Bhattacharya distance. A spectral distance of 1.75 was used as
the minimum cut-off point for good feature discrimination.
5. RESULTS AND INTERPRETATION
A total of I5 class pairs for each texture measure were
analysed. For the three dates this resulted into 45 class pairs
and for the four texture measures a grand total of 180 class
pairs were analysed. From the spectral separability based on
Bhattacharya distance, different class pairs were chosen which
could be separable. For the window size of 3x3, the feature
discrimination was possible only between cotton - urban and
pearl millet - urban in Entropy and Angular Second Moment
only in the first date data. In the window size of 7x7, the
discrimination increased in Contrast, Entropy and ASM both in
the first date and the second date, from the 3x3 window. In the
third date. however, the feature discrimination was not
significant. A total of 36 pairs of spectrally distinct classes
emerged. For a 15x15 window size, a total of 57 pairs of
spectrally distinct classes evolved in Contrast, Entropy and
ASM. In a 25x25 window size, 74 spectrally distinct class
pairs were observed and Correlation measure resulted in 11
pairs of such classes. On increasing the window size to 31x31,
the increment in distinct class pairs over 25x25 window size
was only two. Figure 1 illustrates the number of separable
classes under different window sizes and texture measures.
July 18 data
-—
N
-—
©
i
1
Separable Class pairs
Oo
S
15x15 31x31
Window sizes
ÆCont. ElEnt. BASM E Corr.
3x3 7X7 25x25
Fig 1. Comparison of texture measures under different window
sizes for first date.
By the first date of satellite pass, the crops were already sown
and they were in seedling or early vegetative stage. As the row
spacing in cotton is between 125 and 150 cm., the soil exposure
was very high and it contributed significantly to the
backscattered signal. For pearl millet the row spacing was 70-
75 cm. The soil was wet at many fields and this caused
differential backscatter in these fields. The harrowing of soil in
row direction for different fields caused different. textural
signatures. The entropy and contrast were low because most of
the contribution was from soil and ASM dominated in most of
the window sizes. The fields had heterogeneity with respect to
wetness, roughness and row orientation. The RMS of soil
roughness was of the order of around 5-6 cm. Which was of the
order of the wavelength of microwave.
On the second date, the crop had grown marginally in the case
IAPRS & SIS, Vol.34, Part 7, "Resource and Environmental Monitoring", Hyderabad, India,2002
of cotton and pearl millet was in vegetative stage. Rice was in
early tillering stage. Contrast, Entropy and ASM showed same
number i.e. 8, of distinct class pairs in 25x25 and 31x31
window sizes, but the distinct crop class pairs was more in
ASM than in Contrast and Entropy. Figure 2 shows the changes
in texture measures caused by crop growth.
August 11 data
12
Separable class pairs
25x25
15x15 31x31
Window Sizes
Cont. p] Ent. Bj ASM Ei Corr.
Figure 2. Comparison of texture measures under different
window sizes for second date.
By the third date, the crop had reached the vegetative stage and
the canopy cover over soil was around 60-70 percent. The
texture in the image was due to the different backscatter caused
by volume scattering of the canopy coupled with the soil
background. Maximum number of class pairs was
distinguishable in 25x25 and 31x31 window sizes. However,
ASM and Correlation measures showed a fall in the number of
class pairs. This can be due to the fact that, the average
synthetic field sizes do not exceed 5 ha and with a pixel
spacing of 12.5 m a 31x31 window could cause a change in the
crop type. Secondly, the crop stages were also similar causing
small differences in backscatter. But the distinct crop class
pairs were more in Entropy and ASM.
Sept. 04 data
Separable Class pairs
AR
15x15
3x3 7x7 25x25 : 31x31
Window Sizes
Ei Cont. rj Ent. Bg ASM Ej Corr.
Figure 3. Comparison of texture measures for third date