Recently I finished a project which combined stm32f030r8 microcontroller and MPU9250 sensor to create a system which would detect orientation on all 3 axis using a combination of accel data, gyro data and magnetometer data.
Now, I got an idea to expand the project by building a neural network and training it using Python on a PC and then transfer the trained neural network onto a microcontroller in order to be PC-independent and actually be able to recognize the written stuff on it's own. I am still contemplating how to build a database from the microcontroller data and what the actual neural network structure should be.
My first thought was to build a neural network which would have n inputs (since I'm still figuring out what the database should look like, I don't know n), 5-10 hidden single-layer neurons and an output for each letter and number so that I can recognize for example letter 'A' as as 100...0.
Is this idea feasible on a microcontroller since I'm not actually training the neural network on it? What do you think of the idea so far and could you suggest some sources to read upon, or some courses to help me with this?
Cross-posted to EE