ORTHOPROJECTION TESTS OF HYPERSPECTRAL DATA IN STEEP SLOPE ZONES 
Piero Boccardo*, Enrico Borgogno Mondino*, Mario A. Gomarasca**, Luigi Perotti*** 
* DIGET, Politecnico di Torino, C.so Duca degli Abruzzi 24, 10129 Torino (ITALY) 
e-mail : enrico.borgogno(@polito.it, piero.boccardo(@polito.1t 
** CNR-IREA, Sezione di Milano, Via Bassini 15, 20133 Milano (ITALY) 
e-mail: gomarasca.m(@irea.cnr.Iit 
***Dipartimento di Scienze della Terra, Universita di Torino, Via Valperga Caluso 35, 10125 Torino (ITALY) e-mail: 
luigi.perotti@unito.it 
KEY WORDS: Remote Sensing, Image Geometry, Orthoprojection, whiskbroom, accuracy, Hyper spectral, Neural network, RFM 
ABSTRACT : 
Nowadays hyperspectral data are really important in the environmental field. While the advantages, due to their radiometric features 
are broadly documented a rigorous metric verification is still abset 
steep slopes produces strong deformations on the images that h 
nt especially in uneven areas (mountains) where the presence of 
ave therefore to be preventively corrected. The classification 
potentialities and the high number of bands of the hyperspectral data is already known by operators. Nevertheless these risk to rest 
unusable if a good correction of the scenes geometry is not guaranteed. The mountain zones represent a critical benchmark for both 
warping and orthoprojection algorithms; therefore they have been chosen during this study. Positioning accuracy (planimetric) tests 
have been conducted on airborne sensor MIVIS images (Multispectral Infrared Visible Imaging Spectrometer). Such system is based 
on the whiskbroom digital acquisition technology. The rigorous definition of the projective model still remains an open problem 
subordinated to the external orientation auxiliary data. However the sensor model problem cannot be neglected because of the strong 
geometric deformations of the images that can make them useless or improper for the mapping scales suggested by their average 
geometric resolution. This study shows some orthoprojection results obtained both by commercial software and by autonomous 
procedures (developed by the authors) based on self-calibrating Rational Function Model and on Multi Layer Perceptron neural 
network. A comparison among the different methodologies has been conducted taking care of the geometric accuracy in order to 
define the most appropriate map scale they could be addressed to. A MIVIS image has been select for the test, recorded with an 
across-valley flight on the middle Valley of Susa (Turin-Italy), where the elevation range is about 1800 meters. 
1. INTRODUCTION 
1.1 Main purposes 
This work presents some results about geometric correction of 
MIVIS sensor images acquired over a mountain region where 
image distortions are really heavy and they cannot be neglected. 
Two non-parametric correction methods have been selected: 
RFM (Rational Function Model) and a prototypal one based on 
a NN (Neural Network) approach. Both of them relate image 
and terrain coordinates considering also the terrain elevation 
data for relief displacement minimization. 
Tests demonstrate that many factors determine the quality of the 
final result as the number and the distribution of the GCP 
(Griund Control Points) DEM quality, algorithm 
implementation, presence of unknown distortions. Non- 
parametric methods, alternative to the rigorous ones, allow to 
proceed to correct all the distortion globally without knowledge 
of any auxiliary flight information. 
Different geometric correction tests have been performed in 
order to define how the two methods perform varying internal 
parameters. This is intended to state their reliability for an 
operative approach. 
1.2 MIVIS hyper-spectral sensor 
LARA-CNR (Aerial Laboratory for Environmental Researches 
of Italian National Research Council) has developed and 
assembled the MIVIS hyper-spectral sensor; it operates with 
high geometric and spectral resolution (depending on the flying 
height and generally ranging from 5 to 2m). The MIVIS 
whiskbroom sensor can record 102 bands. All the bands belongs 
to the spectral range from the visible to the thermal infrared, 
including the near and medium infra-red. 
MIVIS is a modular instrument consisting in four spectrometers 
which simultaneously measure the radiance from the Earth's 
surface for a total of 102 spectral bands: 20 in the visible 
spectral region (0.43-0.83um), 8 in the near infrared (1.15- 
1.55um), 64 in the middle infrared (2.0-2.5um) and 10 in the 
thermal infrared (8.2-12.7um). The chosen spectral ranges can 
satisfy operational requirements for botany, agrarian sciences, 
geology, pedology and all the sciences dealing with the territory 
surveying. As far as geology/geomorphology is concerned, if 
rigorous geometric correction and spectral calibration have 
done, MIVIS sensor can be effectively be used for analyzing 
evident superficial features (qualitative and quantitative 
approach ) that could be involved in some type of unstable 
phenomena. 
  
Lower Upper 
Channels Limit (nm) Tem Band (nm) 
1 
2 21-28 8 1150 1550 50 
3 29-92 64 1983 2478 9 
4 93-102 10 8180 12700 3407540 
  
Table 1 - MIVIS images features. 
872 
Ini 
  
  
Th 
fro 
abc 
su 
mo 
Of 
for 
an 
cor 
At 
  
te: 
50r 
and 
cor 
RF] 
Ort 
mel 
(Ini 
and 
lan; 
the 
suc