commonly grown in the region are compiled. Farmers are 
interviewed to learn about their crop rotation and planting 
practices and about their habits to subdivide fields from one 
season to the other. The area of each field is computed using 
the coordinates of the boundaries. 
At the start of the image analysis, the map and the image are 
geometrically registered. The agricultural region is separated 
from other land use by segmentation with the help of the 
digital map base. Other stratification can also be performed 
based on information stored in the GIS. The location of the 
training fields is recalled from the database and the spectral 
response statistics of each crop is generated. A per-field 
(polygon) classification is performed. Spectral response 
statistics are generated field-by-field and are compared, 
through an *IF-THEN" rule table, with the statistics of the 
training samples. This table contains the expected 
occurrence of a crop in a particular field as function of 
geographic location, ownership, soil type, topography, crop 
rotation system, etc. If a match is found with a crop on the 
list within a preset tolerance then the appropriate crop code is 
assigned to this field in the database. In fields, which 
exhibit bi-modal or multi-modal spectral distribution an edge 
detection routine is invoked to determine the new crop 
boundaries. 
In this approach, crop area estimates are based on the pre- 
computed areas of individual farm fields which is superior to 
the pixel count used in the previous method. Besides an 
aggregated figure for the whole agricultural region, area 
statistics can easily be generated for any geographic unit. 
Crop rotation information and land use change statistics can 
also be obtained through the database analysis module, if 
crop inventory from previous years is available. 
4. CONCLUSIONS 
The example presented in the previous section clearly 
illustrates the superiority of the knowledge based analysis 
approach to the conventional per-pixel classification. 
Although the formal requires more elaborate preparation, 
more accurate results are expected which can be stored 
directly in a GIS. Further analysis may be undertaken to 
generate information which is unobtainable or difficult to 
obtain from the conventional hard copy outputs. For 
reoccurring projects, the initial investment in setting up the 
digital database and knowledge base is well justified. 
The kind of knowledge based image analysis which was 
illustrated by the example is, however, only possible if 
image, cartographic and attribute data are merged within the 
same GIS and all the necessary tools are provided for their 
integrated processing and analysis. Research and 
development efforts at U.N.B. have brought this goal to 
realization. 
ACKNOWLEDGEMENT 
This development work has been funded under the Canada/ 
New Brunswick Subsidiary Agreement on Industrial 
Innovation and Technology Development. 
REFERENCES 
Derenyi, E., and R. Pollock, 1990. Extending a GIS to 
Support Image-Based Map Revision. Photogrammetric 
Engineering and Remote Sensing, 56(11): 1493-1496. 
Derenyi, E., 1991. Design and Development of a 
Heterogeneous GIS. CISM Journal ACGC, 45(4): 561- 
567. 
Hutchinson, C.F., 1982. Techniques for Combining 
Landsat and Ancillary Data for Digital Classification 
978 
Improvement. Photogrammetric Engineering and Remote 
Sensing, 48(1): 123-130. 
Foley, J.D. and A. Van Dam, 1984. Fundamentals of 
Interactive Computer Graphics. Addison-Wesley Publishing 
Co. 
Masry, S.E., 1982. CARIS: A Computer Aided Resource 
Information System. Presented at the Institute for 
Modernization of Land Data Systems, Georgetown 
University, Washington, D.C., January. 
Reedijk, W., 1990. The Design and Implementation of 
Raster Handling Capabilities for CARIS. M.Sc.E. thesis, 
Department of Surveying Engineering, University of New 
Brunswick, Fredericton, N.B., Canada 
Strahler, A.H., 1980. The Use of Prior Probabilities in 
Maximum Likelihood Classification of Remotely Sensed 
Data. Remote Sensing of Environment, 10 (1980): 135- 
163. 
USL, 1991. CARIS Products Description. Universal 
Systems Ltd., Fredericton, N.B., Canada, May.