The interstellar object (ISO), known as 1I/’Oumuamua, which buzzed our planet on its way out of the Solar System, caught humanity’s attention for the first time in 2017. There are many theories as to what this object might be because it was obvious from the scant information that it was unlike anything that had ever been observed by astronomers. It was rumored that it might have been an alien probe (or a fragment of a crashed spacecraft) travelling through our solar system. With the ODNI’s publishing of the UFO Report in 2021, public interest in the prospect of “alien visitors” was further stoked.
This action essentially turned the investigation of Unidentified Aerial Phenomena (UAP) into a scientific endeavor rather than a covert operation under the control of government organizations. Scientists are putting forth ideas about how to employ current developments in computers, AI, and sensors to help in the detection of potential “visitors” while keeping one eye on the skies and the other on orbiting objects. This includes a current investigation into the creation of an advanced data pipeline using hyperspectral imaging and machine learning by a team from the University of Strathclyde.
Researchers from the departments of mechanical and aerospace engineering, electronic and electrical engineering, and the Fraunhofer Centre for Applied Photonics in Glasgow made up the team, which was led by Massimiliano Vasile, a professor in those fields. Their study, “Space Object Identification and Classification from Hyperspectral Material Analysis,” has made an online appearance as a preprint, and it is currently being reviewed for publication in Nature Scientific Reports.
This paper is the most recent in a series that discusses uses of hyperspectral imaging for space operations. The Hyperspectral Imager for Space Surveillance and Tracking (HyperSST) project’s first publication, “Intelligent characterization of space objects with hyperspectral imaging,” was published in Acta Astronautica in February 2023. This project is a forerunner to the ESA’s Hyperspectral space debris Classification (HyperClass) initiative and was chosen by the UK Space Agency (UKSA) last year as one of thirteen debris reduction proposals that would get funding.
In their most recent publication, they looked at how this imaging method might be used to the expanding field of UAP identification. This method entails gathering and analyzing information from single pixels across the electromagnetic spectrum, usually to distinguish between various objects or materials captured in photographs. By removing false positives brought on by man-made debris items (spent stages, defunct satellites, etc.), hyperspectral photography combined with machine learning has the ability to focus the search for probable technosignatures:
If UAP are space objects, then studying the spectra will allow us to determine the material composition of each every pixel. By examining the time variation of the spectra, we may also comprehend the motion of the attitude. Both are crucial because they allow us to recognize objects by their spectral signatures and comprehend their motion with only a few simple optical needs.
Vasile and his associates suggest developing a data processing pipeline for UAP picture processing with machine learning methods. They first described how a collection of time-series spectra of space objects, such as satellites and other objects in orbit, is required for the workflow. This includes debris objects, which calls for merging information from other national and international organizations as well as the Orbital Debris Program Office (ODPO) of NASA, the Space Debris Office of the ESA, and others. This dataset needs to be diversified and contain information on various orbital possibilities, trajectories, lighting situations, and accurate geometry, material distribution, and attitude motion information for all orbiting objects at all times.
In order to reduce false positives, scientists would require a substantial database of all human-made objects in space. Vasile and his team developed numerical physics simulation software to generate training data for the machine learning models because a lot of this data isn’t readily available. The next step involved using a two-pronged technique, one based on machine learning and the other on a more conventional mathematical regression analysis used to find the line of best fit for a set of data (also known as the least square method), to link a spectrum to a set of materials that generated it.
The likelihood of discovering a combination of materials was then linked to a certain class using a machine learning-based classification method. Vasile explained that after the pipeline was finished, testing were performed, and the results were positive:
“We conducted three tests, one in a lab using a model satellite constructed of well-known components. These tests came out very favorable. Then, in order to replicate actual orbital object observation, we developed a high-fidelity simulator. The tests were successful, and we learned a lot. Finally, we made use of a telescope to view the space station and several satellites. Because of the current size of our material database, certain tests in this example were successful while others were less successful.
Vasile and his coworkers will outline the attitude reconstruction component of their pipeline in their upcoming publication. They intend to exhibit this pipeline at the upcoming AIAA Science and Technology Forum and Exposition (2024 SciTech), which will take place from January 8–12 in Orlando, Florida.
Further Reading: arXiv