[ Prev | Next ]

Continuous Improvement

• Implement a Neural Net solution for the accurate detection of stars in NorthCam Images.    To help fill a gap in safe protection of the Observatory (specifically the fast approach of snow showers approaching from the north), it was previously proposed to develop a realtime analysis routine in AstroNorthCam that can calculate Cloud Cover or suitable proxy from NorthCam Images, and make that data available to AstroMain.  This was not so easy to achieve due to the difficult of distinguishing cloud during the changing light condition during the dawn/dusk/twilight period (and during Aurora).  A more promising route was then explored to recognise stars and to transfer a count of the  number of stars (NorthCam StarCount) through to AstroAllSky & AstroMain.  However attempts to implement this in May 2025 encountered issues in tuning algorithums that were sufficientally sensitive to stars without having a lot of false-detections (see S1330, 2025-05-27).  It was then proposed to look into using neural net to distinguish between true stars and non-stars.  Working with CoPilot, routines were developed for building and an operating a neural net that could take a 21 x 21 pixel patch surrounding candidate stars objects as input and output a single neuron confidence value (0.0 - 1.0) (converted into 0-100%) that the patch contained a valid star object. The routines now need to be built into AstroNorthCam, tested/debugged and the net trained and test on older data sets, and then released for live testing. This work was done on 2025-08-15 and was  implemented in AstroNorthCam 1.5.4.  Coding was held up by overflow exceptions which were eventually determined to be coming from the following code line when the sum of the three color term values (R, G, B) is > 255.
     gray (double)  = ( C.R (byte) + C.G (byte) + C.B (byte) ) / 3.0 / 255.0
This was eventually resolved with the following fix which converts color term values to integer before addition :
     gray = (CInt(C.R) + CInt(C.G) + CInt(C.B)) / 3.0 / 255.0
  The new version was tested live on the night of 2025-08-15 / 2025-08-16 commencing at around 23:50. This was just in time to see clouds roll-in and the star count fall-off according and unfortunately force the suspension of the S1349A session, but proved the method / solution - see Fig 1 (S1349A) .  A final version (v1.6) was released 2025-08-16. A separate Continuous Improvement ticket has been raised ( S1349A, 2025-08-15) to work out how AstroMain's Obs Manager can utilise the NorthCam Star Count within automated Observatory Management.  The solution has been shown to also work in the presence of Aurora (see Fig 1 below)
   Done  2025-08-16 (AstroNorthCam 1.5.4 / 1.6)

[ Prev | Next ]

Back to Top Issue,  Back to Top

[ Prev | Next ]

Back to Top Issue,  Back to Top