Garbage Waste Segregation Using Deep Learning Techniques

Abstract

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.

Data Augmentation Using Background Replacement for Automated Sorting of Littered Waste

“…One of the most important datasets is the one gathered during the development of TrashNet [ 7 ]. This dataset has been the basis for many other studies [ 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 ]. The TrashNet dataset has been further extended by adding the compost class [ 10 ].…”

“…In fact, these images did not accurately represent the state of wasted goods as most of them represents undamaged or totally unused objects. The TrashNet dataset has been largely used [ 7 , 8 , 10 , 11 , 13 , 15 , 16 , 20 , 21 ]. The set of additional images in CompostNet has the same features of the rest of the TrashNet dataset.…”

“…For example, waste pickers of the most important landfills in Brazil have been, or still are, subject to many diseases, due to working conditions: mainly osteo-muscular disorders, but also viral infections and emotional vulnerabilities [ 6 ]. Therefore, current waste recognition systems [ 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 ] as well as current corpora for training [ 7 , 10 ] focus on the activity of waste sorting in waste management plants. Solid waste runs on conveyor belts and, then, it is sorted into the different types.…”

Garbage Classification Using Ensemble DenseNet169

“…Penelitian ini menggunakan dataset yang digunakan pada penelitian [7], [9]. Dimana dataset ini diambil dari Gary Thung dan Mindy Yang's dataset berisi jenis sampah yang dikategorikan menjadi enam kelas diantaranya : cardboard 403 gambar, glass 501 gambar, metal 410 gambar, paper 594 gambar, plastic 482 gambar, dan trash 137 gambar.…”

“…Kenaikan accuracy terjadi sebesar 3% hingga 5% dari model tunggal, dengan performa accuracy akhir pada metode ensemble sebesar 96%. Sehingga, metode ensemble yang diterapkan mampu meningkatkan accuracy pada dua penelitian sebelumnya dengan dataset serupa yang memiliki accuracy masingmasing sebesar 93,6% dan 94,9%[7],[9] menjadi 96%. Seiring dengan peningkatan accuracy parameter lain juga mengalami peningkatan diantaranya precision, recall, dan f1-score dari masing-masing kelas.…”