A Study on Solid Waste Classification Through Deep Learning Models - A Comparative Approach

Abstract

The study aims to classify solid waste materials into three types using machine learning models and compares the performances of the machine learning models. Three standard models namely ResNet-50, VGGNet-16, and Inception-V3 model have been used in this research along with a newly developed custom-CNN (convolutional neural network) model for waste classification.

Methodology: Datasets of solid waste materials were acquired from ‘Kaggle’ as images for waste management. The images are processed and passed through all four models and the output is classified through image detection and classifier algorithms (XGBoost, Random Forest, and FC-DNN: Fully-Connected Deep-Neural-Network algorithm). The classified images as outputs are then stored under labels N: Non-recyclable, R: Recyclable, and O: Organic. Python’ is used as the programming language software.

Findings and Performance evaluation: The obtained outputs are compared with ‘accuracy rate’ as a metric evaluation technique. Findings showed ResNet-50 model with the ‘Random-Forest’ classifier algorithm is more efficient and reliable than other models with a 72.44% accuracy rate.

Conclusion: The study concludes that ResNet when combined with the Random-Forest model is to perform efficiently in classifying solid waste materials with greater accuracy.