739 
c cali- 
ter-type 
nd forest 
tointer- 
ilopment 
. of MSS 
automated 
;he Envir- 
:torate, 
Ltional 
ilture, and 
.tre for 
:nown by 
¡ification 
and 0VAAC8. 
:lassifi- 
.versity of 
Iunday 
Langham 
l) . It is 
! many 
lalysis of 
fe software 
reloped at 
Saskatchewan. 
;sif ication 
:er) 
: interact- 
>r operation 
i image 
the MICA 
.es are 
astern book- 
Le or modify 
>eful if 
:h prevent 
ist because 
> only 6000 
[) . A look- 
same time 
The MICA 
:tern re 
cognition of five band imagery. 
The Laboratory for Application of Remote Sensing (LARS) 
of Purdue University is well known for its excellent research on aut 
omated classification systems. Dr. Wacker developed a number of pro 
grams for the LARSYS system. He has modified the LARSYS system to 
run on an IBM 370/155 computer at the University of Saskatchewan. 
Most of the LARSYS system is written in Fortran. New programs and 
algorithms can easily be added. Presently available system functions 
(Crain 1974) are given in Table III. The advantages of the LARSYS 
system are its flexibility and its power as a research system. LARSYS 
requires a sophisticated user, knowledgeable in automated classifi 
cation methodology. The Fortran flexibility of the system means 
that the LARSYS programs are quire slow for classification of an 
ERTS frame. Crosson, Peet and Wacker (1974) have successfully used 
the LARSYS system to carry out classification of agricultural fields 
in Saskatchewan. 
Hardware Systems : The most developed hardware system in 
Canada is the CCRS Image 100 system. The Image 100 is intended to 
complement the photointerpretative powers of the users. For that 
reason the system is highly interactive. The system functions are 
given in Table IV. One can classify one ERTS-1 frame, including 
training time, in approximately eight hours. Digital multispectral 
data are loaded from tapes into a solid state memory unit. The user 
selects a training area and the resulting classification and signa 
tures are displayed in four seconds. Preprocessing, such as ratio- 
ing or normalization, and transformations, such as Hadamard or com 
ponents analysis, may be performed in near real-time. Up to eight 
themes may be classified at one time. The resulting classifications 
can be stored on magnetic tape for film production on the CCRS EBIR's. 
Temporal data can be combined to increase classification accuracy. 
Results on a variety of terrain (Economy et_ ad 19 74) indicate that 
classification accuracies of greater than 80% can be routinely achie 
ved. Although the design of the system favors supervised classifi 
cation techniques, one can also carry out clustering analysis of the 
image. 
Two industrial firms, Computing Devices Company and OVAAC8, 
are presently developing hardware interactive systems for image pro 
cessing. Although a number of university investigators have ex 
pressed interest in such hardware based classification systems, 
the high capital costs have prevented the development of such sys 
tems in those institutions which do not have the processing demands 
to achieve the large savings possible through economies of scale.