<aside>
This page will present the planned technical project work among PAY Team subsystems for Summer 2026. Project description & motivation, expected deliverables and timelines are provided.
</aside>
UTAT Space Systems, Payload Team (UTAT SS PAY, PAY) is making a strategic pivot to a nanosatellite mission concept that is primarily focused on student education with a compelling scientific value proposition. Given our resource limitations and tentative faculty support, this pivot will increase our chance of successful project delivery while maximizing valuable learning outcome for everyone on the team.
The goal is no longer to build a perfect cutting edge instrument — instead, we will aim to build a student-doable satellite payload, and crucially aim for perfect execution of our design. The most valuable and tangible learning opportunity will come not from designing the best-in-class hyperspectral imager — we can’t, we aren’t NASA / ESA. It would be to design a device, fabricate it, assemble it, test it, and iterate until the device works as intended. Delivering a product trumps attempting to design a perfect product.
Towards this end, we have envisioned a series of projects in which we will build stepping-stone prototypes while concurrently designing the eventual satellite payload. These prototypes are chosen strategically to directly confront the hardest unsolved problems in the payload design, while also producing useful functioning devices that will benefit the team down the road. These projects will be organized in a way that can be completed in a couple semesters of part-time work, and structured in a manner similar to undergraduate engineering design courses.
We present our list of high-priority payload projects that will define direction for the team for the following several semesters.
SWIR Spectrometer Project: since we are building a hyperspectral imager, it is indispensable to have a spectrometer against which we can check the spectral accuracy of our imager. While we can’t budget for a commercial SWIR spectrometers, we actually have most of the components (worth ~90% of the total cost) that it takes to build it. In doing so, we can build our optical and opto-mechanical knowledge that would directly transfer to our hyperspectral imaging payload design, while also giving Science a new instrument with which they can validate their measurements. Th Data Processing team will write a GUI to display the spectrum from the camera image, and test their keystone & smile correction algorithm approach on real hardware.
Build 2 Hyperspectral Line Scanner: An ongoing project from previous semesters, Build 2 is a visible-range “demo” version of our payload using all-commercial optics and custom-made optomechanics. Optics will finish aligning and adjusting the setup while Data Processing will write a GUI that lets us collect hyperspectral datacubes using the hardware we built. In doing so, we will develop an in-house tool for processing and visualizing hyperspectral datacubes, which we can use to collect more experimental data to improve our scientific models.
FLIR TAU Readout Board: we will finish the efforts currently underway to build a custom PCB that has a microcontroller chip that controls the FLIR TAU SWIR sensor (FPA temperature, integration time, gain) and collects frames via DCMI signal onto an SD card. The microcontroller will run on a finite-state-machine logic and will interface with peripherals using firmware code that we write. Using a microcontroller for camera readout is challenging given the anticipated ~100MB/s data rate, yet is more economical than a full-blown FPGA solution given the power and volume constraints on a 3U CubeSat.
Literature Review of Hyperspectral Imaging CubeSats: as we prepare to write the CUBICS grant, we need to have a strong understanding of our scientific value proposition that makes our satellite project uniquely worthwhile for the CSA to fund. There is a tremendous scope of exciting and impactful scientific applications that are addressable with hyperspectral imaging. While CubeSats provide an accessible low-cost platform for space science, they impose unique constraints within which we need to find a feasible scientific niche. Having a well-substantiated scientific mission provides the central purpose to our work at UTAT Space Systems.
Mini STOP Analysis: Satellites experience substantial temperature fluctuations while in orbit (+40C to -70C), as they are either baked by the sun or freeze in the Earths shadow. Optical systems require precise alignment, and such large temperature swings can significantly affect optical performance. The optomechanical housing must therefore be designed to provide the thermoelastic rigidity to keep optics “in focus”. Performing such a Structural, Thermal, Optical Performance (STOP) analysis on a real optical and optomechanical system is a complex task which requires specialized knowledge and software. In order to gain familiarity, we will develop simplified optical systems with very basic housing geometries and analyze the rough-order-of-magnitude deformations and performance degradations using the tools we have available.
Some subsystems may have self-contained research and onboarding projects which do not extensively interface with other subsystems. We briefly describe those in the following list:
Optics — Monolithic Catadioptric Lens : monolithic catadioptric lenses - made from a solid block of glass with reflective surfaces machined on either side - can reflect light several times internally and “fold” the optical path to achieve a long focal length inside a small volume. Since commercially available variants are virtually nonexistent, we are developing an optical model of this lens that is suitable for use with FINCH EYE. While we do not anticipate including this lens on the satellite, we recognize the potential for contribution to optical engineering research literature and seek to develop the concept into an adequate conference publication.
Science — Hyperspectral Foundation Model (HFM): Foundational models at their core are developed to embed spectra and their conditions into a highly structured, high-dimensional latent space with some intrinsic well-defined manifold. There is currently undergoing research on Geospatial Foundation Models (GFM), but such models admittedly are much more limited in terms of different fields it can be applied to, given that they assume some sort of data cube. With a sufficiently well trained HFM, one could append different decoder network after the latent space to produce several different models that may be capable of: hyperspectral unmixing, data generation, hyperspectral data similarity metric (similar to FID in CV), and atmospheric effect inversion. HFMs in certain context become distinctly superior to GFMs in some cases given that they do not need full data cubes as assumed, but they suffer from less information content, making it harder to train them. This will be the flagship project of the DL sub-team of Science.
Science — Synthetic Data Generation: The synthetic data generation models developed for the ISPRS paper show some potential in being useful, as explained in the paper (the rationale for a decoupling framework). However, the models have been catastrophically bad to be exact. I (Ege) think it is rational to pursue these models further and properly incorporate the physical priors, given their possible significant benefits into making the proposed decoupling framework fully feasible. Additionally, the research of these models with the HFMs are “highly synergized” given that HFMs could end up providing a very crucial similarity metric. This will be the main onboarding/new member project of the DL sub-team of Science.
Science — Advanced Hyperspectral Unmixing Models: Hyperspectral Unmixing Models are obviously the main driving force of Science, given that they are one of the main reasons as to why Science was even formed. The current unmixing models are highly capable in performing well with more well-defined datasets that have well defined endmember classes of not high quantity. However, the primary question is that how can these models be expanded into unmixing scenarios of much higher numbers of endmember classes (a more complex supervised task), or maybe when endmember classes are not defined at all (completely unsupervised task). It is incredibly valuable that we continue in this direction to either develop unmixing models capable of learning intrinsic manifold of more complex mixtures or learning incredibly ill defined but worthwhile to solve.
Science — Generalized and “Fully” Automated Atmospheric Modelling Pipeline: The currently developed atmospheric modelling pipelines are only semi-automated. Although the major parts are mostly automated, we still must have a some tasks that we end up doing manually and are scattered in different codes. The main task will be to build a unified point of entry and exit, that is interactable via config files. Additional to this, we MUST generalize our pre-existing pipeline to more complex modelling scenarios, such as including turbid (or even non-turbid) water, roads, pavements, houses, and cars in the scenarios. Such a task is incredibly hard, and is essentially a massive upgrade to the existing DART program, but will have immense respect from the widespread atmospheric modelling community in its attractiveness.
DataProcessing — DarkNoise: Researching metadata-conditioned dark-frame reconstruction and denoising methods for hyperspectral satellite imagery when on-orbit calibration frames are unavailable. The project focuses on modelling temperature-dependent dark current, fixed-pattern noise, and sensor artifacts using physics-informed machine learning integrated with hyperspectral restoration networks. The goal is to improve radiometric fidelity and downstream analysis quality for small satellites like FINCH, while enabling robust post-processing under constrained calibration conditions and varying orbital thermal environments.
DataProcessing — SuperResolution : Developing super-resolution techniques for hyperspectral Earth observation imagery to improve spatial detail beyond native sensor resolution. The research investigates spectral-spatial deep learning models, multi-frame scan fusion, and physically informed reconstruction pipelines to enhance agricultural mapping and subpixel feature separation. A major objective is enabling more accurate crop residue detection and unmixing from nanosatellite imagery by increasing effective ground resolution while preserving spectral integrity and minimizing reconstruction artifacts.
Detailed project descriptions will be made available by Saturday May 16th and will be discussed at subsystem meetings.
Thank you for continuing with UTAT SS PAY this semester, and we look forward to the learning and engineering work up ahead!
DATA PROCESSING: Summer 2026 Projects Description