29 SUC- 
pspl, 
vledge 
e[i], 
 de[ i ] 
ition, 
! list: 
; Suc- 
pl, 
end of inherit oparation because knowl- 
edge node doesn't exist. 
else pl —p; 
step 1: 
p-uplink ; 
oparation of inherit knowledge; 
goto step 1; 
step 2 
if p=root node, 
stop inherit operation ‚exit; 
else goto stepl 
end of arlorithm. 
Using above algorithm ,a knowledge node 
can be retrieves from knowledge network, or 
added into knowledge network, of course, 
knowledge in parent node can be inherited by 
son node when requiring more knowledge in 
image interpretation. In this way, redundancy 
of knowledge storage is reduced and matched 
speed between data and rules is increased in 
small knowledge base. In knowledge base, 
knowledge of background and interpretation 
can be represented for produce rules and clas- 
sified into several categories according to task 
of interpretation (Qin, 1990). 
3. Intelligent interpretation based on model of 
correlation analysis 
As you know, landuse type is the geo- 
graphical complex, which has many featuresin 
different data plane of TM image. Suppose X 
is a landuse type which has N kind of features 
, X; is ith feature on its data plane. its overlay 
model is X =X; NX.NX:N NX. 
According to overlay model, many data 
planes of TM image are overlayed and formed 
mapping units (Burrough. P. A. , 1986). be- 
tween a mapping unit (the set of the point per 
region)and geographical data there is one —to 
— many relation in the database ,geographical 
data may be geographical attributes ,shape fea- 
tute and spatial relationship feature of landuse 
. the geographical attributes are provided to 
correlation analysis model. 
365 
The idea of correlation analysis comes 
from correlation of many geographical at- 
tribute of the same landuse in different data 
planes. For example ,in specail landform , lan- 
duse type exists a special soil type and vegeta- 
tion. the abstract model of correlation analysis 
is as follows; 
Suppose: there are quastions P — (Q,F) 
and P' — (Q', F^), in which Q is the set of 
possible appearing geographical facts in P, Q’ 
is the set of possible appearing facts in P', F 
and F’ are a kind of binary relation separately 
in Q and Q’,if exist a surjective map 
h: QQ' 
make any ordered pair (q;,q,) EF (qi, q;€ Q) 
if and only if Ch (q:) ,h (q,)) € F* that is, 
they exists a surjective map between F and F', 
Rr’: FOF’ 
then P' is regared as P quastion of homomor- 
phism, P is initial quastion of P',h is a homo- 
morphic mapping from P to P', 
notation as PSP’ 
By change of homomorphism, correlation 
analysis is changed into sign inference based 
on rules. 
Inference is data driven. A task of inter- 
preting image can be divided into several son 
tasks,the task will be implemented when all 
Black- 
board , common data storage area, is used to 
son tasks of the task are completed . 
store initial state, intermediate and last infer- 
ence result. A face can be read from and writ- 
ten into blackboard too. 
The task interpreter interprets and exe- 
cutes rules from knowledge network . A basic 
inference step has the following phases: 
(1) Matching: the data are sent to black- 
board ,the inference engine check the condi- 
tion parts of each production rule once again 
to see if the data match these rules. 
(2) Acting: when match occurs ,the rule 
is triggered and its operationis executed or its 
function for pattern recognization is called.