SmartyBin: A Tangible AI-Based Waste Classification Prototype


Al Shareeda S. Y. A., Hasan H., Alkhayyat A.

2023 International Symposium on Networks, Computers and Communications, ISNCC 2023, Doha, Qatar, 23 - 26 October 2023 identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/isncc58260.2023.10323940
  • City: Doha
  • Country: Qatar
  • Keywords: classification, convolutional neural networks, deep learning, sustainability and environmental support, waste management, zero waste
  • Istanbul Technical University Affiliated: Yes

Abstract

This paper introduces a tangible intelligent trash classification bin that leverages Deep Learning (DL) techniques to efficiently identify and accurately classify diverse waste materials, including metal, plastic, paper, and others. To build a robust classification model, a dataset comprising 4,122 colored labeled JPG images of waste materials was collected from Kaggle. The dataset was divided into a training set, which constituted 70% of the images, and a validation set, which comprised the remaining 30%. The proposed system achieves an impressive validation accuracy score of 94.53% for real-time waste classification. To create an affordable solution, we constructed the trash bin from scratch, incorporating an ESP32CAM camera to capture images of the waste materials. By utilizing the trained Convolutional Neural Network (CNN), the developed bin achieves an approximate 98% accuracy in classifying waste materials, thus significantly enhancing the effectiveness of waste management practices. This paper introduces a tangible intelligent trash classification bin that leverages Deep Learning (DL) techniques to efficiently identify and accurately classify diverse waste materials, including metal, plastic, paper, and others. To build a robust classification model, a dataset comprising 4,122 colored labeled JPG images of waste materials was collected from Kaggle. The dataset was divided into a training set, which constituted 70% of the images, and a validation set, which comprised the remaining 30%. The proposed system achieves an impressive validation accuracy score of 94.53% for realtime waste classification. To create an affordable solution, we constructed the trash bin from scratch, incorporating an ESP32CAM camera to capture images of the waste materials. By utilizing the trained Convolutional Neural Network (CNN), the developed bin achieves an approximate 98% accuracy in classifying waste materials, thus significantly enhancing the effectiveness of waste management practices.