Support Vector Machine Classification to Detect Land Cover Changes in Halabja City, Iraq (1)

1
Support Vector Machine Classification to Detect
Land Cover Changes in Halabja City, Iraq

• presented by
• Jwan Al-doski
• Geospatial Information Science Research Centre
  (GIS RC),
• Faculty of Engineering
• Universiti Putra Malaysia 

2
INTRODUCTION

3
Earth Surface

• The earth's surface is changing as a result of
  natural phenomena or human activity.
• The earths surface changes are divided into two
  categories Land use and land cover ( Barnsley et
  al. 2001).

4
Remote Sensing, Briefly

• Remote sensing is the process of collecting data
  about landscape features without coming into
  direct physical contact with them.

Satellite
Aero-plane
5
Remote Sensing Data

• Aerial Photographs

Satellite Images
6
Landsat Satellite
There are large collections of past and present
Landsat imageries making it possible to analyze
the impact of human activities on land cover
7
Remote Sensing Applications

• Remote sensing satellite imagery satellite has
  been utilized in different application.

8
Land Cover

• Land cover is the physical material at the
  surface of the earth. Land covers include grass,
  asphalt, trees, bare ground, water, etc

9
Knowing about Land Cover changes

• Change detection can simply be defined as the
  process of identifying differences on earth
  surface covers over time
• LC changes is a key toward understanding the
  earth as a system
• LC changes of a region is one of the
  prerequisites for the planning and implementation
  of effective land use policies and schemes for
  sustainable regional development

10
Change Detection Techniques

• Several techniques have been developed and
  evaluated to perform LC changes detection using
  remote sensing imagery However, there are two
  basic methods
• Post-classification Comparison
• Pixel-to-pixel Comparison

11
Post-classification

• Post-classification is the most widely technique
  used in change detection studies. It determines
  the difference between independent classified
  images from each of the dates.

12
Disadvantages of Post-classification

• The accuracy of the post-classification
  comparisons of land cover is dependent on the
  accuracy of single initial classifications
  through time.
• Is necessary to critically evaluate
  classification accuracy of different time images
  prior to employing the classified data in change
  detection studies

13
Advantages of Post-classification

• It minimizes data acquisition errors, image noise
  and sensor differences problems .
• It is the only method that provides from-to
  change information and the kind of landscape
  transformations that have occurred can be easily
  calculated and mapped.
• It can be employed using data acquired from
  sensors with different spatial, temporal and
  spectral resolutions.

14
Classification Approaches

• Image classification is a process to categorize
  all pixels in a digital image into one of several
  land cover classes. Since 1970s, various
  classification approaches developed and employed
  to extract and monitor LULC information and
  changes. There are two categories known as
• Supervised
• Unsupervised .

15

• In order to generate thematic
• maps for this study the
• support vector machine (SVM) supervised
  classification
• was applied

16
support vector machine (SVM)

• The classification maps of the Halabja city are
  created to the highest accuracy possible

17
Main Objectives

• Assessing the effectiveness of Support Vector
  Machine (SVM) supervised.
• Identify what land cover changes have occurred
• To answer this question can we locate bombing
  places as a result of shelling by chemical
  weapons using classification change detection
  techniques.

18
Study Area

• Halabja city is a Kurdish city in the Northern
  part of Iraq with an area bout 5363ha
• It located about 240 km northeast of Baghdad
  within 3510'59.73"N latitude and 4558'59.05"E
  longitude

19
DATA USED
20
Multi-temporal dataset
Landsat5 TM

• Landsat 7 ETM

Images Satellite Instrument Date Pixel Size (M)
1986 Landsat 5TM June14, 1986 30x30
1990 Landsat 5TM June09, 1990 30x30
2000 (Base Image) Landsat 7ETM June28, 2000 30x30
Sources Earth Resources Observation Science (EROS) Center Global Land Cover Facility (GLCF) Earth Resources Observation Science (EROS) Center Global Land Cover Facility (GLCF) Earth Resources Observation Science (EROS) Center Global Land Cover Facility (GLCF)
Spatial Data http//www.diva-gis.org/Data http//www.view-group.net http//www.diva-gis.org/Data http//www.view-group.net http//www.diva-gis.org/Data http//www.view-group.net
21
METHODOLOGY
22

1. Pre- Processing
2.Classification
3. Change Detection
23
RESULTS
24
Classification Images Using SVM
25
ACCURACY ASSESSMENT
26
Error Matrix of LC Map, 1986
LULC Classes Ground Truth (Pixels) Ground Truth (Pixels) Ground Truth (Pixels) Ground Truth (Pixels) Ground Truth (Pixels)
LULC Classes Vegetation Urban Area Bare Land Total Rows User Accuracy()
Vegetation 160 3 4 167 96
Urban Area 4 284 5 293 97
Bare Land 8 3 169 180 94
Total Columns 172 290 178 640
Produce Accuracy () 93 98 95
Overall Accuracy() 96 96 96 96 96
Kappa Coefficient 0.9 0.9 0.9 0.9 0.9
27
Error Matrix of LC Map, 1990
LULC classes Ground Truth (Pixels) Ground Truth (Pixels) Ground Truth (Pixels) Ground Truth (Pixels) Ground Truth (Pixels)
LULC classes Vegetation Urban Area Bare Land Total Rows User Accuracy ()
Vegetation 18 1 0 19 94.74
Urban Area 0 72 5 77 93.51
Bare Land 8 7 92 107 85.98
Total Columns 26 80 97 203
Produce Accuracy () 69 90 95
Overall Accuracy 90 90 90 90 90
Kappa Coefficient 0.8 0.8 0.8 0.8 0.8
28
Change Image Map
Legend
29
Change Masks

• In ENVI 4.8, change statistics and change masks
  for each class in each image are produced Change
  masks were used to show what each class in the
  initial state image changed to in the final state
  image.

30
Vegetation Mask Change
31
Bare Land Mask Change
32
Urban Area Mask Change
33
Urban Area with Bombed Location
34
Example of Change in Urban Area from 1986 to 1990
1986 1990
35
Summary of the LULC Area Changes by Hectare in
Halabja City During 1986 to 1990
  1990 Map 1986 Map 1986 Map 1986 Map 1986 Map
  1990 Map Vegetation Urban Area Bare Land Total Area
Vegetation 373 65 289 727
Urban Area 4 257 3183 271
Bare Land 848 334 9 4365
Total Area 1225 656 3482  
5363 5363 5363 5363
36
Chart Area change (Km) During 1986-1990
2
Classes Area Change from 1986 to 1990 Change
Vegetation -499 -40.7
Urban Area -385 -58.7
Bare Land 884 25.4
37
CONCLUSIONS
38

• land cover change in the Halabja city, Iraq
  examined using two Landsat 5 TM images over a 4
  year time period from 1986 to 1990.
• Post classification base on SVM supervised
  classification algorithm applied to classify into
  three classes urban area, agriculture and bare
  land.
• Land cover maps of the Halabja city were created
  for each image at an average accuracy of 93 with
  an average Kappa coefficient of 0.85 which were
  ideal to examine changes in three land cover
  classes and to disregard the changes from 1986 to
  1990.
• The main overall change trend was the decrease in
  urban area and agricultural classes vice versa
  the bare land class increased by 884 ha from 1986
  to 1990.
• This was the most important finding is bombed
  place is the same urban area changed

39
Reference
40
Thank you