This repository contains the code for the Final Project of the Internet of Things course in the Artificial Intelligence master degree at UniBo.
Machine learning and deep learning have become indispensable part of the existing technological domain. Edge computing and Internet of Things (IoT) together presents a new opportunity to imply machine learning techniques and deep learning models at the resource constrained by embedded devices at the edge of the network. Generally, machine learning and deep learning models requires enormous amount of power to predict a scenario. The aim of TinyML paradigm is to shift from traditional high-end systems to low-end clients. The challenges doing such transition are in particularly to maintain the accuracy of learning models, provide train-to-deploy facility in resource frugal tiny edge devices, optimizing processing capacity, and improving reliability. In this project, we present an application about such possibilities for TinyML in the context of handwritten recognition challenge. we propose the idea to make inference directly on a micro controller that is electronically linked with the input source of the character and obtain the prediction of it. This is possible thanks to the Tiny ML technologies available and in particular thanks to the TensorFlowLite library and its compatibility with the ESP32 microcontroller.
The model is based on the Lenet5 architecture and a modified version of it.
In the modified version the last convolution is substitued with a Dense layer of 256.
We create an application that allows to take the input and collect the prediction of every single letter in order to compose a word, shown on the display with a graphic interface, with three buttons each one associated to an action.
