Asteroid Zoo Talk

Automated detection

  • philmurray by philmurray

    After classifying for a while it has occurred to me (and surely others) that this process could be greatly speeded up by preselecting images that show up moving objects via software. It is not a difficult task to align and merge four frames with a view to removing all static objects and then score the result according to how much 'moving' data there is.

    I recognise this description is simplistic and requires a much deeper treatment but the underlying concept seems sound to me. If only the sky segments that achieve a threshold score are presented to us observers then we can decide if the software has identified a true asteroid or if its noise, artefact or whatever.

    Posted

  • scibuff by scibuff

    Software is already good at identifying the bright (obvious) object. The aim here is to "teach" the software (machine learning) to distinguish real objects from artifact / bg noise and find new ones below 2-3 sigma - something human eyes and brains (thankfully) are still much much better at than any software right now...

    Posted

  • philmurray by philmurray

    Ah, that makes sense. I had not clued in to the object of the exercise and what you say makes sense. Training neural networks or genetic algorithms can be a bitch .. 😉

    Posted