5.9. EXERCISES AND PROJECTS 197
existingfile? Yes or Noî such a response would have to be in Braille for
the operator to choose the desired option. There is much room in this
project to explore the ability of neural networks. The student is urged to
be creative.
15.Project: Several of NASAís planetary missions involve precise and safe
landing of the spacecraft on unexplored planetary surface. As part of
the Mars Landing Program, NASAís Jet Propulsion Laboratory has a
simulated Mars yard where rocks of different sizes are strewn across the
yard to emulate the surface of Mars. This simulated environment provides
a test bed for examining the capability of a lander to select an appropriate
landing site during its descent on to the surface of Mars.
We can describe briefly the process by which a lander may safely land
on the planet Mars. Upon reaching Mars, the Mars orbiter would map
the preselected landing site using its diversity of sensors. Due to the
distance from which such mapping is done, obviously there is not enough
resolution to identify rocks and other debris that may be present in the
selected landing site. Once the lander has been deployed, sensors on board
the lander would begin to acquire visible and radar images of the landing
site and begin to process in real time the ìqualityî of the landing site. The
objective of course is to distinguish ìrockyî regions from ìsmoothî regions
and to provide navigational corrections to the lander as it approaches the
surface. This requires a real-time adaptive neural network processor that
can continuously update itself.
In this project, we expect the student to be very innovative in how to deal
with a simulation-based study of a landerís ability to distinguish ìrockyî
regions from ìsmoothî regions. With the availability of digital cameras
nowadays, it should be easy to acquire images of different resolution and
develop a set of training patterns. We leave the mechanics of data acqui-
sition to the curious and inquisitive mind. Thefirst step however, is to ex-
amine the viability of feedforward neural networks to perform adequately.
We strongly encourage exploration of other neural network architectures
that utilize radial basis functions, principal component analysis, and other
techniques that may be more efficient in real-time training and also allow
for adaptive capabilities.
16.Project: The trend in modern agricultural methods involve the analysis
of multispectral and hyperspectral data of agricultural ecosystems ob-
tained from spaceborne and airborne systems. There is quite a lot of
research activity in this area at NASA as well as at universities and pri-
vate institutions. The primary objective in this analysis is to identify
thehealthoftheecosystembyclassifyingtheìstateîoftheecosystem.
Healthy vegetation, for example, absorbs most of the radiation in the
visible spectrum and reflects only a small portion of the visible spectral
band. Also, healthy vegetation reflects a large portion of the near-infrared
portion of the spectral band. This is indicative of the process of photo-
