Remote Update of an IoT Device Convolutional Neural Network Model Using MQTT

This paper focuses on identifying a solution for an over the air (OTA) update mechanism to upgrade an artificial intelligence model that is running on an embedded system or an IoT device such as a Raspberry Pi in a closed network with limited or no internet connectivity. This paper first focuses on the creation of two machine learning models for image classification of five different types of flowers and focuses on using Kaggle and Google existing datasets for flower classification. The paper explains how an embedded device running the first ML model can be remotely upgraded to run the second more robust and optimized model. The paper highlights how the MQTT protocol used in conjunction with an MQTT local broker server can be used to design and implement an efficient, secure and reliable method for upgrading ML models on an IoT device running the TensorFlow Lite interpreter. This prototype works on a stand-alone network system with or without a cloud service provider or even the Internet.