1. Built-up Extraction from RISAT Data Using Segmentation Approach
Chetna Soni*, Manoj Joseph**, A. T. Jeyaseelan** and J R Sharma**
*Banasthali University, Niwai
**RRSC-W, NRSC/ISRO Jodhpur

2. 6/10/2018
Background
Extraction of Built-up area is important in Urban planning, Urban Sprawl mapping and also in analyzing Urban Heat Islands
Satellites Remote sensing has the advantage of synoptic view of earth with temporal coverage in assessing the spatial and temporal nature of urbanization
In Radar imageries, built up area shows very high back scattering since buildings act as corner reflector which gives multiple bounce
Pixel based classification of SAR images give less accuracy due to the presence of noise like speckles but object-based classification deals with objects instead of pixel.
Segmentation is an optimization procedure to group the pixels to form objects which minimizes the heterogeneity in neighborhood pixels .
Segmentation works on two principles. Bottom up- Grouping of pixels to make larger object. Top down- Cutting down objects to make smaller objects.
Objective
Extraction of Built-up area from SAR imagery using segmentation approach

3. Study Area & Data Software Used Data Specification Study Area
Jodhpur City and surround area.
Data Specification
RISAT-1 FRS-1 (Fine resolution mode)
DoA : 1st July 2012
Polarisation : Dual pol HH-HV
Frequency band : C-band
Incident angle :
Spatial resolution : 8m
Software Used
ENVI- Sarscape
Definiens eCognition/Developer

4. Built-up in SAR Imagery
In SAR imageries built-up is having high backscattering co-efficient than to other features. Backscattering co-efficient values for built-up lies between 7.0 to -7.0 dB
Urban Fallow land crop

5. Methodology
Import RISAT Image
To improve visualization multi-looking with 3*3m (Range*azimuth) window size has been done.
Enhanced frost filtering reduces speckles while preserving texture information of the image.
It identifies single image object of pixel size and merges with neighboring objects based on homogeneity criteria (combination of spectral and shape).
Scale Shape Compactness
Spectral Difference Segmentation refines the previous segmentation by merging objects which are spectrally similar.
Maximum Spectral Difference
Supervised classification with Nearest Neighborhood algorithm has been done.
Merge region algorithm reduces the number of objects by merging the neighboring object of a class.
Multi-looking,  Calibration, Layer Stacking, Filtering
Multi-resolution Segmentation
(Level 1)
Spectral Difference Segmentation
(Level 2)
Training Sites
NN Classification
Built-up Extraction
Merge Region
Validation

6. Scale- Higher value of scale forms larger objects and small value forms smaller one
Compactness- Value assign to compactness give relative weighting to smoothness
Shape- The shape homogeneity is based on the deviation of a compact (or smooth) shape
maximum spectral difference- Define the maximum spectral difference in gray values between image objects
Level level 2
Scale 3 compactness 0.2 shape maximum spectral difference 0.5
Segmentation level 1
Segmentation level 2

7. Optimization of segmentation parameters
Scale Compactness Shape
Scale 10
Compactness 0.8
Shape 0.8
Scale 5
Compactness 0.5
Shape 0.5
Scale 3
Compactness 0.2
Shape 0.2

8. Classification Results
In supervised Classification Training Samples are taken to train the classifier. Class separability is between Built-up and Non built-up
Built-up
Non built-up
Results
Extracted Built-up
from RISAT
Extracted Built-up from
LISS-IV Carto Merge

9. Validation Apartments High court Building 12th July 2012
6/10/2018
Validation
12th July 2012
12th July 2010
Apartments
High court Building

10. Resulted Statistics
Land cover Feature
RISAT Image
(area in sqkm)
LISS IV-Carto Merge Image
Built-up
61.155
57.651
Back scattering varies due to incidence angle and orientations of buildings.