Smart Waste Classification System-Classifying Waste by Image Recognition

Ting, Nathan Chow Yew (2023) Smart Waste Classification System-Classifying Waste by Image Recognition. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text RDS_Nathan Ting Chow Yew_Fulltext.pdf Restricted to Registered users only Download (1MB)

Abstract

Waste is a common by-product in our daily routine. As more waste is produced on a daily basis, sustainable waste management is becoming a concern in most countries, especially the rapidly developing ones with booming populations. This sustainable option is usually the 3Rs of Reduce, Reuse and Recycle. Focusing on the recycling aspect, it requires classifying waste into certain categories such as paper, metals, and plastics since the process of recycling each category of waste differs. The process of classifying waste into recyclable and non-recyclable, then into their sub-categories, can be rather complicated due to the specificity of these sub-categories. This project aims to provide an application that can help people determine whether the waste in their hands is potentially recyclable, then into their sub-categories which not only helps users to separate their waste correctly, but also saves time for workers at recycling centres responsible for picking out wrongly disposed waste at the start of the recycling line. Our application will take in images of the waste from the user where a trained image classifier will identify the waste and determine its recyclability. This project is developed using a mix of Incremental and Agile software process models. Tools and libraries such as Selenium, TensorFlow, TensorFlow Lite, and Android Studio are used in the image scraping, the building and training of a Convolutional Neural Network (CNN) as the image classifier, and the final deployment of the mobile application. The CNN is tested using a collection of 26 images gathered from internet sources and real-life image captures. Overall, even though the working concept and the deployment of the solution is successful, the performance of the image classifier is unsatisfactory, especially when it comes to classifying real-life captured images. The limitations and issues are thus identified with potential solutions noted for future works