ΕΕΛΛΑΚ - Λίστες Ταχυδρομείου

Interest and questions about "Open-Source AI Framework for Thermal Satellite Payload Data Analysis"

  • Subject: Interest and questions about "Open-Source AI Framework for Thermal Satellite Payload Data Analysis"
  • From: Nikitas Karachalios <karachaliosnikitas [ at ] gmail [ dot ] com>
  • Date: Mon, 2 Mar 2026 17:07:16 +0200
Dear Christos, Alexis, and Simon,

My name is Nikitas Rafail Karachalios. I am a 4th-year undergraduate
student at the
Department of Informatics and Telecommunications, National and Kapodistrian
University of Athens, specializing in Machine Learning and Deep Learning. I
am
writing about the "Open-Source AI Framework for Thermal Satellite Payload
Data
Analysis" GSoC 2026 project, as I would love to contribute to it.

My most relevant work is a comparative study of deep learning methods for
satellite image classification on EuroSAT
(
https://github.com/NikitasKrh/Modern-Deep-Learning-Methods-A-Comparative-Study-on-Satellite-Data
).
It covers multiple learning paradigms (supervised, self-supervised,
few-shot,
zero-shot, parameter-efficient fine-tuning), all implemented end-to-end in
PyTorch. The focus was not just on training and comparing models, but on
understanding why they behave the way they do. Each experiment is followed
by a
detailed Q&A section that analyzes the results, explains failure cases, and
connects findings across methods, such as linking representation-level
analysis (t-SNE, cosine similarities) back to classification errors. This
kind
of analysis is something I genuinely enjoy and want to develop further.

I have been going through the TIRAuxCloud codebase carefully, tracing the
data
pipeline (multi-band GeoTIFF loading, auxiliary meteorological channels
stacked
alongside thermal bands, per-band normalization) and the different
segmentation
architectures the project supports. What stood out to me is that adding
auxiliary
meteorological data consistently improves cloud detection over using thermal
bands alone, since it shows that context beyond the raw spectral signal
really
matters. The whole approach of working in thermal infrared instead of
optical
(night-time capability, no snow-cloud confusion) was new to me and sparked
my interest.

I also looked at some of Orion-AI-Lab's other projects, KuroSiwo for
SAR-based
flood mapping, Sen4AgriNet for Sentinel-2 crop segmentation, which gave me a
better sense of the lab's overall work in remote sensing and helped me think
about the questions below.

My background is in image classification, which I see as a solid foundation
for
moving toward segmentation, since many of the core concepts (encoder
architectures, transfer learning, loss functions, evaluation under
different data
regimes) carry over directly. I am actively building on this by studying
segmentation methods and learning to work with geospatial data formats. I
have a couple of questions that would really help me write a focused
proposal:

1. Scope - cloud detection vs. new thermal tasks.
   The project description mentions building a general-purpose framework
that
   covers dataset creation, model training, and advanced analysis tools like
   uncertainty quantification and explainability, with application scenarios
   ranging from cloud detection to thermal event monitoring. Since
TIRAuxCloud
   already has a solid cloud detection pipeline in place, I would
appreciate some
   guidance on how you envision the GSoC work building on top of it. Should
the
   main focus be on generalizing and improving what already exists (for
example
   supporting additional sensors beyond Landsat TIRS and VIIRS, or adding
uncertainty
   estimation), or is there also an expectation to start supporting new
thermal analysis
   tasks during the project?

2. Is there a Discord server, or any other communication channel where
   contributors can discuss the project? I would love to stay in the loop
and
   engage with the community.

I could follow up with a detailed draft proposal once I have your input. In
the
meantime, if there is a starter task I could pick up or a paper you would
recommend to better understand the project's direction, I would really
appreciate
it.

Thank you for your time, looking forward to hearing from you.

Best regards,
Nikitas Rafail Karachalios
----
Λαμβάνετε αυτό το μήνυμα απο την λίστα: Λίστα αλληλογραφίας και συζητήσεων που απευθύνεται σε φοιτητές developers \& mentors έργων του Google Summer of Code - A discussion list for student developers and mentors of Google Summer of Code projects.,
https://lists.ellak.gr/gsoc-developers/listinfo.html
Μπορείτε να απεγγραφείτε από τη λίστα στέλνοντας κενό μήνυμα ηλ. ταχυδρομείου στη διεύθυνση <gsoc-developers+unsubscribe [ at ] ellak [ dot ] gr>.

πλοήγηση μηνυμάτων