Machine intelligence has great potential in astronomical image processing. My image yesterday of the Horsehead and Flame nebulas used the Topaz DeNoise AI application to reduce noise in the faint nebulosity. Today I took that image and used the StarNet module in PixInsight to extract the stars and nebulosity as separate images for additional processing. Stars were reduced with a morphological size reduction and the nebulae were sharpened to bring out a bit more detail. Both tools used trained neural networks to work their magic. The final image is a stack of these two processed layers and the original image.

For comparison here is the original image.

And here the StarNet extracted nebulosity layer.

The Horsehead and Flame Nebulas in Orion from my driveway with a last quarter moon under Bortle 6+ skies. William Optics Redcat 250/51 mm telescope Optolong L-Pro broadband light pollution reduction filter, ZWO ASI 533 MC Pro at -10C, gain 1.01, 112 minutes of exposure in 60” subs. Camera control and polar alignment with a ZWO ASIAIR and iPad. Vixen Polarie single axis unguided mount on a vintage Sanford & Davis tripod.

Processed in PixInsight and StarNet, with Topaz DeNoise and final exposure and crop in Photoshop.

For a peek at the original data. Here is a scaled, stretched, and background neutralized view of the 112 image stack:

Blow a scaled, stretched, and background neutralized view of a single frame: