We have prepared a list of potentially interesting objects identified
semi-automatically in RADARSAT data obtained on 2007 Jan 31 and Feb 03.
The pixel size is 12.5 meters per pixel, so a 44-foot boat would be
unresolved. Nonetheless, we note the presence of larger objects for the
sake of completeness and to facilitate comparisons between datasets.
We may have missed some potentially interesting objects, and some of
the objects in our list may turn out to be artifacts.

We calculated longitude and latitude for potentially interesting
groups using data in the geotiff headers in the tiff files. The
geotiff headers indicate that the coordinate system type is UTM
for zone 10. We converted X, Y pixel locations to longitude and
latitude, using 'GRS 1980/WGS 84' datum and the IDL functions
map_proj_init() and map_proj_inverse(). Following this procedure,
we were able to obtain the longitude and latitude of the Golden
Gate Bridge to within 5" in each coordinate.

We used the read_tiff() function in IDL to read the tiff files.
The X, Y pixel indexing convention may be different in other image
processing software, but the longitude and latitude should be
consistent.

We divided each image (approximately 10k x 10k pixels) into 900 tiles.
Each tile contained approximately 350 x 350 pixels. We constructed a
histogram of 16-bit data values in each tile, ignoring pixels exactly
equal to zero (indicating no data). We fitted a Gaussian to the
histogram of pixel values. We then set the detection threshold at
the mean plus 8.0 times the standard deviation of the Gaussian.
We coalesced neighboring pixels above this threshold into groups.
We then visually inspected each group, using subjective criterea
to decide whether the pixels could be a boat. The Jan 31 data covers
considerable land, which generated many spurious groups that we
rejected by hand. The Feb 3 data included sets of north-south
striations in some tiles, which generated some spurious groups
that we rekected by hand.

For each interesting object, we created a two-panel JPEG image with
a 21x21 pixel zoom and 161 x 161 pixel overview of each object. The
data values in these images are scaled to improve visual contrast.
A title above the panels gives the original filename, date, time,
longitude, latitude, X, and Y. These data are also tablulated in a
separate CSV file.

For each interesting object, we adopt the following naming convention:

  image root name _ round(abs(1000*longitude)) _ round(1000*latitude)

Send corrections and questions to Maria Nieto Santisteban (nieto@
pha.jhu.edu) or Jeff Valenti (valenti@
stsci.edu).