pean 
siteit 
pport 
yéen. 
ipact 
ions, 
3 the 
Ig in 
f the 
onal 
CD- 
itat 
nce; 
pact 
sin, 
ing. 
teit 
Bethel, James 
  
GEOMETRIC REGISTRATION AND CLASSIFICATION OF HYPERSPECTRAL AIRBORNE 
PUSHBROOM DATA 
J. S. BETHEL', C. LEE , D. A. LANDGREBE" 
"Purdue University 
Geomatics Area, School of Civil Engineering 
"Purdue University 
School of Electrical and Computer Engineering 
bethel. changno, landgreb(@ecn.purdue.edu 
Working Group III/5 
KEY WORDS: Mathematical models, Orientation, Hyperspectral, Image registration, Remote sensing, Multispectral, 
Classification 
ABSTRACT 
Innovative geometric modeling techniques involving stochastic constraints and linear feature exploitation have been 
demonstrated to yield good results in the rectification of hyperspectral airborne pushbroom imagery. The unique 
aspects of the platform trajectory are particularly well addressed by these techniques. Supervised statistical pattern 
recognition techniques have been developed specifically to address the unique and challenging aspects of high 
dimensional data. These have resulted in processing strategies which are compatible with common desktop computing 
resources. The combination of such thematic class extraction with effective rectification algorithms delivers a powerful 
tool into the hands of those building urban databases. 
1 INTRODUCTION 
Hyperspectral data analysis has repeatedly been shown to yield high quality thematic class maps in urban areas. 
Rigorous geometric sensor and platform modeling have been shown to yield high quality geometric registrations with 
such imagery. But it has been a rare occurrence for these results to be achieved simultaneously. By using an 
interdisciplinary approach to this problem, we have developed a set of interdependent algorithms which collectively 
produce accurate thematic material and region delineations. Other experience has also confirmed that such 
interdisciplinary data fusion is the way that such analysis should be carried out. Data used for the following analysis 
was acquired by the HYDICE (Hyperspectral Digital Imagery Collection Experiment) sensor. It is a pushbroom 
airborne imaging spectrometer with 210 spectral channels ranging from 0.4 to 2.5 micrometers. It has an instantaneous 
field of view, IFOV, of 0.5 mrad. The ground sample distance, GSD, is generally 2-3 meters. It operates with a gyro 
stabilized platform referred to as the FSP. 
2 GEOMETRIC REGISTRATION 
The basic principle of the geometry in a pushbroom imaging system can be explained by the collinearity condition, 
which requires that a light ray from the ground object to the 
image point should be a straight line. In this case, we need six 
Perspective elements of exterior orientation (EO) for each scan line 
ee because the pushbroom image is collected sequentially line by 
line. Consequently so many parameters may need to be solved 
for that the problem becomes infeasible due to the number of 
required control data in most cases. This situation can be 
addressed by using a priori information about the behavior of 
the platform. 
  
Instantaneous 
2.1  Collinearity Equations 
The geometric relationship between the ground point and 
image point of HYDICE imagery can be simplified as shown 
in Figure 1 for a given scan line. From Figure 1, the 
  
  
  
  
Figure 1 HYDICE imaging system 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 183