Ad Astra

https://docs.google.com/presentation/d/1qvhTnufjPUgbll5kJmXF__cmwAhTOseaV5F_xkfO3m8/edit?usp=sharing

https://github.com/Isistickz/Voss

The basic idea behind VOSS is to take advantage of the fact that plastics reflect some certain wavelength of radiation absorbed by water. We also took advantage of the fact that Sentinel-2 sat actually measures these wavelengths. The data is locally normalized, which sort of allows us to avoid direct Atmospheric processing. The image is pre-trained for cloud masking as well as pixel classification to determine which substance is likey in which pixel. Then we apply the FDI and NDVI to the imagery to obtain both values for every pixel in the image. And based on the results of the FDI and NDVI, we can detect if a material is a plastic, a water body, or another sea material, which we thought would be another cool thing to look out for.

The long-term goal for VOSS is live tracking - to be able to detect in real-time if plastic is present or it isn't in a region of satellite imagery. Especially for the macroplastics, collection/retrieval is the most viable way to get them out of the ocean and VOSS would be a fantastic aid that provides precision to the process of retrieving the plastic waste.

Tools, coding languages

Python

EO-Learn, which also happens to be a python package.

Team AdAstra used the European Space Agency's Sentinel-2 data, available at https://scihub.copernicus.eu/

Initially, we were accessing it through the GUI on the website above, but once we switched to using eo-learn, accessed it via https://www.sentinel-hub.com/

Sentinel-2 is equipped with MSI which basically contains a bunch of bands measuring radiation at different wavelengths. With EO_learn, we could directly select the bands we wanted to work with, in particular, the NIR(Near-Infrared) and the SWIR(Short-Wave Infrared) band.

The journey was interesting and exhilarating overall. It didn't go as planned, but then, we learned a bunch of things.

We chose this challenge because it's important to work on problems that matter and problems that are impactful when solved. Coming from a developing country where basic human needs for a large swath of the population is still hard to come by, this is a population that has profound effects on us now, and much later in the future, will have more profound effects into the future on us as a species.

The thing we first did was comb through documentaries on YouTube to understand the scale and the scope of the problem, and then comb through the scientific literature on the topic of plastic debris in general as well as marine plastic debris, which in all honesty, contained a lot of strange and new vocabulary, but Wikipedia, a simple Google search and YouTube very much came to the rescue. The work of Dr. Lauren Biermann at Plymouth Marine Laboratory really influenced our approach to solving the problem. We then decided what the end -goal had to be, which was to be able to detect plastic debris in a satellite image, using a ML algorithm, and it was pretty much straightforward to walk back from there to figure out all the things that had to be in place for that to be achieved. Then we came up with a list of possible approaches as well as the sub-tasks which we then apportioned to each member of the team to work on.

None of us has worked on any remote sensing project before, but two guys on the team have done some ML projects. It quickly became apparent to us that the bulk of the work wasn't in building, training, or even testing the model, but in the stage of acquiring the data and processing it into a format that we can do ML on.

After initially downloading the data set we wanted to work with, there was the need to take care of the effects caused by the absorption and scattering of rays in the atmosphere(Atmospheric correction). None of the software we got worked. Going by our guesses and debugging trials, the fault was with our machines and we spent quite some time looking for a workaround. Then we discovered that we could use EO-Learn and all its packages to process and work with Sentinel data without the need for Atmospheric correction.

All in all, it was a very interesting experience for us all.

1. Biermann, Lauren & Clewley, Daniel & Martinez-Vicente, Victor & Topouzelis, Konstantinos. (2020). Finding Plastic Patches in Coastal Waters using Optical Satellite Data. Scientific Reports. 10. 5364. 10.1038/s41598-020-62298-z. 
2. Sentinel-2 data - scihub.copernicus.eu.
3. EO-Learn - https://eo-learn.readthedocs.io/en/latest

#marinedebris #plasticwaste #AI #ML

This project has been submitted for consideration during the Judging process.