The detection is done with an artificial intelligence already trained from numerous calibration images where the position of the emission lines has already been identified..
Model's training is therefore done overall by showing it a set of raw NeAr spectral image data on which some lines have been identified. The algorithm will then learn to recognize them.
The performance of the model will increase with the data we can provide, the more data, the better the training.
If you do spectroscopy with an Alpy600, you can participate in its improvement simply by providing your calibration images.
The model currently detects 3 to 4 ArNe lamp lines with a 99% accuracy for very visible lines. But there is still work to for use it, for example in an educational framework for the discovery of spectroscopy.
In order to have a slightly more performant model, it is preferable to have additional lines detected.
Similarly, other types of machine learning models are being explored.
This project is purely experimental and aims to study the possibilities of machine learning technologies associated with the field of amateur spectroscopy.
Current automatic calibration solutions, particularly in the Alpy600, work very well and on many lines. The objective of this project is not to look for alternatives, but to study the possibilities of AI.
(See here for better algorithms : Demetra - Shelyak)
You do astronomical spectroscopy ? You have a spectroscope? So you can help !
Detection model training is based on a set of calibration image data. The more calibration images there are for training, the more precisely the model learns.
You can improve the model simply by sending your calibration file via the button above.