740
FUTURE DEVELOPMENTS IN CLASSIFICATION
To date the emphasis in automated classification has been
on supervised methods with spectral information. The operational
classification systems will develop during the next year by incor
porating effective clustering algorithms with the use of multitem
poral data. Registration of digital ERTS imagery, important for
classification and change detection, will be initially restricted
to small areas where variations in scale are minimal. Registration
of complete ERTS frames for operational classification systems will
not be carried out until hardware systems have been developed which
permit rapid geometric co-ordinate conversion of entire frames. The
inclusion of such a geometric correction device in the present image
analysis programs would not only make the registration of temporal
imagery easier, but would increase considerably the mapping accuracy
of the resulting land-use maps.
It is anticipated that use of spatial information to per
mit shape discriminations will be made in the near future. The in
clusion of spatial measures with spectral information can be carried
out for ERTS. It is necessary, however, to reduce the ground re
solution of the image by segmenting the scene into cells, each of
which is large enough (about 5x5 pixels) to enable an accurate
determination of the spatial parameters to be made. This reduction
in resolution would decrease the cost effectiveness of the approach,
but these methods may still prove beneficial for mapping of geologi
cal features.
To date automated classification systems have not made
extensive use of digital sensor measurements collected by aircraft.
New developments in airborne radar units (SLAR) and multispectral
scanners will stimulate investigators to develop systems of resource
mapping which include satellite and/or airborne digital sensors.
Such research will be carried out by Canadians from many different
disciplines and institutions. We hope that the co-operation amongst
all scientists in the study will continue to grow.
