Some classes
ng IAS may not
the resource
several forest
;roup.
or raster form
rpretation to
classes being
nay suffice to
y spectral and
classify some
'mportant here
osts can rise
Is used in the
:, the GIS and
ystems. Both
acities , many
ays. Analysts
least one year
relating maps
:nd 4. These
g (artificial
use of expert
!, but adds to
m applications
YSTEM (LDIAS)
launched the
, the Canada
ving data from
(TM) and the
ensor produces
the MSS. The
nough, 1977),
handle a full
ge power were
analysis with
to begin a
TM analysis,
research and
, the LANDSAT
I , which would
(185 km by 185
ile permitting
In addition,
optical scan-
. From this
my activities
nd functional-
Figure 2. The
1, is a Digit-
11/785 with 8
of disk stora-
ch with 512 by
a IP8500 dis
es II displays
to a Gandalf
ts one to work
al. The next
2, a DEC VAX
1.6 gigabytes
:put subsystem
on. There are
¡ach with two
speed graphics
processing, there is a Graphics Processor. One can
connect to a remote GIS through an automatic dial-up
device. LDIASl and LDIAS2 are connected together
via an Ethernet network, two dual-ported disks, and
a dual-ported tape drive unit. The two computers
are not clustered together because: a) this was not
an available product from DEC when the project
started; b) the two computers are at different oper
ating system versions. The VMS operating system for
LDIASl is changed with each release from DEC; that
for LDIAS2 is changed with less frequent releases
from Intergraph Corporation.
The third computer in LDIAS is a DEC VAX 11/730
which is the control computer for the Fast Multi
dimensional Processing System (FMPS). FMPS is the
batch production system for full scene analysis.
Interactive analyses are conducted on LDIASl and
LDIAS2. FMPS was developed for CCRS by Canadian
Astronautics Limited (CAL). FMPS has an Aptech 24-
megabyte-per-second bus with a mass memory of 3
megabytes. Attached to this bus are the VAX 11/730,
a 100-megaflop array processor (Star Technologies
ST-100), a CAL Parallelepiped Preprocessor Unit
(PPU), and two 256-megabyte disk drives which are
dual-ported with LDIASl. FMPS also has an Ethernet
connection to the other computers.
Although not formally part of the LDIAS project,
there are three DEC AI VAXstations connected to the
same communications network. Each workstation has a
71-megabyte disk, a tape cartridge unit, and a bit
mapped display. These workstations use VMS as the
operating system and support rapid development of
expert systems in Prolog or Lisp. The expert sys
tems discussed in section 5 make use of this
hardware.
The system architecture is intended to support the
software functions briefly described below.
2.3 Major Software Functions
The LDIAS is used to support research and develop
ment in remote sensing, computational vision, sensor
development and GIS integration, and to support the
development of applications for renewable and non
renewable resources. This latter group use one of
the two analysis workstations. A workstation
consists of an image display, a map display, a
digitizing tablet, and two terminals. The work
station is operated for applications development by
a team of two: a resource expert (user), and an
analyst. The analyst is knowledgeable about the
software, hardware, and analytical procedures.
The primary remote sensing inputs to LDIAS are
geocoded TM computer compatible tapes (CCTs) from
the MOSAICS system (Link et al 1985). These geo
coded products correspond to four 1:50,000 map
sheets in the Universal Transverse Mercator (UTM)
projection. The TM scan lines are oriented east-
west with north at the top of the corrected image.
The image has been resampled on a regular, 25m, UTM
grid with a 16-point truncated sine function.
Digital maps come from provincial agencies or from
the Surveys and Mapping Branch of our department.
The LDIAS software is run using a supervisor
program. The user is presented with the Master Menu
shown in Table 1. Each item in this Master Menu is
itself a menu of tasks, or menus, or both. The user
can obtain help in the form of on-line documentation
at any step, even if he or she is in the middle of a
program. All user responses to program queries can
be recorded and subsequently used to produce an
automatic sequence of task executions; that is, a
batch command file. A man-machine interface was
developed to support bilingual dialog (English and
French), scrolling of user sessions on the terminal,
interactive or batch modes, standard prompts,
on-line help, and effective error recovery. An
image data base structure, UNIDSK, was implemented
to support images with 8000 by 8000 pixels, and 21
channels, each of which could be quantized from one
bit to 128 bits in integer or real formats. In
order to support our research, flexibilty in the
software was very important. LDIAS currently con
tains more than one million lines of code,
documented and debugged.
TABLE 1
LDIAS MASTER MENU
Label Description
A Input
D Radiometric Corrections
F FMPS - Fast Multidimensional
Processing System
G Geometric Corrections
I Generate Spatial Texture
J Acquire Spectral Signatures
K Perform Segmentation
L Classify and Cluster
M Map Input/Output
N Filter and Enhancement
0 Diagnostics
Q Accuracy Assessment
S Classification Filtering
U Utilities
Z Output
The structured design and analysis methodology of
Gane and Sarson (1977) was used for linking
research activities and software development.
Logical data flow diagrams for all of the software
are on-line. The software is managed through a
central computer repository. A major part of the
documentation is automatically extracted from
software for on-line help and generation of
technical memoranda. In this way, we avoid most
incompatibilities between software versions and
corresponding documentation versions. For some
pixel processing operations, we have found a stan
dard program shell or skeleton useful. This image
processing skeleton is integrated with an editor
with knowledge of the computing language, FORTRAN
77. Through these techniques the LDIAS project
team's software productivity has been greatly
enhanced.
An analysis project begins with the user specify
ing their desired outputs. This specification
establishes the most probable analysis procedure.
The TM or MSS imagery are read in, together with
digital terrain models if they exist. Corrections
can be applied for radiometry, atmospheric effects,
viewing geometry, image scale, and output project
ions. Identification of training and test areas can
be made on the display of the TM imagery, from paper
maps or photographs, or from GIS files. For TM, it
is essential to utilize spatial features, in
addition to the spectral features. Segmentation can
be performed using these spatial and spectral
features. Parametric and non-parametric, supervised
or unsupervised, classifiers can be used. The LDIAS
supports up to 256 classes at any one time for
remote sensing imagery. There is a wide range of
enhancement tasks available. The outputs may be
digital updates to the GIS, digital or paper maps,
area summaries, or photographs. At each stage in
the user session, the user or analyst can assess the
accuracy. The theoretical estimate of the classifi
cation accuracy to be achieved, for example,