Real-time Food Waste Classification Using Machine Vision

Lim, Keng Yung (2025) Real-time Food Waste Classification Using Machine Vision. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text Lim Keng Yung - FULL TEXT.pdf Restricted to Registered users only Download (5MB)

Abstract

This project aims to reduce food waste, minimize waste sorting costs, and improve real-world datasets for AI-based classification. A food waste detection system was developed using YOLOv8, targeting Sustainable Development Goals (SDGs) 6, 13, 14, and 15 by addressing environmental issues such as water pollution and ecosystem degradation. Previous efforts achieved around 70% accuracy due to limited real-world datasets. This project improves performance to nearly 80%, especially for the "NonBiogasReady" category, which is prioritized for separation in waste treatment. The dataset comprises 2,020 original images—860 labeled as ‘BiogasReady,’ 760 as ‘NonBiogasReady,’ and 400 with both. Each train image was augmented with 14 variations, increasing the dataset to 9,030 ‘BiogasReady,’ 7,980 ‘NonBiogasReady,’ and 4,200 mixed-class images. After augmentation, a total of 21,210 images were used, split in a 7:2:1 ratio for training, validation, and testing. The system supports real-time detection by identifying food waste items and their coordinates. It achieved an accuracy of 0.801, with a precision of 0.675, recall of 0.728, and F1 score of 0.700, demonstrating reliable classification and localization. These results indicate that YOLOv8 is a suitable and scalable model for practical food waste management applications.