Recently, the topic of on-orbit data processing has been gaining in popularity, as it offers not only up to a 100-fold reduction in the data size without loss of relevant information but also the possibility of real-time application, which is key in crisis management.  

However, to be efficiently used, firstly, the delivered hyperspectral data require fast processing to enable their analysis in as close to real-time as possible. Secondly, they necessitate compression, as their size significantly exceeds the downlink capabilities. 

Taking all that into account, the current biggest challenge of processing data in space is developing the technology that would enable their efficient processing and compression already on board of the satellite.  

What actually is data processing? 

On-board payload data processing includes the data acquisition, selection, compression or reduction (depending on the specific scenario), as well as their storage. 

Data acquisition and selection involves gathering pertinent information and analyzing it. Hyperspectral imaging* is typically transmitted by a satellite in the form of raw data, that is, sets of two-dimensional images generated for individual wavelength bands. In case of difficult weather conditions, such as heavy overcast, Earth observation is impossible, due to lack of visibility of the objects that interest us. Without the on-board image selection capabilities, an inconceivable amount of useless data would be transmitted to Earth – data for which we would have to wait for a very long time (owing to their size, for one). For that reason, it is imperative to use various processing and compression techniques in order to reduce the amount of data. In the simplest terms, data compression involves changing the method of recording information to minimize redundancy and, effectively, the volume of the data set. When it comes to data storing, currently, the biggest problem is posed by huge amounts of the generated raw data unable to reach Earth quickly due to their size. Consequently, they need to be stored for a long time on board of the satellite, which has limited storage capabilities. 

Example: 

For example, a single, unprocessed hyperspectral image takes up 2 GB, whereas the speed of data transfer to the ground station is 50 Mb/s. As a result, transmitting one image to Earth can take almost 7 min., while the length of a typical communication session is 5-10 min. 

The example above demonstrates how crucial the speed of data analysis is if you want to use the downlink throughput most efficiently. Processing data using traditional digital signal processors (DSPs) is very time-consuming due to their low computing power. The dynamically evolving DPS technologies show that one of the most effective solutions are currently field-programmable gate arrays (FPGAs), which, combined with artificial intelligence, work much more efficiently, resulting in faster data analysis than before. 

Artificial intelligence in orbit 

On-board payload data processing using artificial intelligence enables automatic selection of the gathered information and, consequently, downlink of the data with key importance for the user first. The volume of the data transmitted to the ground station will be significantly reduced, and the used algorithms will facilitate capturing relevant changes in the observed areas thanks to the capability for real-time analysis of different types of information. The information sent that way will not only shorten the reaction time to the observed phenomena, but it will also allow the end-user to acquire processed data ready for further analysis. 

Example: 

The goal is to reduce the size of the transmitted data to a bare minimum. For instance, roughly 7 GB are needed to record the results of imaging an area of 40 km x 40 km at the spatial resolution of 25 m/pixel for spectral measurements using 150 wavelength bands/channels. However, through processing, it is possible to achieve at least a 100-fold reduction of the amount of the data – for example, by using image segmentation algorithms, which allow us to generate a map that shows the position of specific objects at the scene. This will contribute to shortening the time between the moment of the occurrence of a specific phenomenon, such as fire or flood, in the observed area and the moment we receive information about it.  

On-board computer 

If delivering already processed data to the ground station is so crucial for fast acquisition of strategic information, why can’t we do it on board of the satellite using a dedicated on-board computer? 

To answer this question, the KP Labs engineers conducted long research, which resulted in the development of technology facilitating efficient data processing in space. Thanks to its advanced architecture, the data processing unit Leopard created for the mission Intuition-1 offers smooth performance in space. It has been equipped in field-programmable gate arrays (FPGAs), which, combined with artificial intelligence, enable gathering key information from hyperspectral imagery and transmitting the already processed data to Earth. The used technologies increase the throughput to even 3 billion operations per second, making real-time data compression and classification possible. 

If you are interested in the details of our solution or if you want to learn more about the mission we are planning, we will do our best to answer your questions. Contact us at BD@kplabs.pl

Leoopard

*multispectral and hyperspectral imaging – techniques of capturing image data in colour photography in full-colour space in the visible spectrum as well as in microwaves, near- and far-infrared, and ultraviolet. A multispectral image consists of multiple bands that can divide the 3 primary colour bands: R (red), G (green), and B (blue), into bands of any chosen spectral range. 

Sources: 

htts://www.youtube.com/watch?v=LNkl4WvCc8M&t=9s 

http://www.esa.int/Enabling_Support/Space_Engineering_Technology/Onboard_Data_Processing/What_is_On-board_Data_Processing 

https://www.researchgate.net/figure/A-current-typical-onboard-data-processing-system-design_fig3_309914377 

https://ui.adsabs.harvard.edu/abs/2013EPSC….8..355M/abstract