Real-time solution to recycling, supercharged with machine learning
Le, Tuan (2022)
Le, Tuan
2022
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2022060315184
https://urn.fi/URN:NBN:fi:amk-2022060315184
Tiivistelmä
Within the past few years, machine learning has been rapidly growing and has become the key element in a wide variety of departments, including health care, science, technology, finance and many more. In parallel, new tools and libraries are introduced to make machine learning even more accessible and encourage more products developed with machine learning capability.
Embracing the same vision, this thesis discusses a popular platform for machine learning, known for its comprehensive resources and ease of use – TensorFlow. Consequently, a smart mobile solution for recycling problems was created to utilise the power of image classification with machine learning to simplify the process. The thesis also documents how TensorFlow Lite was chosen and integrated for its on-device machine learning optimization. Android Neural Networks API was also implemented to push the performance even higher. The final product is an Android app that implements an image classifier to recognize objects from the camera feed in real-time and suggests the proper actions to recycle them.
This research aims to study the performance of machine learning on mobile devices and the ease of access that libraries like TensorFlow provides to machine learning while aiming to align the technology closer to daily life by introducing an application that solves our recycling problems.
Embracing the same vision, this thesis discusses a popular platform for machine learning, known for its comprehensive resources and ease of use – TensorFlow. Consequently, a smart mobile solution for recycling problems was created to utilise the power of image classification with machine learning to simplify the process. The thesis also documents how TensorFlow Lite was chosen and integrated for its on-device machine learning optimization. Android Neural Networks API was also implemented to push the performance even higher. The final product is an Android app that implements an image classifier to recognize objects from the camera feed in real-time and suggests the proper actions to recycle them.
This research aims to study the performance of machine learning on mobile devices and the ease of access that libraries like TensorFlow provides to machine learning while aiming to align the technology closer to daily life by introducing an application that solves our recycling problems.