Work

TrashNeXt: Classification of recyclable water pollutants using deep transfer learning method

Year: 2024

Type: article

Abstract: An automatic waste classification system embedded with higher accuracy and precision of convolution neural network (CNN) model can significantly the reduce manual labor involved in recycling. The ConvNeXt architecture has gained remarkable improvements in image recognition. A larger dataset, called TrashNeXt, comprising 23,625 images across nine categories has been introduced in this study by combining and thoroughly analyzing various pre-existing datasets. The deep transfer learning (DTL)-based proposed model achieved the highest accuracy of 94.97% compared to other CNN models by applying image augmentation and comprehensively fine-tuning hyperparameters. Additionally, the trained and optimized weights are utilized to classify water-bound liter objects. • Extensive evaluation of the proposed solid waste dataset applying various DTL models. • A New ConvNeXt-based DTL model, trained on the proposed dataset and benchmarked against TrashNet. • Additional training on Plastic and Foam_Ruber classes to sub-classify the floating trash on water.

Source: Case Studies in Chemical and Environmental Engineering

DOI: 10.1016/j.cscee.2024.101073

Authors: Jahid Tanvir, Sk. Tanzir Mehedi, Bikash Kumar Paul, Monir Morshed

Institutions: Mawlana Bhashani Science and Technology University, University of Asia Pacific, Daffodil International University, Virginia Tech

Language: English

Topic: Advanced Neural Network Applications

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

SDGs: Responsible consumption and production, Clean water and sanitation

Open Access status: gold

Estimated APC paid: $760