NORMA eResearch @NCI Library

Waste Classification by Using Deep Learning and Transfer Learning

Pote, Suchal Suhas (2022) Waste Classification by Using Deep Learning and Transfer Learning. Masters thesis, Dublin, National College of Ireland.

[thumbnail of Master of Science]
PDF (Master of Science)
Download (3MB) | Preview
[thumbnail of Configuration manual]
PDF (Configuration manual)
Download (805kB) | Preview


Improper waste management directly impacts numerous habitats and species as well as air pollution and climate change. If waste can be identified at an early stage and classified, it can then be recycled or disposed of depending on the type of waste. This waste segregation helps to lower the cost of treatment, uses fewer land resources, and is also beneficial for social, economic, and ecological factors. This project aims to efficiently categorize the most common types of waste at the early stage by using deep learning frameworks on waste datasets. This separated waste can be further recycled or disposed of. The TrashNet dataset was previously used in many studies that classified only six types of waste. This research is being carried out in order to broaden the waste categories for the classification of other common types of waste. This is achieved by customizing the TrashNet dataset with additional classes. The transfer learning method is used to construct ImageNet pre-trained deep learning algorithms Vgg16, InceptionV3, Xception, and DenseNet201 to classify the 10 various forms of waste. The model’s performance is evaluated in order to determine accuracy and loss. All the models suffered from overfitting in the first attempt. The ‘Dropout Regularization’ and ‘Batch Normalization’ in Keras are used for all deep learning algorithms to prevent overfitting of the model. Overfitting was not observed in the second attempt, also the accuracies of each model were increased by 4% to 5%. The performance of DenseNet201 is satisfactory with a validation accuracy of 89%, whereas Xception has achieved a validation accuracy of 87%. With a validation accuracy of 84%, InceptionV3 has good performance. Additionally, Vgg16 performs averagely with a 74% validation accuracy. The performance of all four models has been compared with each other to decide the most efficient algorithm. Several common types of waste can be successfully categorized for recycling or disposal using the suggested method.

Item Type: Thesis (Masters)
Staikopoulos, Athanasios
Uncontrolled Keywords: Deep learning; Waste; Transfer Learning; Vgg16; InceptionV3; Xception; DenseNet201
Subjects: Q Science > QA Mathematics > Electronic computers. Computer science
T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science
T Technology > TD Environmental technology. Sanitary engineering
Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions: School of Computing > Master of Science in Data Analytics
Depositing User: Tamara Malone
Date Deposited: 23 May 2023 17:01
Last Modified: 23 May 2023 17:01

Actions (login required)

View Item View Item