Date of Award
6-2023
Document Type
Thesis
Publisher
Santa Clara : Santa Clara University, 2023.
Department
Electrical and Computer Engineering
First Advisor
Hoeseok Yang
Abstract
TinyML is a rapidly developing field of machine learning that focuses on deploying complex neural networks on to edge devices such as microcontrollers and phones. The goal of this project is to train and deploy an image classification neural network on top of a Raspberry Pi Pico4ML that can be used in an application to assist the visually impaired. In this project, we will use the TensorflowLite for Microcontrollers platform in order to train and convert a convolutional neural network from Keras into a lightweight tflite model that can be deployed to the Pico4ML. Our tasks include training the image classification model on the Cifar10 dataset, modifying the model to run on a smaller subset, and changing the clock speed in order to optimize performance. Our goal is to optimize power consumption and inference speed while maintaining an accuracy that could be deemed safe. By optimizing these aspects of our model, we can create an efficient and meaningful application for the visually impaired.
Recommended Citation
Chun, Robb and Hasan, Anonna, "Low Power TinyML For Image Recognition" (2023). Electrical and Computer Engineering Senior Theses. 86.
https://scholarcommons.scu.edu/elec_senior/86