42 is the answer home page
Our approach
1.-Processing chain in Sentinel-2
Medical imaging with no patient risk can be applied to this discipline. In fact, there are complete numerical runtimes designed for space application based in medical imaging.
The most interesting part of this kind of applications is that this is always a big data problem, that must be treated as a large scale problem to provide successful results. This is processed in a chain where products (computing refining steps) are provided.
One of the problems when delivering remote sensing products is the time needed to prepare the L1, L2 products since the time they are received from the satellite. The complete operation since the acquisition of the image implies waiting for the satellite to arrive to a discharge station after one or several passing orbits, the reception and transmission of high resolution hyperspectral images, sending the raw data to the ground segment for processing and wait until the processing chain completes L1 and L2 products are delivered to the public.
There are still demanding applications such as precision agriculture, that requires as fast as possible results from very well located areas and specific surveilance that requires faster in time that wait to the full processing chain that is intended in normal satellite operation.
For this purpose our group is currently develping on-board processing applications. This will provide, faster thant standard chain processing results. This applications are focused to hyperspectral images.
On-board processing
Tipically, on-board processing is performed on a satellite platform using small electronics that consume the least power possible, whereas at the same time provide a limited computing capabilities, but enough to perform a series of complex but fast algorithms, ready to be used almost instantly to third party applications.
There are a large number of corrections that must be applied to an hyperspectral imaging (geometric correction, atmospheric correction, calibration). For large swath sensors like CHIME the visible surface is large enough to provide a wide range area where some ground applications, such as crop monitoring does not require precision but confirmation about these events.
Many of this information is contained at TOA level (Top of Atmosphere signal), and includes based on detection of special absortions bands.
- Enhanced cloud, sea detection
- Soil moisture
- Chlorophyll absorption
- Chemicals and aerosol industrial production
- Active volcanoes emissions
This research area includes a large number of technical disciplines availables in our research center but not integrated in a standard aplication.
Modeling and TOA generation
Fast modeling by realistic models, including one-year-period, of several scenarios. Currently we have delivery the Scene Generator Module (SGM), for CHIME mission and simulation over one year in complex scenarios (hihg frequency Digital Elevatio Models (DEM), vegetaton, clouds, snow, and human activities), has been achieved thanks to Artemisa.
The generation of the synthetic scenes using our High Performance SGM receive input from several sources and diverse scientific domains, e.g. land cover map databases, atmospheric radiative transfer models in-LUT stored radiances, reflectance maps, vegetation simulation using a model or maps, digital elevation models. The different inputs needed to realistically simulate the natural variability in a synthetic scene are heterogeneous in nature, size and format, and the SGM handles each of these inputs in an optimal way.
Atmosphere and aerosol generation, mainly used for coupling surface and atmosphere for TOA generation, but generally we can use diferent atmosphere models without increasing the computational cost due to the LUT, format used during generation.
Image modelling expertise, this is the main focus of our research, as we colaborate with the IPL, in this projects we have access to the state of the art about atmosphere modeling and databases in different scenarios, clouds, carbon, volcano activities, and most important, vegetation processing.
Our processing chain has improved in near a factor 100 the computing time expectations, this provides a great opportunity to generate large scale models train and select diferent applications.
Data augmentation is always a problem for confident Machine Learning algorithms, but with the high performance and high thoughput to generate realistic TOA images, we can perform an exhaustive study surpassing the state of the art of on board processing algorithms that provide results in the limits of the precision.
Another approach of this processing model is the ability of provide changes of the firmware based on the satellite location, we can triggen an specific model to provide sea applications, such as the orbits around the oceans (Pacific, Atlantic) where ground detection does not provide extra information, and change the firmware whenever the satelite goes by continental orbits.
In special devices implementation
As our group comes from High Energy Physics, we are also interested in the development of the state of the art of hardware comming from our experiments.
Currently we hold projects where speed communication, or trigger information of a large scale detectors such as ATLAS or KM3Net, requires decisions to store or not the data generated.
This expertise helps us in the current areas:
- Algorithm capabilities (based on the technology)
- Trigger and electronic feedback (based current projects)
- Near real time application implementations