Ny Altho
RENT AA than 1
INERTIA, 5 = > ) ("risque (8) withir
i=l j=1 is exe
image
where N = number of digital intervals. e
a clas
This difference moment of the co-occurrence mátrix is a scalar whose value (in m)
represents a measure of the contrast or the amount of local variation present to man
in the subregion over which the co-occurrence matrix was computed. Statis
Experimentation with the distance parameter led to my selection of d = 2 as Mean v
most appropriate. The size of the moving window used for the inertia computation the ay
had to be a compromise between being large enough to provide stable estimates buted
of the co-occurrence matrix elements, yet small enough not to severely distort Sov
the values computed for the narrow fields in the study area. The output image 2 fan
obtained using average inertia with d = 2 for the texture measure is shown 1mage
ingFigure 6. (The "average" is taken over the four orientations, 9 = 0”, 45°,
90^, and 1357).
The variations in gray tone in this
image correspond to variations in tex-
ture in the input image. This same
procedure was also applied to Xu
digital data from the other two flight
directions (data sets 710 and 711).
What makes the co-occurrence approach
so valid is its characterization of
the spatial interrelationship of the
gray tones in a textural pattern in a
manner which is invariant under mo-
notonic gray tone transformations.
Moreover, as implemented here, no a
priori knowledge of the location of
i field boundaries is required. The
ze : approximation of the local mean value
Figure 6. Texture Image Representing (see Fig.4) uses the digital SAR
Average Inertia From Data Set 709 Xu imagery jtself. to capture the shape,
size and location of the individual
agricultural fields.
CROP CLASSIFICATION
The goal of reliable crop separation in the presence of bi-directional re-
flectance effects on the signal returns from different viewing angles has
existed since the early days of remote sensing. In addition to dielectric
constant, a SAR sensor responds primarily to the relative roughness of vege-
tation. Therefore, crop types which differ significantly in their vertical
development at the scale of the radar wavelength should be able to be dis-
criminated. Moreover, plant density, relative height, and whether or not it is
a row crop, all contribute to the textural pattern of a specific crop type as
well as to its tone. While field-to-field differences are clearly seen in the
raw SAR data, the key question is: can digital features be developed from the
SAR returns to form the basis for reliable discrimination between the crops
of interest?
256
HMM SE
