Fine-Tuning DenseNet121 for Effective Waste Categorization: Insights into Automated Sorting Systems

The rapid expansion of industrialization and urbanization has resulted in a notable increase in garbage generation, so effective waste management systems become ever more important. Automated solutions are especially important as conventional garbage sorting techniques are labour-intensive and prone to human error. This work investigates the application of convolutional neural network architecture DenseNet121 to categorize waste into six groups: cardboard, glass, metal, paper, plastic, and rubbish. We seek to maximize classification accuracy by using DenseNet121's dense connection and thereby minimize model parameters. We trained DenseNet121 for the task using a dataset taken from Kaggle and obtained an accuracy of 88%. The outcomes show the possibilities of the model in automating waste classification, so supporting sustainable cities, resource economy, environmental preservation, responsible manufacturing and consumption habits, and enhanced recycling procedures. We then address the design, training approach, and performance criteria, stressing the advantages and difficulties of including such systems in major waste management projects.