UNVEILING THE POTENTIAL OF DEEP TRANSFER LEARNING MODELS FOR WASTE CLASSIFICATION: A COMPREHENSIVE ANALYSIS

Abstract

Waste segregation has emerged as a pressing global issue in recent years, with significant environmental implications. A staggering 242 million tonnes of plastic waste were generated worldwide in 2016, constituting 12% of total solid waste, as highlighted by a study conducted by the World Bank in 2018. The persistence of non-biodegradable waste in the environment poses significant challenges and threatens ecosystems. Conversely, biodegradable waste has minimal environmental impact, undergoing degradation through natural forces like fire, water, air, microorganisms, and soil.

In the field of deep learning, Convolutional Neural Network (CNN) models have proven instrumental in training and testing large image datasets. These models utilize convolutional layers comprising filters, pooling layers, and fully connected layers to process input images. To address the issue of insufficient training data, transfer learning, a powerful technique in deep learning, has gained prominence due to its wide range of applications.

This paper focuses on reviewing transfer learning models within the Keras library and utilizing CNNs to classify waste into biodegradable and non-biodegradable categories. Furthermore, the study investigates the impact of crucial hyperparameters such as activation function, batch size, learning rate, and optimizers on waste image classification. To identify the optimal hyperparameters, the CNN model's train, test, and validation accuracy are considered.

The paper is structured as follows: the second section provides a comprehensive review of related works, while the third section outlines the dataset and methodologies employed for waste classification. Finally, the concluding section presents a summary of the study's findings. This research contributes to the development of effective waste management strategies by leveraging deep learning techniques for accurate waste segregation and classification