Smart Waste Sorting Using Deep Learning and Image Augmentation

Abstract

The global challenge of waste management has caused significant environmental and health concerns. As the volume of waste continues to increase, manual sorting methods are proving inadequate and potentially hazardous. This research aims to develop an automated waste categorization system using deep learning algorithms to provide an innovative solution for waste classification and recycling. We have used images from trashnet dataset to train and test our model as trashnet dataset contains images from six different waste category that are close to real world. We implemented image augmentation techniques on the dataset to enhance the model's learning capacity. The classification task was conducted using the ResNeXt-50 convolutional neural network, pre-trained on the ImageNet dataset. The model underwent training and validation across 50 epochs to ensure its accuracy and generalization capabilities. The methodology employed demonstrated robust performance, achieving a maximum validation accuracy of 99.17%. T The model exhibited better generalization performance than some of the recent works reported in the literature, which proves its applicability for the waste classification task. The innovation of this research resides in the integration of image augmentation with the ResNeXt-50 architecture for waste classification, coupled with the achievement of substantially higher accuracy compared to previously reported models. This research demonstrates the potential for developing scalable, automated waste sorting systems utilizing deep learning techniques.