IAPRS & SIS, Vol.34, Part 7, “Resource and Environmental Monitoring”, Hyderabad, India,2002
TEXTURE FOR CROP DISCRIMINATION IN SAR
C. Patnaik', M. Chakraborty and S. Maity
Agricultural Resources Group
Space Applications Centre (ISRO)
Ahmedabad 380 015
KEYWORDS: Radar, texture, Grey Level Co-occurrence Matrix (GLCM), crop discrimination, spectral distance, window size.
ABSTRACT
Synthetic Aperture Radar images have a high texture and texture is one of the important characteristics used to identify objects or
regions of interest in an image. One of the methods of measuring texture mathematically is based on grey level co-occurrence matrix
(GLCM). An attempt was made to discriminate major crops in Surendranagar district of Gujarat, India, using texture measures.
Multi-temporal Radarsat Standard Beam 5 was taken for the study. Plant parameters and soil parameters were collected during
ground truth synchronous with the satellite pass.
Four texture measures, which were studied, are: Contrast, Entropy, Angular Second Moment (ASM) and Correlation. The window
sizes chosen were 3 x 3; 7 x 7; 15 x 15; 25 x 25 and 31. x 31. The multi-temporal data was calibrated and georeferenced, but not
filtered. Texture measures were used on all three-date data separately and the output images generated. The statistics obtained from
output image were used for computing the spectral distances using Bhattacharya distance. In this study it was found that Entropy
and ASM resulted in the maximum number of distinct class pairs for all the window sizes. However, ASM in all the three dates and
all the window sizes resulted in better feature discrimination than Entropy. The window size of 25x25 was found to be optimal for
the current study.
1. INTRODUCTION
In India, the kharif (monsoon) season accounts for most of the
agricultural activity of the year and getting optical remote
sensing data for crop inventory is an uncertainty. Microwave
remote sensing helps here but vegetation discrimination in SAR
under Indian agricultural conditions is comparatively difficult
as compared to optical data. This is due to the fact that
microwave data is sensitive to crop moisture, geometry,
roughness and other soil parameters. However, with multi-
temporal, multi-incident, multi-frequency or multi-polarised
data, vegetation discrimination is possible.
SAR images have a high texture and texture is one of the
important characteristics used to identify objects or regions of
interest in an image. Unlike spectral features, which describe
the average tonal variation in the various bands of an image,
textural features contain information about the spatial
distribution of tonal variations within a band. These textural
measures are derived from a grey level co-occurrence matrix or
difference vector computed for each rectangular window of
user specified dimensions and spatial reldtionships of the input
image. Based on the textural information contained in radar
images, a study was carried out to find out the feasibility of
using texture as a crop discriminator when large area crop
inventory has to be done.
The objectives of the current study were
Identify the appropriate texture measure
Decide the optimum Window size
to use texture as a crop discriminator.
2. DESCRIPTION OF TEXTURE MEASURES
In the understanding of images, pixel colour and brightness are
“ E-mail : cpatnaik@rediffmail.com 079-6774003
commonly used parameters. À less often used parameter is the
texture (graininess). As opposed to colour and brightness
(which are associated with | pixel) texture is computed from a
set of connected pixels. Texture information in an image is
contained in the local spatial relationship that the intensity
values have to one another.
There are several paradigms for measuring texture
mathematically. A commonly used one is based on the grey
level co-occurrence matrix (GLCM). A GLCM is a matrix of
relative frequencies, P(ij), with which two neighbouring
pixels, separated by a distance d and angle 6, occur on an
image, one with grey level i, the other with grey level j. Here, P
is the normalised symmetric GLCM of dimension N x N where
N is the number of grey levels. A grey level can represent the
intensity at each pixel of the textured image for analysis. The
texture of an image is related to the grey level joint probability
distribution, which is approximated by the co-occurrence
matrix. Particularly, the amount of dispersion that the GLCM
elements have about the diagonal characterises the texture of
the local region. The texture measure is put at the centre of the
window at the appropriate position in the output image.
Feature extraction of a textured region can be based on a
statistical approach, which has three categories: a) the use of
power spectrum and autocorrelation function; b) the use of
grey-level statistics, e.g. histograms and co-occurrence
matrices; c) the use of local feature statistics, e.g. the statistical
distribution of corners. (Fletcher, 2002).
For a local window size X by Y and spatial vector = (dX, dV).
the total number of pixel pairs used to build the GLCM is:
2*(X-|dXp*(Y-|dY])
