1. Introduction
Sensors have long been employed by man to obtain information which
the unaided human could not obtain or not obtain as easily. The develop-
ment of modern sensors has been so successful that the human being has
been virtually abandoned as a direct sensor and has been replaced with
devices of greater speed and capability. These devices are mounted in
vehicles of ever increasing speed and greater altitude so that large
areas of the earth can be sensed at faster rates while, at the same
time, resolution is improved so that finer detail is obtained. The
devices operate in spectral bands in and outside the visible region to
produce more information and all this information is disseminated at
ever faster rates over transmission media of growing capacity.
In view of these facts, it is appropriate to consider devices to
aid or replace man in the extraction of desired information from sensor
acquired data. My topic is a limited aspect of this general area of
automated processing and interpretation of remote sensing data of earth
resources and land use, namely automatic information extraction techniques
and processing methods for multispectral scanners, which are modern non-
photographic imaging sensors which operate in the ultraviolet, visible,
and reflective as well as thermal infrared regions of the spectrum. I
will discuss some of the latest advancements in automatic multispectral
scanner data processing, some of the major problems and tasks which
remain, and my recommendations for their solution. As a recent replace-
ment reporter, I ask your indulgence if I appear to give an unbalanced
preseutation in favor of my co-workers because I am more familiar with
the details of their work, although in any case I could not hope to cover
the many contributions of an expanding number of workers in this field
in the short time allotted.
The objective of these techniques is to have fast, reliable, and
accurate automatic procedures for extracting information about the
amount, location, and condition of objects of user interest and to
implement these techniques in a cost effective manner on an operational
basis for data from aircraft and satellites -- in short, to have a
practical tool for planners and decision makers in many disciplines.
Photographic interpretation has been defined as "the act of examing
photographic images for the purpose of identifying objects and judging
their significance" [1]. Woolnough [2] has suggested that machines
perhaps are best at "identifying objects" in aerial photographs and
humans are needed for "judging their significance". Likewise, processing
of multispectral scanner data is now and will remain for some time a
"semi-automatic" process requiring a man-machine interaction to achieve
the desired results. However, one objective is to reduce the man-machine
interaction (which slows the process) to an optimum point through
displays and controls which do not unduly impede but allow the man to
guide the processing.