AUTOMATIC EXTRACTION OF INFORMATION FROM 
MULTISPECTRAL SCANNER DATA * 
Jon D. Erickson 
Environmental Research Institute of Michigant 
P.O. Box 613 
Ann Arbor, Michigan 
Abstract 
The present status of automatic multispectral scanner data processing 
and interpretation is evaluated with problem areas and suggested approaches 
discussed. The rationale for automatic information extraction is given. 
The basis of spectral discrimination is examined indicating sources 
of systematic variability which tend to obscure the spectral attributes 
of the classes of interest. 
Automatic processing functions, techniques and methods, and equipment 
are discussed with emphasis on preprocessing for feature enhancement 
and invariance, which accounts for sources of systematic variability and 
reveals the underlying class signature otherwise obscured. Techniques for 
carrying out the major functions of preprocessing, signature extraction, 
discrimination, display, and post-processing are examined in detail. 
Procedures are reviewed for adaptive discrimination and for estimating 
the proportions of classes within a single ground resolution element. A 
special purpose multispectral parallel-channel processor system is 
presented. 
Applications and results are discussed briefly in terms of agricul- 
tural and geological surveys to indicate the substantial success achieved 
to date. 
It is concluded that despite the potential demonstrated in the many 
successful feasibility tests in various application disciplines that 
operational uses of information derived from multispectral sensing of 
earth resources and land use will await development of cost effective 
processing systems. 
  
*Supported in part by NASA Contract NAS9-9784. 
tFormerly the Willow Run Laboratories of the Institute of Science and 
Technology of The University of Michigan. 
1