Incorporating Client-Specific Information to Improve Image Classification for Food Waste Management
Keywords
Loading...
Authors
Issue Date
2025-01-01
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Currently, the food industry is one of the largest contributors to greenhouse
gas emissions, and an estimated one third of all produced food goes to waste.
Orbisk attempts to reduce the amount of food that goes to waste by devel
oping a food waste recognition system, which is deployed in restaurants in
the hospitality industry. Using a camera and artificial intelligence, their
smart bin automatically captures recognizes what kind of food is thrown
away and in what quantity by segmenting and classifying the different items
in a picture. This research aims to improve the ingredient classification task,
by exploring the possibilities of incorporating client-specific information. A
thorough data analysis is conducted to get insights on each client’s data,
and to select several high-potential clients, and one cluster of clients, which
are used for experimentation. Two methods of incorporating client-specific
information to improve classification accuracy are explored: post-processing
the model output to remap food items based on a client’s historical data,
and fine-tuning the general model on client-specific data. Results demon
strate that both post-processing and fine-tuning methods significantly im
prove classification accuracy for some clients, whilst not proving effective for
others. While challenges remain in generalizing these improvements across
diverse clients, this research provides a foundation for future efforts to opti
mize image classification models by integration of client-specific information.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen