Table 3: Estimates of CIU for each crop type
Oats Wheat Peas Canola Canary Barley Flax Fallow
T 2 5 2 5 2 5 2 5 2 5 2 5 2 5 2 5
A 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
S 0 8 171 [691 [70 |365 |191 |516 |17 |74 |188 |740 |21 110 [45 1180
Sd 0 3 0 5 0 46 10 55 16 12" l0 13 TO 23 10 24
CIU * * 3 * 58 [.87. 1.79 ].89 j* * 79 1.95 1.29 |.74 1.41 1.83
* means that the CIU is not calculated, since there is no rule for the corresponding crop.
Table 4: Crop Rotation Rules with Adjusted Certainty Factors Based on Two
OW
Years' Inven
Data
Table 5: Crop Rotation Rules with Adjusted Certainty Factors Based on Five
Years' Invento
Data
Table 6: Accuracy of classification results
Channel Time Basis MLM 2 year 2 year * 5 year 5 year *
whole 57.8% 57.4% 63.5% 64.5% 64.9%
CHH 1989 field set
Rule applied | 55.1% 54.5% 63.1% 64.6% 65.2%
field set
*- where the adjusted certainty values were used in the classification.
results in a slightly lower classification accuracy than
that of the MLM method; however, the accuracy is
significantly increased (by 8.6%) after the certainty
factors are adjusted using the values of CIU. This
suggests a possibility that the less time periods are used
in eliciting time-dependent knowledge, the greater
increase in classification accuracy could result through
the consideration of the UIU problem.
CONCLUSIONS
Consideration needs to be given to the Uncertainty In
Uncertainty (UIU) problem existing in the knowledge
either generated from databases or provided by human
experts. A model has been developed in this paper in
order to estimate the UIU values. Methods have also
been addressed for estimating the variables involved
in the model. A case study has shown that the
proposed model is effective in improving classification
accuracy based on multiple knowledge sources.
Further research is needed to estimate the reliability or
accuracy of time-dependent knowledge provided by
944
human experts. More experiments are also needed to
further test the effectiveness of the proposed model.
REFERENCES
Frost, R. A., 1986, "Introduction to Knowledge Based
Systems". William Collins & Co. Ltd., NY.
Huang, S & A. Zhang, 1991, "Improving Crop
Classification Accuracy of SAR Imagery by Introducing
Crop Rotation Knowledge Using AI Methodology", A
technical report prepared for Canada Center for
Remote Sensing, 36p.
Kenk, E., M. Sondheim, & B. Yee, 1988, "Methods for
Improving Accuracy of Thematic Mapper Ground-
Cover Classifications", Canadian Journal of Remote
Sensing, 14(1), pp. 17-31.
Middlekoop, Hans & L. L. F. Janssen, 1991,
"Implementation of Temporal Relationship in
Knowledge Based Classification of Satellite Images".
Photogrammetric Engineer & Remote Sensing, 57(7),
pp. 937-945.
