412 
Using line printer maps, electrostatically printed gray-scale maps, and 
color CRT display maps, the results from the four techniques were ana 
lyzed and compared to selected ground-truth features. Both changed and 
unchanged features were evaluated to determine how well the given tech 
nique enhanced the temporal information. The results from factor rota 
tion on the remaining test sites were evaluated by visual interpretation 
of the corresponding aerial photography. 
Another variation of factor analysis was also attempted on the initial 
test site. The rotation of the underlying factors was performed on 
the first three principal components and on just the first two principal 
components. Because the first two or three principal components often 
account for the preponderance of the variation in a data set, this 
method may eliminate noise in the data and may lead to a more useful 
transformation. Again, the results were compared to the ground-truth 
features. 
5. RESULTS 
After reformatting and geometric correction of the data set, the four 
techniques for detecting change were applied to test site 3, the sewage 
treatment facility. The final results of each method were displayed 
and visually compared to determine which method detected change correct 
ly and consistently. Changed and unchanged features, field-verified by 
the DMA for test site 3, were used as references for correct and consis 
tent change-detection. The changed features included equalization 
basins, new secondary clarifiers, a detention pond, new multi-media 
filters, and new lime reaction tanks. The unchanged features included 
forested areas, grass areas, gravity sludge thickeners, and the prime 
clarifier building. 
The first method, post-classification comparison, gave poor results. 
The main reason was poor classification for both the early and late 
scenes. The BW scene, which was density sliced into 9 classes, pro 
duced a photograph-like map (Fig. 1). Many of the cover types, however, 
were confused because their photographic densities overlapped or were 
very similar. This was particularly true with grass areas, roads, and 
the secondary clarifiers, all having similar density levels. This prob 
lem also occurred in the color scene classification. Several of the 13 
Figure 1. Classification of the black-and-white scene.