Deep Learning-based Education Waste Classification for School Students

Abstract

The purpose of this project is to overcome the problem of waste management in Indonesia through the development of an interactive web-based learning application for elementary school students. The program classifies garbage into 26 categories under 9 parent classes through the use of deep learning, a Convolutional Neural Network (CNN) built in Tensorflow.js that takes real-time video input. The model was trained and tested on 5,661 labeled images in a controlled laboratory setting and achieved an accuracy of 86.73%. The accuracy rate is 71.20% after the test with 35 kids of grade 4 to 6 of a private elementary school in Indonesia with adjustments of illumination, object blocking and camera quality. The software includes gamification elements such as mini-games and digital badges as rewards. Pre-and post-tests showed an average increase of 29% in students’ understanding of waste classification. User Acceptance Testing confirmed that all major features worked as designed. This paper presents an AI-based educational game for kids combining state of the art image recognition with real world environmental teaching. This program promotes sustainable development target 4: Quality Education by promoting green practices in an engaging digital environment.