Classification of Wastes by Utilizing Image Processing and Convolutional Neural Networks


İnsel M. A., Baş N., Yücel Ö., Sadıkoğlu H.

7th Eurasia Waste Management Symposium, İstanbul, Turkey, 21 - 23 October 2024, pp.1-8

  • Publication Type: Conference Paper / Full Text
  • City: İstanbul
  • Country: Turkey
  • Page Numbers: pp.1-8
  • Yıldız Technical University Affiliated: Yes

Abstract

Effective waste management is essential for safeguarding human health and ensuring a clean environment. A critical component of waste management is the systematic separation of waste based on its categories. In this study, the classification of wastes is carried out by utilizing convolutional neural networks. Firstly, a novel dataset of 3035 waste images that contains 7 different categories (battery, glass, metal, organic, trash, paper-cardboard, plastic) is constructed by combining TrashNet image dataset (2527 images) and our own dataset (508 images). Then, the dataset was randomly divided to train (80% of data) and test (20% of data) sets to perform the hold-out validation. Finally pre-trained convolutional neural network ResNet50 is trained on the train set, and the model’s validity is checked by comparison with the test set. It is shown that ResNet50 is able to classify the wastes with high accuracy value (99.1%), even if the training is stopped at comparatively low epoch numbers. Our results indicate that the wastes can successfully be classified by deep learning from image data and the prevalent future direction regarding this topic is the detection and classification of wastes directly from videos in continuous processes.