This project will pioneer novel deep-learning architectures for spectral reconstruction from thermographic hyperspectral interferometric data and apply it to detect gas-phase ammonia. Specifically, we will develop and train suitable deep-learning architectures to bypass the need for numerical matrix inversion, which would otherwise be needed to reconstruct the thermal infrared spectra, which has hitherto been insufficient to reconstruct the spectrum of ammonia.

Furthermore, we will use the developed AI models to test the detection of gas-phase ammonia from simulated environments, aimed at, ultimately, direct in-field measurements. The detection of gas-phase ammonia with AI-driven thermographic hyperspectral imaging would be paradigm-shifting in the field of ammonia emissions measurement, particularly as the project migrates to conducting measurements outside of the laboratory setting, which ensures research quality at the highest
international level.

Additionally, the developed AI architectures for the spectral reconstruction of thermographic hyperspectral interferometric data would be innovative not only to the field of
ammonia sensing but broadly impact in any field working with this type of hyperspectral data, including forensics, building inspections, defence and more, by enabling spectral reconstructions superior to those obtained by conventional methods. Thus, offering new insights into thermographic hyperspectral payload integration with UAVs and gas-phase ammonia detection, the project has both research quality at an international level and significant innovation height.

The purpose of the project is three-fold:

1. Develop an AI architecture correlating thermographic hyperspectral interferometric measurements (intensity vs. mirror separation) to the reconstructed
spectrum (intensity vs. wavelength/wavenumber) as conventional Discrete Fourier Transforms is not compatible with the novel hyperspectral camera technology.

2. Measure gas-phase ammonia in a laboratory environment that emulates aspects of in-field ammonia emissions.

3. Collect thermographic hyperspectral imagery in-field using the developed computational frameworks.

Participating organisations will gain forefront knowledge on future technologies for ammonia emissions quantification and monitoring, opportunities to scope the technology towards application in agricultural settings, and gain new competencies in handling, deploying and interpreting thermographic hyperspectral cameras and their output data.

The successful project will be a driver towards creating more tech- and data-driven workplaces and will support the broader society by tackling an underestimated source of anthropogenic pollution in the 21st century.

Results from the experiments and the developed AI Architecture has a high level of novelty, and thus it is expected to be presented in a scientific publication(s). The collaboration between SDU, Qtechnology and Agro-alliancen also creates a network of competencies with the potential to introduce future cutting-edge solutions for monitoring ammonia gas.

By the end of this project, it is very likely that the partners will apply for a larger research project within the Green Theme at Danmarks Frie Forskningsfond or Innovationsfonden Grand Solutions.

Furthermore, the potential business case in providing the monitoring of agricultural fields in the future, creates an opportunity for a spinout or startup by the end of the project

The Ministry of Higher Education and Science has financed this project with 552.960 DKK.

The project runs from August 2023 to August 2024.