Hyperspectral Imaging (HSI) is powerful technology that facilitates comprehensive analysis of chemical and biological properties of Earth surface on a large scale, with proven applications in environmental monitoring, farming, forestry, mining and water quality management. In the context of industrial agriculture, HSI allows for data-driven optimisation of cultivation practices, giving dramatic reduction in use of water, fertilisers and crop protection chemicals, and resulting in significant improvements in the quality and quantity of the agricultural produce.
Considering the increasing urgency in these domains of human activity, the importance of HSI technology for food security, sustainable agriculture and forestry, marine and maritime and inland water research, as well as climate action, environment, resource efficiency and raw materials is evident.
However, HSI technology still struggles with the vast amounts of data generated: Processing of hyperspectral images is too demanding to be performed on-device, and current storage and transmission solutions are strained beyond their limits. Compression can solve this problem, but no algorithm has yet been adopted by industry, hinting at an unfavourable cost-benefit ratio or a lack of applicability. In this project, Dotphoton, specialised on image processing and compression using methods from quantum information science, and Gamaya, who has extensive expertise in hyperspectral imaging, team up to apply novel, information-preserving compression to the data challenge of HSI.
The breakthrough of information-preserving compression comes from a detailed modelling of the entire image acquisition pipeline, from the physics of light emission and scattering via elaborate image sensor characterisation to image file generation. For RGB images, a 10:1 compression has been achieved. We can prove that information loss is negligible, and images remain perfectly suitable for all applications.
The goal of this project is to push this compression technology into the realm of HSI. We will do so in three steps. First, the technology needs to be adapted to the particularities of hyperspectral sensors. Next, we validate the implementation by applying highly sophisticated image processing algorithms based on machine learning on the decompressed images. The aim is to achieve competitive performance on these images compared to non-compressed images. The final step of the project lays the foundation for broad adoption by investigating how the technology can be complemented and further optimised to match the needs of industry. Factors like sensor parameters, hardware implementation and analysis overhead will be studied to identify impact-maximising next steps after conclusion of the project.
The success of this project will significantly reduce complexity for all technologies using HSI, and effectively catapult them 6 years into the future in terms of bandwidth and storage costs.