AN AUTODIGITIZING PROCEDURE
FOR GROUND-DATA LABELLING OF LANDSAT PIXELS*
Ma~tin Ozga and Richard Sigman
Economics and Statistics Service
United States Department of Agriculture
Washington, D.C., USA
automatic digitizing procedure
pixels. The described procedure is based
on video scanning of.line tracings
aerial photographs containing identified
The key feature of
p roe@dur@. - 15-th-a-t--t-n----t:h·@-d"8-t1I--~
Insteao the procedure directly maps scan
labels into LANDSAT pixel labels.
An evaluation of the procedure
alternative manual digitization methods,
Currently, the Economics and Statistics Service (ESS) of the
of Agriculture (USDA) is engaged in research on crop area
estimation using LANDSAT data.
Besides LANDSAT data,
investigated uses ground data collected by USDA enumerators for sampled
areas, known as segments.
This ground data is collected
(JES), which is a major data source for ESS's operational
(non-LANDSAT) crop statistics program.
For each field in a segment, the
ground .cover, along with other information, is
This segment data can be used to label LANDSAT pixels for supervised
training of a pixel classifier if one is able to identify the LANDSAT
to the fields in each segment.
Currently, a major step in this
labelling of LANDSAT pixels
is the -manual digitization
into polygons using a coordinate digitizer.
Table I lists some of
the manual digitization characteristics for JES-segment photographs.
method, described in this paper, replaces manual digitization
scanning of field boundaries into a raster format.
2. CURRENT METHOD
Using the present met