rite Xu e e Load
In order.to reduce the signal film histories to a digital image, this facility
consists of a coherent optical processor interfaced:to an image dissector under
computer control. The data at the output plane were digitized with an image
dissector using 256 gray levels and the resulting samples were recorded on
computer compatible tape (CCT). A sample spacing of. 0.75 m was used in both
slant range and azimuth (1.5 m resolution in slant range sampled twice in
accordance with the Nyquist theorem). Investigations done at ERIM (Ref.2)
showed that the slant range and azimuth impulse point response (IPR) achieved
during these flights of the SAR 580 system was not quite as good as the nominal
values of 1.5 m for range and 2.1 m for azimuth. The measured IPR (3 db width)
was 2.3 m in slant range and 2.2 m in azimuth for X-Band (3.2 cm) and 3.3 m
in slant range and 3.9 m in azimuth for L-Band (23.5 cm). A one-dimensional
convolution resampling operation was used to transform these data from slant
range to ground range coordinates. In order to be able to overlay the data of
different wavelengths and flight passes, which were chosen perpendicular to
each other by the SWISSAR Investigation Team, I resampled to a 1.5 mby 1.5m
grid system on the ground. The L-Band imagery was found to be of lesser quality
than X-Band imagery. In view of this and the resources required to carry out the
digitization, I chose to concentrate my current study to the seven data sets
depicted in Figure 2 instead of all twelve which were available.
Figure
Data S
à An affine transformation with coefficients
Palio: determined by a least squares fit was
nau used to transform SAR slant to ground
Test Area range corrected imagery from each data
M set (flight direction/-wavelength/po-
larization) to a coordinate system based
upon a 1:25'000 scale topographic map. In
order to determine the best linear trans-
SE formation between the map coordinate
714 XHV (M > M.) and radar image coordinate
A (Ri; Rp)» I estimated the six coefficients
da using a set of 10 common points on both ;
the map and each of the seven data sets. Figure
These points (typically field corners) Data: Sc
were chosen well distributed throughout
the entire test area. Their locations
were determined visually using
1024 x 1024 images. For flight pass 709
Ou) the least squares fit based upon
the control points resulted in the
following coordinate transformation
Eqs; "T and 2.
hs
711 XHH «4
Figure 2. Flight Direction and
Wavelength/Polarization Selection
R
|7 7.15221 M, - .44533 M, + 760.05 (1)
Ry
-.24885 M, + -07889 M, + 285,71 (2)
Applications of this kind of transformation to the control points themselves
provide an estimate of the accuracy of the least Squares fit. The regression
coefficient, r2, was 0.99976 for predictions in scan line direction and 0.99940
for predictions in pixel direction. The mean Square error was 8.13 in the scan Figure
line direction and 6.14 in the pixel direction, which corresponds to a mean Data Se
error of lessthan 4 pixels (6 meters). This procedure was repeated for each
232
uM MEE [5 8 5 5 1 0 I ROS SSR 7
a OO”