Date of Award
Doctor of Philosophy (PhD)
The amount of Internet-of-things (IoT) devices is rapidly expanding. This also triggered the necessity of smart IoT devices which are capable of conducting any task by itself. Deep learning techniques are also booming due to the increased computing power and refined algorithms. The advantage of deep learning is that it can be tuned into any application without the manual feature extraction process. Now, the combination of deep learning with smart IoT devices/edge devices can result in any application that can be used in machine vision, vision inspection, autonomous vehicle, and many more. These applications can be automated which requires human operation. Now, combining deep learning and edge device together and running the application can be a difficult task. The main reason is that deep learning requires large computation power and edge devices does not have that capability. This study focused on this problem. Ie used techniques to encrypt and compress data which is essential for the edge devices. In addition, we developed novel methods to protect user privacy for data collection and learning on edge devices. Also, we conducted a study to evaluate different edge devices for different application purposes with certain compression technique of the models. Lastly, we conducted a real life experiment of collecting data, creating different models and evaluating it on different edge devices.
index terms - IoT, computer vision, deep learning, machine learning, quantization, autoencoder, mobilenet v1, mobilenet v2, inception v3, face mask detection
Prairie View A&M University
Rights© 2021 Prairie View A & M University
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
Date of Digitization
J. B Coleman Library
Reza, S. R. (2023). Deep Learning For Resource Constraint Devices. Retrieved from https://digitalcommons.pvamu.edu/pvamu-dissertations/27