Technology

Synthetic-aperture radar satellite imagery

While the first SAR images were already taken in the 1960s, there has been a strong acceleration in the launch of new satellites over the last decade. This has allowed companies like ICEYE to build impressive constellations which are continously imaging the Earth's surface. The active mode of sensing allows observation during night-time or cloudy circumstances, the single most acute limitation of optical satellite imagery, by using radio frequency wavelength to penetrate clouds.

SAR imagery is highly useful for detecting man-made objects contrasted with vegetation, ground or seawater, detecting for example ships, weapon systems, bunkers, aircrafts and more. Also oil spills are particularly visible as well as its uses in change detection such as deforestation, illegal settlements or activity. Additionally, using multiple SAR images it is possible to accurately (in the order of millimeters) measure ground level changes and subsidence useful for predicting mining or dam failures and glacier degradation.

Hyperspectral satellite imagery

The recent advent of multispectral and now hyperspectral imaging has created the possibility of a hitherto unmatched resolution both spatially and spectrally. While multispectral payloads typically measure in the order of 10s of wavelength bands, hyperspectral payloads can measure up to 450 bands over a wide range (400 - 2550 nm) of wavelengths such as the Tanager constellation of Planet.

This allows the exact spectral profiling of objects of interest and thus accurate detections over large areas. Examples include the detection of sargassum seaweed and contrast it with other floating species.

Radio frequency triangulation from HawkEye360

Accessing source of radio frequency, especially in remote areas, allows the detection and location of for example GSM and VHF signals by triangulation while sweeping over large areas at a time. This permits the detection of even small boat and speedboats at open sea which might be involved in illicit behaviour, as well as illegal logging or suspicious activity near critical infrastructure.

The constellation of HawkEye360 is unparalleled when it comes to RF scanning from space. Their wide frequency band between 60 MHz and 18GHz allows for scanning between the VHF (maritime radio) and Ku (military) bands. HawkEye360 operates over 30 satellites in clusters of three to triangulate the exact position of an emitted signal.

Automatic identification system with global converage

Ships are equipped with a messaging system avoid collisions in busy straits or near ports by transmitting their position and other relevant data to other vessels nearby. While not designed to be picked up from space, the constellations of ORBCOMM and Kpler allow real-time and global tracking of all emitted AIS signals. This includes identification numbers like MMSI and IMO but also their position, speed, course, heading, cargo, dimensions, ETA and more, which is complemented with ownership and historic data from shipping registers to obtain all relevant information.

Pattern analysis of historic ship trajectories, including entrance into sanctioned zones, as well as alerting about real-time anomalous behaviour such as spoofing or intermittent signals is a essential source of information in locating and tracking dark vessels. Joined with satellite imagery and RF triangulation, suspicious AIS data can be cross-referenced and verified conclusively.

Artificial intelligence computer vision models

By employing artificial intelligence models in the field of computer vision and object segmentation, we enable the automatic detection of objects of interest, such as vessels at sea or oil spills, with high accuracy and at the scale required for satellite imagery. We use specialized high-performance computing clusters with state-of-the-art machine learning (AI) models that allows our solution to provide rapid and trusted reports.

Our vast database of satellite imagery permits the accurate and continuous training of our artificial models. We use modern GPU clusters to train the models at scale and instant inference results. For each of our products we develop a specialized deep learning model based on research and on the specific requirements. This has a clear advantage over standard statistical models such as CFAR since it is able to include the context and subtleties when classifying segments of the data.

High-performance computing with custom pipeline

The size of satellite imagery data can exceed 10 GB per scene which puts severe contraints on traditional or public solutions to process those images. We have built a tailor-made solution for geospatial processing with highly optimised image processing and analysis, reducing processing time from hours to minutes. This permits us to deliver timely results in time-critical missions while avoiding legal issues related to licence and cyber security risks.

Using machine code (ASM) and parallel processing, up to 64x speeds can be achieved in high-concurrency environments. Additionally, our in-house experience in cyber security enables us to adhere to the highest security standards and the zero-trust model to limit and disable unauthorised intrusions. This includes the use of the SELinux permission management developed by the NSA, OWASP security standards and recommendations, automated vulnerability testing from open databases and general strong focus on building robust, performance and secure code.