ABSTRACT Slums are becoming inevitable phenomenon of the urban fabric in the developing world. Many cities in developing countries lack detailed information on the emergence and growth of highly dynamic slum developments. Due to the lack of spatial and temporal data, informal settlements are often not spatially documented. There are no maps indicating the position, patterns, size, complexity and influence of the settlements. The challenge therefore lies in having appropriate methods to identify and monitor the spatial behaviour of informal settlements. One possible solution is to use remote sensing imagery as the primary data source. This project explores different approaches to extract slum in which the gray level co-occurrence matrix(GLCM) texture features are extracted from small individual blocks to classify slums and formal built-up areas in very high resolution satellite imagery ( WorldView-2 image of Madurai city). The next approach employs the utility of homogenous urban patches (HUPs) for which the information extracted from the GLCM variance is aggregated for better accuracy. The result evaluated using collected ground-truth information and visual image interpretation shows that an accuracy of 81% is achieved. KEYWORDS – Gray level co-occurrence matrix (GLCM), Homogenous Urban Patches(HUP), texture features, slum classification, accuracy. I. INTRODUCTION A slum is a collection of households living in close proximity to one another in a number of buildings such that the households share one or more deprivations of access to improved water; access to improved sanitation facilities; sufficient-living area; structural quality/durability of dwellings; and security of tenure. According to UN- HABITAT (2010),the world’s slum population is expected to reach 889 million by the year 2020.The intent of the United Nations Millennium Development Goal 7, Target 7D is ―to have achieved a significant improvement in the lives of at least 100 million slum dwellers‖ by 2020. Reliable identification of slums and tracking of their growth has always been a difficult task for urban administrators in the developing world. There is no doubt that slum areas are expanding in major urban centres of developing countries worldwide, but significant challenges remain in evaluating them and measuring expansion of built-up areas with scientific methods. Although shape-based measures (fractal dimension, lacunarity, mathematical morphology) and texture measures (gray-level co-occurrence measures) have been used to identify individual slum communities in the past two decades, minimal progress has been made on development of geospatial measures that can distinguish differences between informal and planned settlements using readily-available very high resolution (VHR) multispectral imagery. The United Nations Global Urban Observatory project has acknowledged there is a lack of data and an immaturity of applicable methodology to measure durability of informal settlements and also emphasized the need to ―be able to identify and define slums spatially in a consistent manner to be able to use geographical targeting for slum intervention programs‖. The objective of this research is to develop a small set of statistically significant indicators that distinguish settlement type to be used by organizations such as the UN HABITAT and urban planners without the need for field work or surveys. This research advances the ability to distinguish informal (slum) from formal areas by analyzing shape (form), texture, vegetation, lacunarity of a built-up areas using geographic information systems (GIS) and remote sensing image analysis. Recent advances in computing power and the increasing availability of remote sensing imagery have revived renewed interest in remote sensing as potential data ware for monitoring informal settlement behaviour. Results showed that mapping high-resolution data using purely spectral information resulted in relatively low map accuracies while using image segmentation and classification tree approach increased map accuracy. The classical approach used in extraction of slum is based on Object Based Image Analysis which is carried out in two steps, namely segmentation and supervised classification. The main limitation in Object Based Image Analysis is due to the complexity of satellite images, segmentation is a difficult task. Machine learning approaches have been employed successfully for extraction of slum in the recent time. As machine learning techniques are pixel based it has become a major drawback for VHR imagery. From the detailed review of prior literatures, informal settlements appear to exhibit more dirt roads, less vegetation, less road accessibility, higher SLUM EXTRACTION APPROACHES FROM HIGH RESOLUTION SATELLITE DATA – A CASE STUDY OF MADURAI CITY G.Girija 1 , R.Immaculate Nikhila 2 1,2 UG Student, Department of Electronics and Communication Engineering,St.Joseph’s College of Engineering, Chennai. 1 [email protected], 2 [email protected]International Journal of Pure and Applied Mathematics Volume 119 No. 12 2018, 14509-14514 ISSN: 1314-3395 (on-line version) url: http://www.ijpam.eu Special Issue ijpam.eu 14509
6
Embed
SLUM EXTRACTION APPROACHES FROM HIGH ...slums in the Madurai city along the Vaigai banks and railway tracks mostl y concentrated in Arapalayam, Periyar bus terminal area, Karumbalai
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
ABSTRACT
Slums are becoming inevitable phenomenon of the urban fabric
in the developing world. Many cities in developing countries lack
detailed information on the emergence and growth of highly
dynamic slum developments. Due to the lack of spatial and
temporal data, informal settlements are often not spatially
documented. There are no maps indicating the position, patterns,
size, complexity and influence of the settlements. The challenge
therefore lies in having appropriate methods to identify and
monitor the spatial behaviour of informal settlements. One
possible solution is to use remote sensing imagery as the primary
data source. This project explores different approaches to extract
slum in which the gray level co-occurrence matrix(GLCM)
texture features are extracted from small individual blocks to
classify slums and formal built-up areas in very high resolution
satellite imagery ( WorldView-2 image of Madurai city). The
next approach employs the utility of homogenous urban patches
(HUPs) for which the information extracted from the GLCM
variance is aggregated for better accuracy. The result evaluated
using collected ground-truth information and visual image
interpretation shows that an accuracy of 81% is achieved.
International Journal of Pure and Applied MathematicsVolume 119 No. 12 2018, 14509-14514ISSN: 1314-3395 (on-line version)url: http://www.ijpam.euSpecial Issue ijpam.eu
14509
texture and contrast, more heterogeneity of spaces between
built-up areas, steeper slopes and less favourable terrain
geomorphology for dwellings than formal settlements.
Including measures from each of these categories to correctly
identify settlement type is therefore a worthwhile test, and
should contribute to the collection of replicable, measurable
indicators that can be used to determine the presence of
informal settlements in a variety of urban areas. In this paper, we explore the use of gray level co-occurence
matrix(GLCM) texture features in the extraction of slums and
produce useful information for pro-poor policies. The
technique of combining HUPs and GLCM texture feature
addresses the limitation of pixel-level analysis which is not
suitable for pro-poor policies.
II. CASE STUDY AREA
The study area considered here is the Madurai city in the
South Indian state of Tamil Nadu. The urban population in
Tamil Nadu as per 2011 census is 349.50 lakhs and among
them 20% is living in slums. Madurai is the third largest city
in Tamil Nadu next to Chennai and Coimbatore and the
second largest municipal corporation in Tamil Nadu next to
Chennai. According to 2011 provisional census data, Madurai
city had a population of 1,016,885 (before expansion of city
limit) within the corporation limits, with 509,313 men
(50.08%) and 507,572 women (49.92 %). The urban
agglomeration had a population of 1,462,420. Madurai
metropolitan area constitutes the third largest metropolitan
area in Tamil Nadu and the 24th in India. (TNSCB Report,
RAY 2013)
Longitude 78 6’ 42.34‖E - 78 7’ 17.07‖ E
Latitude 9 55’ 3.46‖ N -9 54’ 33.33‖ N
Fig.1 Study Area
The image is taken by World View-2 satellite sensor which
was launched at September which provides a resolution of
0.46 meter for panchromatic and 1.85 meter for multispectral
images(multispectral bands - red, green, blue, and near-
infrared bands) for enhanced spectral analysis mapping and
observing applications. In this work, the Worldview-2 MSS
(Multispectral data) of Madurai city in the year of January,
2010 is considered. Initially, 331 slums within the Madurai
Corporation area were identified for relocation of the
residents. But later, the officials scrapped 135 of them from
the list as they were already developed. There are nearly 200
slums in the Madurai city along the Vaigai banks and railway
tracks mostly concentrated in Arapalayam, Periyar bus
terminal area, Karumbalai and Alwarpuram.We have
considered the slum dataset of Karumbalai and Managirislum.
Fig.2 Slum dataset-Karumbalai and Managiri slum
III. PROPOSED METHODOLOGY
The steps involved in our research are broadly classified into
four stages: Segmentation, Feature extraction, Classification
and Accuracy Assessment.
Segmentation involves two approaches, one involves dividing
the entire image into individual blocks and another involves
formation of Homogenous Urban Patches(HUP). Feature
Extraction stage involves the computation of features using
statistical (GLCM) method. After extracting different features,
classification is performed by considering a particular GLCM
feature. Finally, in the accuracy assessment stage, the
classified result is verified with the ground truth data with the
help of error matrix.
Fig.3 Proposed methodology Flow chart
International Journal of Pure and Applied Mathematics Special Issue
14510
A. GLCM FEATURES CALCULATION
GLCM is a tabulation of how often different combinations of
gray levels co-occur in an image or image section. The gray
level co-occurrence matrix is found for the whole dataset
image and the corresponding texture feature values like ASM,
contrast, correlation, energy, homogeneity and variance are
calculated. However, we usually do not want a single measure
for a whole image. The texture measure calculation is done to
a GLCM derived from small areas on the image. We then look
at a different small area and record its texture measure to
cover the whole image and find quantitatively how the pixel
relationships differ in different places.
In order to implement this technique, the entire dataset image
is divided into small individual blocks where each block
consists of 16*16 pixels and 8*8 pixels. The gray level co-
features are calculated. Five training data of slum and formal
areas are considered and GLCM texture feature values are
computed for these training data. The average of each texture
feature values corresponding to slum and formal areas are
calculated. These average values are compared with the
texture feature values of each block. The blocks whose values
lie closely with the average value of slum are classified as
slum areas. Thus classification is done based on the
comparison.
B. HOMOGENOUS URBAN PATCHES(HUP)
Homogenous settlements are also referred to as homogenous