Carvajal, Juan (2024) Model learning for parameter cartography applied to hyperspectral astrophysical images PFE - Project Graduation, ENSTA.
![]()
| PDF 4Mb |
Abstract
Integration of a model for parameter mapping within an astrophysical context of hyperspectral imaging. In this work, a surrogate spectral model based on autoencoder is trained to allow both spectral reconstruction and mapping prediction of the physical parameters associated with the spectra. Using as dataset, simulations inspired by supernova remnants, a benchmark is performed between two network architectures for the prediction model. The best model is integrated together with a non-stationary unmixing algorithm to retrieve both the data cubes corresponding to the physical components and the parameter maps from the hyperspectral images. We demonstrate that a model can be trained that allows the construction of the parameter maps without suffering a significant loss in the quality of the physical component separation.
Item Type: | Thesis (PFE - Project Graduation) |
---|---|
Uncontrolled Keywords: | Hyperspectral images; Machine learning; Source separation; Interpolating autoencoder; Semi-blind unmixing |
Subjects: | Physics, Optics |
ID Code: | 10384 |
Deposited By: | Juan CARVAJAL |
Deposited On: | 04 oct. 2024 17:39 |
Dernière modification: | 04 oct. 2024 17:39 |
Repository Staff Only: item control page