Training vectors for pattern recognition

Hi all,

Absolute newbie here.
I am an EE student working on a solar panel analyzer gizmo. I aim to embed an A.I. algorithm (ANN) on a STM32 MCU to recognize the shape of a solar panel characteristic extracted by hardware (I vs V matrix), and provide a bill of health for the panel. Haven’t got a proj. ID yet.
At the moment I’m having a bit of a headache with the training vectors.
My current format is [expected power output in %, I0, V0, … I50, V50]. For example I have a collection of 100 vectors like this: [100%, 8.499, 0.01, … 0.09, 43.983], [90%, 8.485, 0.009, … 0.087, 43.992], etc.

In the examples provided on the edgeimpulse site, the first value is a time increment, which I gather is also an input, but in my case the first element is the desired output.
I was wondering at what stage in the training I need to provide the expected output for each vector?

Best regards,
Bogdan

I think my application is similar to OCR - I was wondering if anyone would like to share their insights with OCR training vectors please? So far I’ve had a read through OCR-related forum threads but without much success.

Cheers,
B.

Hi @barefta

Can you share the links you tried? OpenCV character recognition was one option, or have you tried via BYOM? Bring your own model (BYOM) | Documentation

Best

Eoin