891 
detection of those fields in which two or more crop types are 
grown. Based on updated DFBM this way, two classification schemes 
have been developed (Csornai et al, 1988). The first one is a 
simple reclassification[REC] of a per point classification (e.g. 
maximum likelihood CMXLj) on the basis of majority of classified 
pixels within the field boundaries, the second incurs information 
on boundaries at the very beginning of the classification, 
collecting the statistics based on pixels each field (Table 3). 
Categories 
#FieIds 
Area (Ha) 
Accuracy 
CMXLj 
in percent 
CRECj 
Winter wheat 
9 
746 
84.4 
100 
Alfalfa 
1 
54 
60.6 
60.6 
Corn 
14 
764 
69.5 
87.6 
Sugar-beet 
3 
168 
60.2 
71.7 
Peas 
3 
88 
86.6 
100 
Pasture 
4 
749 
64.4 
86.7 
Settlement 
1 
187 
71 
100 
Total 
3b 
2756 
72 
90.7 
Table 3. Table of accuracy in classification for a sample 
co—operative (Foldes) 
Both these two schemes proved to be effective. Based on some 
preliminary results (Csornai et al, 1989) it can be 
concluded that in inventory type of surveys the basic 
classification accuracy can increase by 10—12 percent 
(Table 4). 
Categories 
# p i x e 1 s 
Per—point 
CPPC) 
Reclass. 
CREC] 
Per—field 
CPFC] 
Wheat 
41150 
87 7. 
91 7. 
98 7. 
Maize 
37442 
73 7. 
78 7. 
90 7. 
Sugar—beet 
12485 
65 7. 
83 7. 
81 7. 
Potato 
4186 
67 7. 
83 7. 
77 7. 
Alfalfa 
3687 
65 7. 
B5 7. 
96 7. 
Soil 
2026 
87 7. 
91 7. 
45 7. 
Water 
3568 
71 7. 
100 7. 
76 7. 
Average PCC 
80.1 7. 
86.8 7. 
91.6 7. 
Table 4. Comparison of the classification accuracies on a 
small area.