Image classification on an ARM Cortex M Microcontroller

Prototyping and Deploying a neural network for image classification using MNIST data on an ARM Cortex M Microcontroller

In this example, we have a Simulink model based on the shallow network with five layers described in Loren’s blog below for image classification using MNIST data:
https://blogs.mathworks.com/loren/2015/08/04/artificial-neural-networks-for-beginners/
We have three versions of the model – in double precision, single precision and a fixed-point version. These models can then be tested using live data from an Arduino Due using the Simulink® Support Package for Arduino® Hardware and deployed to the Arduino board as a standalone application.
The single precision and fixed-point versions were generated using Fixed-Point Designer as described in the links below. The fixed-point model uses no more than 16 bits and the accuracy of the model is above 94%.
https://www.youtube.com/watch?v=sxSodI0pwPw
https://www.youtube.com/watch?v=zX44UvyLeAc
https://www.youtube.com/watch?v=nkZAB7LIRXI&t=12s