I am working with a small team of amateur radio astronomers in the UK (Radio Astronomy – British Astronomical Association) to set up a citizen-science scheme to record meteors. This is achieved using radio reflections off the plasma tails left by meteors entering the Earth’s atmosphere. Currently analysing these reflections is pretty manual and we would like to automate some of this, possibly using ML. There are two problems we would like to overcome. 1) Counting actual meteor hits 2) Rejecting aircraft echos.
The latter is one of the main challenges and is currently been done manually, as well as on a zoo basis.
I am sure, looking at the work others (and myself) have done with ML in Tensaflow that certainly the aircraft echos could be identified, and hopefully the actual meteors too. Here is a link to some of the data we are collecting with what we are looking for. Zooniverse
My questions is: 1) Do you think this would be possible? 2) are there any of you who may be interested in helping out on the project?
Nice initiative!
- I think it would certainly be possible to have a classification task, given that you already have quite a lot of data. To reject the aircraft echos, I was thinking about anomaly detection using auto-encoders, again given the fact that you probably have a lot of data on actual meteor reflections.
Either way, I think there are several ways to tackle this project
1 Like