It's not a dot, in a long exposure photograph, it's a line. You don't delete it spatially, you'll want to subtract off the extra photon counts along that line so you can retain whatever was in that part of the image. Now, I think that's possible - find lines one diffraction-spot wide, fit to its brightness and subtract it off. They should be damned near straight lines of constant brightness given how fast satellites move, and the transverse profile should be exactly given by your telescope's point-spread-function, so you should be able to locate them and model them for subtraction very well. I think it can be done.

Photon shot noise is sqrt(N) though, so even after the subtraction you're left with more noise along that line than elsewhere. So that's not ideal. But increased noise is less of a concern that "totally ruined data".

Another caveat is that the lines have to not be so bright as to saturate your CCD, otherwise you truly have lost data that subtraction won't help with.

Edit: it turns out astronomers take many shorter exposures [1] and simply reject the pixels that are too far away from the mean in any given exposure when combining them. This seems mostly better than what I was proposing, though it does mean increased readout noise since you read out multiple times and so for very sensitive measurements I would still expect astronomers might want to do true single long exposures.

[1] https://www.skyandtelescope.com/astronomy-blogs/imaging-foun...

You can't just subtract a bright signal and hope to see the weak signal underneath it, or we'd just subtract daylight during the day.

Depends how bright, I do it every day in a different context and it can work pretty well. The brighter, the more of that noise I was getting at. Sunlight is very, very bright , but satellites less so. I'd need to know how bright to know if the resulting noise would swamp the signal. Perhaps the fact that astronomers aren't doing this is evidence that it does.