Open AccessGarbage Waste Segregation Using Deep Learning Techniques
Author(s) -
G Sai Susanth,
L.M. Jenila Livingston,
Lawrence G. Livingston
Publication year - 2021
Publication title -
iop conference series. materials science and engineering
Language(s) - English
Resource type - Journals
eISSN - 1757-899X
pISSN - 1757-8981
DOI - 10.1088/1757-899x/1012/1/012040
Subject(s) - garbage , waste management , convolutional neural network , municipal solid waste , workforce , environmental science , classifier (uml) , computer science , engineering , artificial intelligence , economic growth , economics
Waste segregation is one of the primary challenges to recycling systems in major cities in our country. In India, 62 million tons of garbage is generated annually. Of this 5.6 million tons of wastes consist of plastic materials. About 60 percent of this is recycled every year. In addition, 11.9 million tons are recycled from 43 million tons of solid waste produced. Though the numbers sound good, a serious problem in the recycling industry is the segregation of waste before recycling or any other waste treatment processes. In India, at present situation waste is not segregated when collected from households. So a lot of workforce and effort are needed to separate this waste. In addition to this people working in this industry are prone to various infections caused due to toxic materials present in the waste. So the idea is to decrease the human intervention and make this waste segregation process more productive. The proposed work is aimed to build an image classifier that identifies the object and detects the type of waste material using Convolutional Neural Network. In this work, four different models of the CNN, such as ResNet50, DenseNet169, VGG16, and AlexNet, trained on ImageNet, are used to extract features from images and feed them into a classifier to make predictions and distinguish a type of waste from its corresponding category. The experimental results showed that the performance of DenseNet169 was significantly greater than all four models and the performance of ResNet50 is closer to DenseNet169.