18-08-2012, 04:52 PM
Digital Image-Processing Activities in Remote Sensing for Earth Resources
Digital Image Processing Activities in Remote Sensing for Earth Resources.pdf (Size: 4.87 MB / Downloads: 51)
Abstract
The United States space program is in the throes of a
major shift in emphasis from exploration of the moon and nearby
planets to the application of remote sensing technology toward increased
scientific understanding and economic exploitation of the
earth itself. Over one hundred potential applications have already
been identified. Since data from the unmanned Earth Resources
Technology Satellites and the manned Earth ResourcOebss ervation
Satellites are not yet available, the experimentation required to
realize the ambitious goals of these projects is carried out through
approximation of the expected characteristics of the data by means of
images derived from weather satellite vidicon and spin-scan cameras,
Gemini and Apollo photographs, and the comprehensive sensor
complement of the NASA earth resources observation aircraft.
The extensive and varied work currently underway is reviewed
in terms of the special purpose scan and display equipment and efficient
data manipulation routines required for high-resolution images;
the essential role of interactive processing; the application of supervised
classification methods to crop and timber forecasts, geological
exploration, and hydrological surveys; the need for nonsupervised
classification techniques for video compaction and for more efficient
utilization of ground-control samples; and the outstanding problem
of mapping accurately the collected data on a standard coordinate
system.
An attempt is made to identify among the welter of “promising”
results areas of tangible achievement as well as likely bottlenecks,
and to assess thec ontribution to bee xpected of digital image-processing
methods in both operational and experimental utilization of the
forthcoming torrent of data.
I. ISTRODUCTIOS
T H E O B J E C T of this survey is to give an account of
experimental developments in digital image processing
prompted by the major environmental remote sensing
endeavors currently underway, such as the already operational
weather satellite program of the National Oceanographic
and Atmospheric Agency (NOAA), the projected
Earth Resources Technology Satellite (ERTS) and Skylab
experiments, the NASA Earth Resources Aircraft Program
( E R A P ) , a n d t h e D e p a r t m e n t of the Interior’s Earth Resources
Observation System (EROS).
Sources of Information: The most comprehensive and
readily accessible source of material in this area is the seven
volumes published so far of the Proceedings of the International
Symposium on Remote Sensing of Enyironment, held annually
under the auspices of the Center for Remote Sensing Information
and Analysis of the Uni\-ersity of llichigan.
Other useful sources of information are the NAS.-l-MSC
Annual Earth Resources Program Reziews, the Proceedings
of the Princeton Cnielersity Conference on Aerospace Methods
for Rer’ealing and Ez~alxating Earth‘s Resources, the publications
of the American Society of Photogrammetry and of the
Society of Photo-Optical Instrumentation Engineers, the
Jozunal of Applied Meteorology, the Proceedings of the I E E E
Manuscript received January 31, 1972: revised June 30, 1972.
The author was with IBM Thomas J. \Vatson Research Center,
Yorktown Heights, S.Y. 10598. He is now with the Department oi Computer
Science, University of Sebraskn, Lincoln, Seb. 68508.
(pertinent special issues in April 1969 and in July 1972), the
IEEE Transactions on Computers and the IEEE Transactions
on Man, Machines, and Cybernetics, the Journals of Remote
Sensing of Environment and of Pattern Recognition, and the
proceedings of several symposia and workshops on picture
processing and on pattern and target recognition. Previous
introductory and survey articles include Shay [185], Colwell
and Lent [37], Leese et al. [120], Park [167], Dornbach [48],
and George [67].
As is the case with most emerging fields of research, the
assiduous reader is likely to encounter considerable redundancy,
with many experiments republished without
change in the electrical engineering and computer literature,
in the publications dealing with aerial photography and
photogrammetry, in the various “subject matter” journals
(agronomy, meteorology, geophysics), in the pattern recognition
press, and in the increasing number of collections devoted
exclusively to remote sensing.
A depository of relevant published material, government
agency reports, and accounts of contractual investigations is
maintained by SASA at the Earth Resources Research Data
Facility at the Manned Spaceflight Center in Houston, Texas
(Zeitler and Bratton [ 2 2 3 ] ) . The Facility also maintains a
file of most of the photographs obtained by the NASA satellites
and earth observation aircraft, and by other cooperating
agencies, institutions, and organizations. Provisions
are made for convenient browsing through both the printed
material and the vast amounts of photography. The Center
publishes Mission S z m m a r y R e p o r t s and detailed Screening
and Indexing Reports of each data-collection operation and
acts in principle as a clearinghouse for the exchange of such
material. A l l of its holdings are cataloged by subject, location,
and author, but in its periodically published computer compiled
Inde? [155]; documents cannot, unfortunately, be located
by either author or title. An annotated liosft references
to the literature is, however, also available [154].
For background information, the book Remote Sensing,
embodying the report of the Committee on Remote Sensing
for -Agricultural Purposes appointed by the National -Academy
of Sciences, is recommended as much for its comprehensive
coverage (the chapters on “Imaging with Photographic
Sensors,” “Imaging with Sonphotographic Sensors,” “Applications,”
and “Research Seeds,” are particularly interesting)
as for the quality of its photographic illustrations [161].
The reports of the other Committees are also available [lSj].
The International Geographic Union is compiling a sur\-ey
of current work, including a list of participating scientists,
in geographic data sensing and processing. The long-range
plans of the United States, as presented to the Committee on
Science and .Astronautics of the U. S. House of Representati\-
es, are set forth in [197], [38], and [60].
Contents of the Paper: .Although much of the current ac1178
PROCEEDINGS OF THE IEEE. OCTOBER 1972
tivity is sponsored by NASA, mosto f the early work in remote
sensing was initiated by military intelligence requirements;
in particular, the developmento f imaging sensors was greatly
accelerated by the deployment of high-altitude photoreconnaissance
aircraft and surveillance satellites. Very little information
is, however, available in the open literature about
the actual utilization of the collected imagery. The few published
experiments for instance, in the Proceedings of the
Symposium on Automatic Photo Interpretation (Cheng et al.
[SI]), deal almost exclusively with idealized target recognition
or terrain classification situations faremoved from presumed
operational requirements. In view of the scarcity of
up-to-date information, this aspect of remote sensing will be
discussed here only in passing despite its evident bearing and
influence even on strictly scientific and economic applications.
We shall also largely avoid peripheral applicaotif odni gital
computers to the collectioonr preparation of pictorial material
intended only for conventional visual utilization, as in the
calculation of projective coefficients in photogrammetry or
the simulation of accelerated transmission methods independent
of the two-dimensional nature of the imagery. Nor
shall we be concerned with statistical computations arising
from manually derived measurements, as in models of forest
growth and riparian formations based on aerial photographs,
or in keys and taxonomies using essentially one-dimensional
densitometric cross sections or manual planimetry.
Omitted too is a description of the important and interesting
Sideways Looking Airborne Radar all-weather sensors.
Such equipment will not be included in the forthcoming
satellite experiments. Its potential role in remote sensing is
discussed by Simonett [187], Hovis [94], and Zelenka [224].
The diffuse and unstructured nature of terrestrial scenes
does not lend itself readily to elegant mathematical modeling
techniques and tidy approximations; an empirical approach
is well-nigh unavoidable. The first ERTS vehicle is not, however,
expected to be launched until the second half of 1972,
and the Skylab project is scheduled for 1973, hence, preparatory
experimentation must be based on other material. Although
none of the currently available sourceso f imagery approximates
closely the expected characteristics of ERTS and
Skylab, some reflect analogous problems, and several are of
interest on their own merits as large scale data-collection systems.
These sensor systems, including both spaceborne and
airborne platforms, are described in Section 11.
A large portion of the overall experimental effort has been
devoted to developing means for entering the imagery into a
computer, for storing and retrieving it, and for visual monitoring-
both of the hardware available for scanning and displaying
high-resolution imagery, and of the software packages
necessary for efficient manipulation of large amounts of twodimensional
(and often multiband) imagery in widely disparate
formats. These matters are discussed in Section 111.
Section IV is devoted to image registration, the difficult
problem of superimposing two different pictures of the same
area in such a way that matching elements are brought into
one-to-one correspondence. This problem arises in preparing
color composites from images obtained simultaneously
through separate detectors mounted on the same platform, in
constructing mosaics from consecutive overlapping pictures
from a single sensor, in obtaining a chronological record of the
variations taking place in the courseo f a day or a year, and in
comparing aspects of the scenery observed through diverse
sensor systems. The most general objective here consists of
mapping the images onto a set of standard map coordinates.
Section V is concerned with the application of automatic
classification techniques to the imagery. The major -problem
is the boundless variability of the observed appearance of
every class of interest, due to variations inherent in the features
under observation as well as in atmospheric properties
and in illumination. The difficulty of defining representative
training classes under these circumstances has led to renewed
experimentation with adaptive systems and unsupervised
learning algorithms. From another point of view, the classification
of observations into previoasly undefined classes is an
efficient form of data compression, an objective of importance
in its own right in viewof the quantityo f data to be collected.
By way of conclusion, we attempt to gauge the progress accomplished
thus far in termofs w hat still remains to be donif e
automatic digital image processing isto play a significant part
in the worthwhile utilization of the remote sensing products
about to become widely available.
The remainder of this Introduction lists some of the proposed
applications for ERTS and Skylab, outlines the functional
specifications for the image collection systems designed
for these platforms, and describes the central data processing
facility intended to accelerate widespread utilization of the
ERTS image products.
A. Objects of the United States Remote Sensing Program
I t is too early to tell whether expectations in dozoefn s pecific
application areas are unduly optimistic [185], [38], [60].
Certainly, few applications have emerged to date where
satellite surveillance has been conclusively demonstrated to
have an economic edge over alternative methods; it is only
through the combined benefits accruing from many projects
that this undertaking may be eventually justified.
Typical examples of proposed applications are crop inventory
and forecasting, including blight detection, in agriculture
[61], [169]; pasture management in animal husbandry
[97], [32]; watershed management and snow coverage measurement
in hydrology [135], [22]; ice floe detection and
tracking in oceanography [93], [196]; demarcation of lineaments
and other geographic and geomorphological features in
geology and in cartography [219], [59]; and demographic
modeling [ 2091.
Much of the digital image processing development wortko
date has been directed at removing the multifarious distortions
expected in the imagery and in mapping the results oan
standard reference frame with respect to the earth. This
process is a prerequisite not only to most automatic classification
tasks but also to mouf cthh e conventional visual photointerpretation
studies of the sort already successfully undertaken
with the Apollo and Gemini photographs [37].
The pattern recognition aspects of the environmental
satellite applications are largely confined to terrain classification
based on either spectral characteristics or on textural distinctions.
Object or target recognition as such is of minor importance
since few unknown objectso f interest are discernible
even at the originally postulated 300 ft per line-pair resolution
of the ERTS-A imaging sensors.
B. Plans for ERTS and Skylab
The ERTS satellitesw ill be launched ina 496-nmi 90-min
near-polar (99") sun-synchronous orbit. The total payload is
about 400 Ib.
The two separate imaging sensor systemosn ERTS-A (the
first of the two Earth Resources Techno!ogy Satellites) consist
1) of three high-resolution boresighted return-beam vidiFOR
EARTH RESOURCES 1179
cons sensitive to blue-green, yellow-red, and deep-red solar eter/scatterometer altimeter, will also be on board, as well
infrared regions of the spectrum, and 2) of an oscillating- as an optical telescope [215], [168].
mirror transverse-sweep electromechanical multispectral The multispectral data will be recorded on board in PCM
scanner with four channels assigned to blue-green, orange, on 20 000 BPI 28-track tape and returned with the undevelred,
and reflective infrared (.IR). ba nds. ERTS-B will carry a oped film at the end of each manned period of Skylab.
fifth hlSS channel in the thermal infrared.
The target of the vidicon tubes is exposed for a period of
12 ms/frame; the readout takes 5 s. This design represents a
compromise between the requirements of minimal motion
smear, sufficient illumination for acceptable signal-to-noise
ratio, and low bandwidth for transmis$ioonr recording. In the
oscillating-mirror scanner high signal-to-noise ratio is preserved
through the use of multiple detectors for each band.
The field of view of both types of sensors will sweep out a
100-nmi swath of the surface of the earth, repeating the full
coverage every 18 days-with 10-percent overlap between
adjacent frames. Every 100-nmi squarwei ll thus correspond to
seven overlapping frames consistingo f approximately 3500 by
3500 picture elements for each vidicon and 3000 by 3000 elements
for each channel of the mirror scanner, digitized at 64
levels of intensity.
The resolution on the ground will be at best 160 m/linepair
for low-contrast targets in the vidicon system a2n0d0 m/
line-pair in the mirror scanner [126], [159], [160], [12]. A
comparison of the various resolution figures quoted for the
Gemini/Apollo photography and for the ERTS/Skylab
sensors, and more pessimistic estimates of the resolving power
of the ERTS sensors, can be found in [34].
The pictures will be either transmitted directly to receiving
stations at Fairbanks, Alaska, Mojave, Calif., and Rosman,
N. C., if within range, or temporarily stored on video tape.
The vidicon data will then be transmitted in frequencymodulated
form in an analog mode while the scanner information
is first digitized and then transmitted by pulse-code
modultation (PCRI) [67]. Canadian plans to capture and utilize
the data are described in [198].
The center location of each picture will be determined
within one half mile from the ephemeris and attitude information
provided in the master tracking tapes which will also be
made available to the public.
The sources of geometric and photometric distortion and
the calibration systems provided for both sensors are described
in some detail in Section IV, where digital implementation
of corrective measures is considered. \Ye note here only
that estimates for digital processing on a n I B h l 360167 computer
of a single set of seven ERTS images ranges from 2 min
for geometric distortion correction only to 136 min for complete
precision