Object based
methods are also able to handle high resolution satellite imagery
which aggravates the classification process for most pixel based
methods [1]. Classification techniques include conventional
statistical algorithms, such as discriminate analysis and the
maximum likelihood classification, which allocate each image
pixel to the land cover class in which it has the highest
probability of membership [4]. One of the major disadvantages
of these classifiers is that they are not distribution free [5]. In the
same vain, traditional pixel-based classification methods have
difficulty with high resolution satellite imagery, resulting in a
“salt and pepper” appearance.
In this paper, an attempt has been made to practical classify
high resolution satellite imagery into accurate spatial groups
with the following classes using the object based approach for
cadastral, environmental studies and management. The classes
are farmlands, bare lands, built-up areas, and others. In this
study, the classification of the imagery has been done using
color k-means clustering algorithm, where the imagery was
classified into various classes with a view to determine the most
optimum clusters based on apriori knowledge of the imagery,
and then the land cover classification was performed. The aim
has been to identify and classify farmlands for statutory
environmental functions. Initially the imagery has been
georectified to assume the planer surface that could be needed
for environmental quantitative image analysis [5]. The paper is
organized as follows: In section 2, we describe the related work
of the paper while in Section 3, we describe the color
segmentation scheme and the implementation process. In
Section 4, we give experimental results and an evaluation
method was presented in Section 5. Finally in Section 6, we
draw a concise conclusion.
Related Work
Lonesome M. M. [6], used a region-based approach for
doing image classification. The main goal was to develop an
alternative procedure for an object-based image classification.
The procedure significantly reduced the mixed pixel problem
suffered by most pixel based methods. Wen C. et al. [7],
presented a Satellite image classification method using color and
Satellite Imagery Land Cover Classification using K-Means Clustering
Algorithm Computer Vision for Environmental Information Extraction
Babawuro Usman
Department of Computer Science, Faculty of Computing and Mathematical Sciences, Kano University of Science and Technology,
Wudil, Kano State, Nigeria.
ABSTRACT
Segmentation and classification of high resolution satellite imagery is a challenging problem
due to the fact that it is no longer meaningful to carry out this task on a pixel-by-pixel basis.
The fine spatial resolution implies that each object is an aggregation of a number of pixels in
close spatial proximity, and accurate classification requires that this aspect be subtly
considered. K-means clustering algorithm is a better method of classifying high resolution
satellite imagery. The extracted regions are classified using a minimum distance decision
rule. Several regions are selected as training samples for region classification. Each region is
compared to the training samples and is assigned to its closest class. The procedure
significantly reduces the mixed pixel problem suffered by most pixel based methods. In this
paper, we used K-means clustering algorithm to classify satellite imagery into specific
objects within it for cadastral and environmental planning purposes, thereby eliminating the
above mentioned problems and getting better classification accuracy with the overall
performance for accuracy percentage as 88.889% and Kappa values as 0.835.
© 2013 Elixir All rights reserved
ARTICLE INFO
Article history:
Received: 10 August 2013;
Received in revised form:
2 October 2013;
Accepted: 21 October 2013;
Keywords
K-means algorithm,
Clustering,
Satellite Imagery,
Classification.
Elixir Comp. Sci. & Engg. 63 (2013) 18671-18675
Computer Science and Engineering
Available online at www.elixirpublishers.com (Elixir International Journal)
Tele:
E-mail addresses: wuroabu@yahoo.com
© 2013 Elixir All rights reserved