Here is my attempt. Processing:
- split to R, G, B
- ABE, function degree 1, subtraction individually for each channel
- use image statistics, activate stdev and check how much noise there is in the three channels. Question is use green only or blue and green. It turns out your blue is not much worse than green.
- Linear Fit blue to green
- Pixel Math: O3 = 0.5xG + 0.5xB
- rename R to Ha
- I skipped deconvolution. The data is not deep enough.
- for both individually clone and make a mask that separates stars from background, invert mask, use wavelets for linear denoise.
- stretch Ha: 300 in arcsinh followed by a gentle HT
- stretch O3: 500 in arcsinh, follewed by a gentle HT, O3 is much dimmer, don't overstretch
- for O3: scripts-analysis-extract wavelet layer, 4. Keep the residual, close the layers. Stretch residual to a mask for individual boost of the nebula only, remove stars from the mask with clone stamp
- LHE on O3 to taste using that mask (I used the default settings) plus HT
- make the same residual mask for Ha but do not yet boost Ha, it is strong enough
- use PixelMath for combination so that the stars are neutral or ideally a few blueish and a few redish
I used R=0.85Ha, G=0.75O3, B=0.07Ha+0.75O3
- use HT in RGB/K to stretch more, use R, G, B individually to adjust star color and background using the white point and black point sliders.
- use an inverted L-mask to apply TGV denoise 30 or so
- using the residual mask from Ha apply more LHE to taste
As you can see I failed getting the O3 out. NB is not my strong side but maybe you can pick up a few steps that help you.