ISSN (O) 2278-1021, ISSN (P) 2319-5940
IJARCCE
International Journal of Advanced Research in Computer and Communication Engineering
ISO 3297:2007 CertifiedImpact Factor 7.918Vol. 11, Issue 10, October 2022
DOI: 10.17148/IJARCCE.2022.111002
© IJARCCE This work is licensed under a Creative Commons Attribution 4.0 International License 15
Land Cover Segmentation of Multispectral
Images Using Multiresolution Algorithm
Herlawati
1
, Prima Dina Atika
2
, Rahmadya Trias Handayanto
3
Faculty of Computer Science, Universitas Bhayangkara Jakarta Raya, Jakarta, Indonesia
1,2
Computer Engineering Department, Universitas Islam 45, Bekasi, Indonesia
3
Abstract: Semantic segmentation is needed by regional planners to know the composition of land cover in their area, so
that they can take the right policy. Several methods from manual to automatic have been researched, both based on colour
and pattern. Each method has their strong and weakness, so it is necessary to make the right choice when applying the
method. Currently, multispectral imagery is still very rarely used, even though sources of information from the internet
are easy to find, i.e. Landsat imagery from the United States Geological Survey. This study uses two methods for
segmenting three-channel multispectral images (red, green, and blue), namely iterative self-organizing clustering
(ISOCLSUT), which is based on a colour sensor, and a multiresolution algorithm, which is based on colour and pattern.
For the experiment, the pre-processed satellite image of Karawang district was segmented using the ISOCLUST as well
as multiresolution algorithm. The experimental results show that land cover segmentation with multiresolution algorithm
is better than ISOCLUST for RGB but for more than three channels, i.e., seven frequency channels, ISOCLUST shows
better performance compared to real image conditions.
Keywords: Satellite Imagery, Multispectral, ISOCLUST, Multiresolution
I. INTRODUCTION
In the Geographic Information System (GIS) there are two land use maps, namely land use and land cover. Land cover
describes how a land is covered by a biophysical environment such as buildings, trees, water, and the like, while land use
describes how human socio-economic activities on land such as housing, business, industry, schools and the like [3]. A
land cover, for example a building, in a land use can be residential, industrial, or business.
For land cover, the term classification is usually given the term segmentation. Moreover, in the computer science
literature a similar term to classification and segmentation is object detection. In the segmentation process, each pixel of
the satellite image will be segmented into several categories such as buildings, vegetation, waters, and other types of land
cover.
Currently, the main source of land cover data is imagery derived from remote sensing such as satellites, drones, unmanned
aerial vehicles (UAVs). Image processing is needed to produce a raw image into a segmented image. There are two types
of processing involved: colour pixel based and object/pattern based [2], [3].
Initially, the map was made by direct survey to the area. However, since the development of remote sensing, most maps
are drawn using satellite imagery. Applications that are currently developing such as Google Map, WAZE, among others,
use Global Positioning Service (GPS) on smartphones connected to these applications. These applications work using
existing techniques and methods in the field of Geographic Information Systems (GIS). In other words, remote sensing
captures images and represents them with GIS tools. [4].
Digital image processing is needed in managing satellite capture from remote sensing. Satellite images have hyperspectral
characteristics, namely one image capture has a number of frequency bands, for example Sentinel-2, and Landsat-9
images have 13 frequency bands [5]. Several remote sensing applications are available that are capable of classifying
based on colour differences from satellite imagery. The application works well for land cover but is not able to distinguish
buildings that have a specific use (land use), whether settlements, factories, roads, terminals, airports, and others. Dyna-
Clue and Idrisi/TerrSet are applications that are often used in land cover classification. Other applications then emerged
to perform pattern/object-based land cover segmentation known as Object-Based Image Analysis (OBIA), such as
eCognition. For example, OBIA can distinguish between a river and a lake, although they are both water because they
have different object shapes.