Automatically identifying satellite trails

Hi!

I'm participating in a project that calls for uploading entire field images to a project server where the photometry is done. My images are usually stacks of very short exposures and occasionally I catch satellites that have so bright trails that they show up in the stacked image and potentially confuse the photometry.

So I would like to identify the frames with satellite tracks in them and exclude them from stacking, and since we are talking 100s of frames, an automatic procedure would be nice. 

So I wonder if anyone is using an off-the-shelf solution that is superior to a very simplistic procedure I've come up with:

I'm experimenting with running my raw frames thru ImageMagick's [1] "-hough-lines" [2] filter to identify straight white lines against a black background, after increasing the contrast. So under Linux or in cygwin [3] on Windows, it comes down to a very simple command line like

ls -1 directory/*.fits | xargs -Ixxx -P4 convert xxx -white-threshold 0.25% -hough-lines 4x4+300 xxx.mvg

which will run (with 4 jobs in parallel to use multicore CPUs) over all *.fits files in a given directory and will create for each frame a text-file with a vector-graphics representation of any identified bright track (you might need to adjust the parameters for the white-threshold and minimum line length for your images)

For frames without an identified track, the output will look like this

$ cat V1491Cyg_00242_19_23_50Z_V.fits.mvg
# Hough line transform: 4x4+300
viewbox 0 0 3096 2080

However for a frame with a detected sat trail, it will have one or more lines beginning with "line" like this:

# Hough line transform: 4x4+300
viewbox 0 0 3096 2080
line 3182.26,0 2425.2,2080 # 369
line 3190.77,0 2433.71,2080 # 442
line 3195.03,0 2437.97,2080 # 443
line 3181.59,0 2465.39,2080 # 1486
line 3162.39,0 2486.56,2080 # 1212
line 3149.02,0 2513.1,2080 # 381
line 3156.34,0 2520.42,2080 # 419
line 3157.39,0 2521.47,2080 # 419

So after executing the command line above, you can identify frames with satellite tracks like this :

egrep "^line" directory/*.mvg | cut -f1 -d\: | sort | uniq | xargs -I xxx basename xxx .mvg

V1491Cyg_00049_18_32_21Z_V.fits

OK, so let's check whether  that frame indeed was the culprit:

Ok, that worked. Now this was a fairly obvious trail and this method needs more testing and tweeking (e.g. maybe use a sigmoid-contrast filter up front?). And it didn't even involve artificial intelligence and training a neural network ;-)

But I guess there are alternatives that people out there use.

The command line above took about 0.24sec per frame (3000x2000 pixel frame, so 12MB each), run on an SSD and a CPU with 4 cores. But runtime will be slower for very crowded fields. 

Any feedback is welcome...this might get more relevant in the future with those mega-constellations like SpaceX's Starlink etc quickly populating the night sky with more moving light pollution.

Clear Skies

HBE

[1] https://imagemagick.org/index.php

[2] https://imagemagick.org/script/command-line-options.php#hough-lines

[3] https://www.cygwin.com/