The present document constitutes the Final Report of the DeepLIM project. The Final Report is meant to represent a self-contained document that explains, in medium-level details, all the activities and results of the projects. DeepLIM (Deep Learning for Inversion Models) is an ESA funded activity that investigates the use of Generative Artificial Intelligence to help in the generation and processing of data relative to remote sensing applications. The two key applications carried out in the activity, as agreed at proposal stage, are: 1. Thermal Infra-Red (TIR) use case. For this use case it is possible to identify the following areas of research and development: a. Synthetic data generation: in this field, the activity explored the generation of synthetic data related to DART-like signatures, TASI-like signatures, conversion from TASI to DART, and 2D scene generation. b. Inversion modelling: in this field, one Deep Learning -based inversion model has been developed and compared with two state of the art retrieval methods (GSW and TES). 2. Vertical Total Electron Content (vTEC) maps. For this use case, the following areas of research and development were explored: a. Single vTEC map generation: the capability of generating a single vTEC map in a given point in time. b. Sequential vTEC map generation: the capability of generating a sequence of vTEC maps starting from a given point in time. The activity has two goals: 1) Goal 1: to develop generative models that can augment and extend the capabilities of generating new datasets for the two use cases 2) Goal 2: to explore the capabilities of Deep Learning technology for performing inversion modelling on the Thermal Infra-Red use case. The activity has been carried out over a period of 18 months, and consisted in the research, design and development of the following Deep Learning models: o GANDART - Deep Learning model to generate synthetic DART data o TASI2DART - Deep Learning model to generate DART signatures from TASI campaign data o DLIM - Deep Learning based inversion model for TIR data retrieval o vTEC Generator - Deep Learning model to generate vTEC maps The consortium is composed by AIKO, prime contractor, the National Research Council of Italy - Institute of Atmospheric Sciences and Climate, and the Barcelona Supercomputing Centre. The Abdus Salam International Centre for Theoretical Physics provided support for Use Case 2. ESA contract reference: 4000128370/19/NL/AS.
Final Report
Elisa Castelli;Enzo Papandrea;Alessio Di Roma
2021
Abstract
The present document constitutes the Final Report of the DeepLIM project. The Final Report is meant to represent a self-contained document that explains, in medium-level details, all the activities and results of the projects. DeepLIM (Deep Learning for Inversion Models) is an ESA funded activity that investigates the use of Generative Artificial Intelligence to help in the generation and processing of data relative to remote sensing applications. The two key applications carried out in the activity, as agreed at proposal stage, are: 1. Thermal Infra-Red (TIR) use case. For this use case it is possible to identify the following areas of research and development: a. Synthetic data generation: in this field, the activity explored the generation of synthetic data related to DART-like signatures, TASI-like signatures, conversion from TASI to DART, and 2D scene generation. b. Inversion modelling: in this field, one Deep Learning -based inversion model has been developed and compared with two state of the art retrieval methods (GSW and TES). 2. Vertical Total Electron Content (vTEC) maps. For this use case, the following areas of research and development were explored: a. Single vTEC map generation: the capability of generating a single vTEC map in a given point in time. b. Sequential vTEC map generation: the capability of generating a sequence of vTEC maps starting from a given point in time. The activity has two goals: 1) Goal 1: to develop generative models that can augment and extend the capabilities of generating new datasets for the two use cases 2) Goal 2: to explore the capabilities of Deep Learning technology for performing inversion modelling on the Thermal Infra-Red use case. The activity has been carried out over a period of 18 months, and consisted in the research, design and development of the following Deep Learning models: o GANDART - Deep Learning model to generate synthetic DART data o TASI2DART - Deep Learning model to generate DART signatures from TASI campaign data o DLIM - Deep Learning based inversion model for TIR data retrieval o vTEC Generator - Deep Learning model to generate vTEC maps The consortium is composed by AIKO, prime contractor, the National Research Council of Italy - Institute of Atmospheric Sciences and Climate, and the Barcelona Supercomputing Centre. The Abdus Salam International Centre for Theoretical Physics provided support for Use Case 2. ESA contract reference: 4000128370/19/NL/AS.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.