Hyperspectral Image Processing
Hyperspectral
images are now available for a wide range of applications: monitoring, mapping
or helping with disaster management.
A major
challenge lies in the interpretation of these images, a task which is referred
to in computer science as the classification of images. For each pixel, we are
provided with a set of intensities, one for each bandwidth. These intensities
are somehow related to the surface reflectance and hence to the type of land
cover or of land use. And instead of being able to model precisely this
extremely complex relationship between intensities and interpretation, the
scientific literature provides an abundance of techniques to capture
information from the data themselves with the help of the ground truth.
These
lectures aim at describing some of these techniques: what are their objectives,
what kind of information they use, how reliable are their predictions? To study
these techniques, we will consider toy examples, sometimes get involved in the
mathematical technicalities and sometimes consider simple algorithms. Some
ideas developed in these lectures come from textbooks for university students,
many others stem from research papers and related questions.
I would
expect these lectures to help getting more familiar with how proposed
techniques are described in research papers. Throughout these lectures we will
consider in the context of binary classification of hyperspectral
images the following issues: learning regarded as an optimization problem, can
we be positive about machine learning predictions, why is there a need for some
strange concepts? We will have a look at some segmentation issues stemming from
the computer vision community.
Assignment assignment.pdf
The evaluation will mainly take into
account the compliance with the assignment or at least an exact compliance with
part of what is requested. The formulas I am expecting are those that someone
implementing your proposition would have to use. They should be explained not
in terms of how in general one would use them but how they should be used when
implementing your proposition. I will take into account the following expectations
listed by order of decreasing importance.
0. The proposal should be clear enough
for someone not having taken the course, to be able to implement it. It should be
described using pseudocodes interacting together. The
computations involved in these pseudocodes should be described
in formulas. The notations need not be the same as in the lecture, however they
must be precisely defined.
1. Formulas even if they happen to be
wrong or contradictory, they should primarily make sense (for instance, a
matrix should not be added to a scalar, a parameter used as a counter in a
summation should not appear outside this summation).
2. Whereas in the literature, a single
technique may actually be used in different ways in different proposals, the report
should be unambiguous as to what exactly is being proposed.
3. What is claimed in the report in
terms of complying with the constraints should be correct.
4. The computations included in the
report should be correct (it is advised to make some simple numerical tests).
5. The technique used to fuse the
different sources of information (spatial context, different bandwidths...)
should be beneficial, in that performances should not
be increased when one source of information is lacking.
Notebook (only started) with in appendix, all
exercises and all Octave/Matlab
code to yield the displayed figures in the slides
Slides used during
the lecture
First lecture (March 13th 2024)
Second lecture (March 14th 2024)
Third lecture (March 15th 2024)
Fourth lecture (March 16th 2024)
This is my mail Gabriel.dauphin@univ-paris13.fr
(please mention HIP2 in the subject of emails).
GABRIEL
DAUPHIN