As a preliminary test of this approach, the average signal level was 
determined for each spectral band in the angle-corrected data for both 
data sets. The ratios of these signals were then computed. It was felt 
that these ratios could be used to make the adjustment described above. 
To check whether this would indeed work, signatures were extracted from a 
limited number of fields in Segment 203. On comparing the means of these 
signatures to those extracted from similar object classes in Segment 212, 
it was found that the ratios of the mean signatures for all objects were 
essentially equal to the ratios of average signal levels in each channel. 
Based on this limited substantiation of the approach, a recognition map 
was generated for Segment 203 using the adjusted signatures from Segment 
212. 
The six spectral bands which were used in producing this map were the 
same as were used in processing data from Segment 212. A detailed analysis 
was made of the recognition results in all the fields in a 3-kilometer 
portion of Segment 203. The recognition map for this area is shown in 
Figure 29. The field identifications for this area are included in 
Figure 30. 
The accuracy of the recognition results for Segment 203 data using 
signatures from Segment 212 data is less than that achieved on Segment 
212 data. For corn, the detection rate was 55%, while the false alarm 
rate was 10%. These figures for soybeans were a detection rate of 50% 
and a false alarm rate of 15%. Typical rates of correct classification 
for normal training within a segment was 80-85% in late July and early 
August. 
Considering the fact that the signals generated when viewing the 
two scenes were quite different, it is felt that the application of 
signatures from Segment 212 to recognize objects in Segment 203 was a 
reasonably successful effort. We believe that these results give reason 
for optimism regarding the feasibility of operational remote sensing 
crop survey systems. The signature extension capability illustrated 
here and made possible by careful feature enhancing preprocessing can 
eliminate a need for much costly ground observation and subsequent 
retraining of the processor. 
Of the many applications which have been successfully demonstrated 
to date I will present only one other example here. A growing body of 
literature is available which describes various applications in detail. 
A new technique for geological remote sensing has been developed by 
Vincent [57, 58]. The next three figures display infrared images of 
specially processed scanner data gathered by the University of Michigan 
aircraft at an altitude of 3000 feet over an area near Pisgah Crater in 
Southern California. The data was gathered at 0800 local time on October 
30, 1970. Each of the Figures present analog images of two single wave- 
length channels of data and an image of the ratio of radiances in the 
two channels. The ratio image is ‘sensitive to differences in chemical 
composition of the rock targets [59, 60].