Making the most of bad data

[Hicks]

On Tuesday I bought a set of LRGBC baader filters for imaging and as luck would have it, for the first night in weeks we had a spell of clear skies the following day (same day my narrowband filters arrived, I guess they even filter out the clouds in the box

Focus offsets need more testing, but I managed to capture (NGC7331) around 70x60s exposures in L,R,G and B. Having had my best night guiding I went to take flats/darks and to my horror realised I'd forgotten to connect the dew heater tape to my heater controller and the corrector was all nice and foggy.

Looking back on all the L images via subframe selector you can see the SNR take a gradual/increasing hit after around the first 30 frames. Fortunately I did 1L, 1R, 1G, 1B and then looped around again so I do have data to work with pre-fog. What I'm interested to know is what people feel about including the "bad" data perhaps heavily weighted to favour the good?

How bad was it? Well here's the first image of the night and the last, the difference is quite striking.

The most frustrating part is, I greatly improved my guiding after the dew had formed*. To the point it was close to guiding out the harsh periodic error that hit every 9 minutes. How would you tackle such a mismatch of data? Dump the dew'd frames and stick to the few good frames from the first 30 (poor guiding/more bloated stars)? Or try to weight in the later frames where guiding was good, stars are tighter and the background is increasingly foggy each frame

tbh the main reason I'm even bothering to process this and not chalking it up to a just one of those nights, is that it could be a month+ before we see another clear night and I've much to learn about RGB processing in pixinsight, never done it before. So, some data works better than none even if the results will be limited.

I did a calibration, integration and combine of all 72 LRGB images last night both the good, the bad and the ugly with no weighting or processing beyond   a bit of histogram stretching I got:

Was a little surprised by how that turned out considering how bad (in different ways) the data is from either half.

Colour is very washed out. Is that down to lacking in data? there's about 72 minutes of R, G and B. Also I'm under bortle 8 skies I believe.

There's a gradient from left to right too, not sure if that's due to including the dew'd frames or perhaps some local light pollution.

I've not done any processing as far as sharpening/masking or trying to boost any of the colour. Any scripts/processes I should read up on for pulling a bit more out of this data?

* Lesson learned, added check dew heater cable attached to checklists and also added a 30-60 minute periodic check of the corrector to ensure heater is actually keeping up

Edited October 4, 2019 by Hicks
Attached images rather than linked.

[vlaiv]

I'm in favor of using all the data, but for that to happen, you need some clever algorithms.

I know that PI has sub selector and that it can assign weight based on estimated noise. So maybe you should try that.

On the other hand, having worked on algorithms for exactly that purpose, I can tell you that there is no single weight for entire sub that will produce optimum result. Combining different SNR sources requires weight per SNR and SNR depends on both signal and noise, so measuring noise in one part of the image will not give you complete picture of SNR across the image and how to combine different subs together for optimum result.

Have a look here for actual comparison of rejecting subs vs stacking them with different weights (but my approach is different than PI sub selector so you might get different results if you work with PI):

I also have part of stacking workflow developed that sort of deals with LP gradients. Let me see if I can find that thread as well. Here it is:

I have efficient residual gradient removal tool as well

If you wish, you can post your subs (or put them somewhere online like google drive or such) and I can put them thru my workflow to see what we can come up with in terms of end result

[Hicks]

Thanks for the links, I'll have a read.

If you want to have a play with the subs, I've pm'd you a link to the download (821MB 7zip). Note with the flats, first 100 were compromised by dew also. Last 50 were taken with a clear corrector.

Be interesting to see how your process stacks up (pun intended). I'll hopefully find some time next week to re-process and learn more about RGB in pixinsight to improve on my attempt :)

[Hicks]

Just looking at the top left image and wondering if the eggy shape to the stars is likely polar align error causing field rotation.

Anything stand out as a likely cause? PA, tilt or focal reducer spacing? If it's PA I can improve that as I only went to a 125s DARV exposure.

I think the next clear night, I could do with getting a few subs of a rich star-field and very short exposures to eliminate PA/guiding.

Edited October 6, 2019 by Hicks

[vlaiv]

Btw, what gear did you use to capture this?

 

[Hicks]

LX90 8" fork mounted with F6.3 reducer. Moonlite focuser, SX Filter wheel, Lodestar OAG and SXVR-H9 main ccd.

Polar alignment was only rough, with a 125s DARV exposure showing overlapping lines but I expect 185s+ would show more refinement needed. Guiding was pretty poor until second half of the night when I managed to make some improvements. Still more to practice there. Seeing conditions were reported as 1.75" which is almost double the 1.04"/px I was imaging at, might have been worth binning.

 

Edited October 6, 2019 by Hicks