ability 
d their 
ted by 
shown 
° more 
er was 
itively 
5” had 
1 PCA 
ctions. 
ations 
lower 
ted to 
ers of 
tively 
'uracy 
Cohen, Yafit 
  
Table 1 and 2 present confusion matrices of both unsupervised and knowledge-based classification. By using 
knowledge-based classification rules high classification accuracy was achieved for all crops, except for the clover. 
Despite significant improvement achieved for clover (>20%) it still wasn't recognized sufficiently. High confusion was 
found between clover and vetch. From experts and literature it was found that both legume types are often grown together 
in the same field. In addition, clover is hardly grown in Israel recently and most of its plots are very small. For these 
reasons it might be right to unite clover and vetch. A united class of them has high accuracy and reliability. 
Main improvements were achieved through imagery splitting criteria generalized by agricultural knowledge. Rainfall data 
was utilized mainly to overcome confusion between legume types, wheat and legume types and wheat and herbaceous. 
Most splitting criteria based on rainfall data were combined with NDVI imagery data. Confusion between orchards and 
NTF was considerably decreased (12%) with the usage of enhanced values of spectral PCA alone. Soil-based splitting 
criteria achieved an improvement of 8% in orchards recognition in comparison to unsupervised classification in a limited 
area. 
4 CONCLUSIONS 
The integration of imagery data, environmental properties, agricultural and expert knowledge through “split-and-merge” 
rules has high potential for high crop recognition in Mediterranean areas. The classification process included detailed 
unsupervised classification for the recognition of crop sub-features, inductive learning of relations exist between spatial 
data and classification refinement by knowledge-based rules. It seems that each phase in the process has a considerable 
contribution for high recognition quality. 
In this paper splitting criteria determination was made in hierarchical way. Imagery data got priority on environmental 
properties external to the image, in order to maximize image-based recognition and to utilize other information sources 
for refinements. In that way most rules were based on imagery data (spectral and temporal) and generalized with the 
incorporation of agricultural and expert knowledge. Environmental properties were shown to be very efficient in winter 
crops recognition and in distinguishing between crops and natural vegetation formations confusion. 
Fully established knowledge-based system is under development in order to utilize these findings and improving 
recognition quality. 
ACKNOWLWDGEMENT 
We wish to thank the Ministry of Agriculture of Israel for the Farmer organization of Israel for financing this project. 
5 REFERENCES 
Adinarayana, J. & Rama-Krishna, N., (1996).Integration of multi-seasonal remotely-sensed images for improved landuse 
classification of hilly watershed using geographical information systems. Int. J. of Remote Sensing 17:1679-1688 
Alty, J. L. & Coombs, M. J., (1984). Expert systems — concepts and examples. NCC publication, Manchester, England. 
Efrat, E., (1998). Towards land policy changes in Israel. Ofakim in Geography 48-49: 147-155 (in Hebrew) 
Frost, R., (1986). Introduction to knowledge base systems. Chapter 1 & Chapter 8. Macmillan, New York. 
Groom, B. G., Fuller, R. M. & Jones, A. R., (1996). Contextual correction: techniques for improving land cover from 
remotely sensed images. Int. J. Remote Sensing 17:69-89. 
Hinton, J. C., (1996). GIS and remote sensing integration for environmental applications. Int. J. GIS 10: 877-890. 
Huang, X. & Jensen, J. R., (1997). A machine-learning Approach to automateed knowledge-base building for remote 
sensing analysis with GIS data. PE & RS 63: 1185-1194. 
Janssen, L. L. F. & Middelkoop, H., (1992). Knowledge-based crop classification of a landsat TM image. Int. J. Remote 
Sensing 13: 2827-2837. 
Kartikeyan, B., Majumder, K. L. & Dasgupta, A. R., (1995). An expert system for land cover classification. IEEE Trans 
Geoscience & RS 33: 58-66. 
Kontoes,C. C., Wilkinson, G. G., Burrill, A., Goffredo, S. & Mégier, J., (1993). Int. J. GIS 7: 247-262. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 285