inadequate. The Multivariate Interactive Digital Analysis System (MIDAS)
system is an attempt to solve the problem of real time multispectral data
processing in an operational system. We have been aware of this problem
since the initial design and operation of the M-7 scanner and processing
equipment at ERIM over 10 years ago. A real-time processor (SPARC - SPectral
Analysis and Recognition Computer) was completed in 1967, allowing parallel
decision operations using a multivariate maximum likelihood algorithm thus
making possible for the first time, rapid processing of 12-band data (Kriegler,
1968). It was evident that a considerable amount of assistance and time
was needed by the user in setting up such a machine and controlling its
operation. This led to the evaluation of a hybrid system employing a
general-purpose digital computer for control by the user of the system
(Marshall,1973). Processing ofL remotely sensed data is an interactive
process in which the man and machine must, in fact, be considered as the
real processing system.
This is not apparent in general purpose computer processing systems
using software classifiers since the machine is so slow by comparison to
parallel digital approaches that an operator is easily able to keep pace
with the task (hours or days is the normal time interval). In this MIDAS
system, where the parallel-pipeline digital processor is substantially
faster, the time required by an operator is three or four times longer than
that used in processing. Increases in processor speed will then provide
little improvement in throughput. It becomes evident that well designed,
interactive display and control subsystems will, in reality, offer the
greatest gains in throughput.
From another point of view, the objective of providing greater speed also
provides much lower cost. This is true directly for the operational situation
but also for research and development. It has been estimated that processing
costs in an operational system can be reduced by about a factor of twenty or more
from some present day feasibility processing costs. Costs may be determined by the
time required for training. For ERTS costs as low as $500/frame may be possible.
MIDAS represents a breakthrough in the field of multispectral scanner image
analysis by providing a low-cost capability for user-oriented, interactive,
near-real-time, digital analysis to produce thematic mapping with instantaneous
or multi-temporal data. MIDAS accepts data from aircraft or spacecraft MSS
in the form of high-density digital instrumentation tape, computer-compatible
tape, or analog tape, and makes use of proven multispectral processing
techniques (including signature extension feature extraction) within an
innovative hardware approach resulting in a cost-effective, user-controlled
system for multispectral analysis and recognition. Its hardware and software
are intended to require a minimum amount of instructional training for
successful operation. MIDAS is intended to provide multispectral analysis
for applications in disciplines such as agriculture, urban planning, forestry,
geology, pollution detection, hydrology, and others. Features may be extracted
that are spectral, spatial, temporal, and (possibly) polarization-dependent,
thus giving a very general and powerful capability.