552 
AIRBORNE IMAGING SPECTROMETER DATA ANALYSIS 
APPLIED TO AN AGRICULTURAL DATA SET 
K. Staenz and D.G. Goodenough 
Canada Centre for Remote Sensing 
2464 Sheffield Road, Ottawa, Ontario, Canada K1A 0Y7 
ABSTRACT 
This paper describes the ISDA (Imaging 
Spectrometer Data Analyzer) preprocessing steps 
necessary to bring the imaging spectrometer data 
acquired with the Programmable Multispectral 
Imager (PMI) over an agricultural test site near 
Zurich, Switzerland to an acceptable data quality 
level for information extraction purposes. Such 
preprocessing steps include data quality assess 
ment, correction of data in the spectral, as well 
as in the spatial domains, the proper radiometric 
alignment of the five PMI cameras, and relative 
and absolute normalization. Special emphasis was 
directed towards the viewing angle effects result 
ing from the large PMI scan angle range of 72.5 
degrees. The removal of this effect is very 
important for information extraction purposes with 
respect to spectroscopic analysis and classifica 
tion of airborne imaging data. Preliminary work 
in the classification of imaging spectrometer data 
of agricultural targets (e.g. wheat, corn, 
potatoes, etc.) is presented. The two most likely 
methodologies for classification of such data, 
feature selection followed by classification and 
'full' spectrum classification, will be discussed 
with special emphasis on data reduction using the 
band-moment analysis approach. 
Key Words: Imaging spectrometer, data quality 
assessment, radiometric camera 
alignment, data normalization, data 
reduction, band-moment analysis, 
feature selection, classification. 
INTRODUCTION 
Application of remote sensing technologies to 
agriculture has resulted in numerous investiga 
tions of the spectral behaviour of agricultural 
objects (Buechel et al., 1989; Horler et al., 
1984; Brown et al., 1980). Most of the remotely 
sensed data acquisition to date has been conducted 
within well-defined, non-contiguous bands. There 
have also been efforts made to collect spectral 
data with spectroradiometers on the ground in 
order to generate a full spectrum of the 
agricultural object. Unfortunately, such ground- 
based measurements usually do not provide 
sufficient spatial coverage under similar 
illumination conditions. In order to combine the 
spatial as well as the spectral resolution and, 
therefore, to increase the potential of remote 
sensing substantially, a new generation of sensor 
systems, such as the airborne imaging spectrometer 
(Vane and Goetz, 1988), has been developed. 
These instruments, such as NASA's AVIRIS (Airborne 
Visible/Infrared Imaging Spectrometer) or 
Moniteq ' s PMI ( Programmable Multispectral Imager), 
acquire data simultaneously in many narrow 
spectral bands (up to 288), usually covering the 
VNIR/SWIR (Visible and Near Infra-Red/ShortWave 
Infra-Red) region of the electromagnetic spectrum. 
The volume and complexity of imaging spectrometer 
data, however, require new approaches in image 
processing, especially in data handling, prepro 
cessing, and information extraction. These areas 
are addressed in the Imaging Spectrometer Data 
Analyzer (ISDA), a software package developed at 
the Canada Centre for Remote Sensing for an 
efficient analysis of imaging spectrometer data 
(Staenz and Goodenough, 1989). 
Discussed in this present study are the results 
of an analysis involving an agricultural data set 
with emphasis on the different ISDA preprocessing 
steps required to bring the imaging data to an 
acceptable quality. These steps are extremely 
important for information extraction purposes and 
the utilization of the full potential of imaging 
spectrometer data. Preliminary results of a 
classification approach are also presented. 
DATA ACQUISITION AND TEST SITE 
The data set was acquired with the PMI sensor over 
an agricultural test area, approximately 30 km NW 
of Zurich, Switzerland on July 28, 1986 by the 
German Aerospace Research Establishment (DLR). 
The PMI sensor characteristics are summarized in 
Table 1. This instrument works in a rake mode 
providing a total of 40 profiles over the wave 
length range from 430 to 805 nm in 288 contiguous, 
2.6 nm wide, spectral bands sampled at 1.3 nm 
intervals (Hollinger and Gray, 1987). The over 
flight was conducted under nearly cloud-free 
conditions (horizontal visual range approximately 
15 km) from an altitude of 500 m resulting in a 
swath width of approximately 730 m by a scan angle 
of 72.5 degrees as well as a ground resolution at 
nadir of 3.3 m across and 10.4 m along track. The 
aircraft heading is approximately 33.0 degrees 
azimuth and the solar illumination geometry is 
given by a zenith angle of 29.2 degrees and an 
azimuth of 194.3 degrees. 
Major crop types within the flat test site include 
wheat, barley, corn, and potatoes. Ground 
reference information, such as plant height and 
density, age (planting data), degree of fertiliza 
tion, etc., was collected for these different 
object types. In addition, black and white, as 
well as false colour, aerial photography was used 
in support of the ground data collection, for 
preprocessing of the imaging spectrometer data and 
for information extraction. 
PREPROCESSING 
Problems were encountered during the process of 
analyzing the 16-bit PMI data set with respect to 
the data quality of the sensor, calibration and 
the aircraft attitude variations (Borstad et al., 
1985). In order to maximize the accuracy of 
information extraction, it is important to address 
these problems. Figure 1 summarizes the