This can be got by applying a general purpose noise reduction program (here Neat Image) which can be guided by selecting a noise sample from otherwise flat area:
There's set zero color noise, maximum luminance noise, sharpen and the noise sample was taken from the dark area.
G'mic filter collection has detail splitting filters which try to separate different patterns to different layers. Here's one result. The layers which seemed to contain unwanted patterns are closed. This remained.
I guess this is quite the same as what's presented in another answer, but this is only a guess. I haven't modern image processing math knowledge.