is it possible to read analog linear scale?

[mortenx]

Is it possible to read a linear scale like a thermometer or rotameter?

[Slider0007]

Hi @mortenx,

I've never used the software for this scenario, but in principle it could be possible. But for sure, you would have to train your own model.

I assume, training a linear scale based on dig-cont / ana-cont model approach (producing the same model output dimension, result mapped to sin/cos --> https://github.com/jomjol/neural-network-analog-needle-readout) could work without any software modification. At least raw value without post-processing could be used in any downstream automation system (e.g. home assistant) for further processing.

[mortenx]

Is there any tutorial somewhere on how to train a linear model for this specific thing? Do you think this neural-network-analog-needle-readout also works on linear? , I've never done it but I might try.
I need to graph the rotameter data from a gas measurement experiment at university.
I think it would also be a way to get the data from a button like a fader.

[Slider0007]

It will not work out of the box, therefore you need to train your own model. But the approach can be used as basis to train. In principle the process is always the same indepentent what you want to train / detect. Anything can be trained and detected (quality depends a lot on image quantity and quality / labeling)

As a first step, please check this discussion. Another user were also asking for model training recently: jomjol#2337

You could also try to search for other discussions (https://github.com/jomjol/AI-on-the-edge-device/discussions) or check the docu repo https://jomjol.github.io/AI-on-the-edge-device-docs/ for further infos.

General workflow:

1. Image collection with rectangle digit ROI (images neend to cover the whole value range, well distributed over the range)
2. Image labeling (specifiy, which value the respective image is respresenting)
3. Training (starting with model defined in https://github.com/jomjol/neural-network-analog-needle-readout)
4. Model verification by using it as digit model in software (to be able to use a rectangle ROI). Never used it that way, though...