# Comparative Analysis of Multiple Deep CNN Models for Waste Classification

**Authors:** Dipesh Gyawali, Alok Regmi, Aatish Shakya, Ashish Gautam, Surendra Shrestha  
**arXiv:** [2004.02168](https://arxiv.org/abs/2004.02168)  
**Date:** 2020-04-05  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2004.02168v2/

## Key points

- Research focuses on automatic waste sorting and collecting for recycling
- Current waste management approaches are inefficient and require human intervention
- Project aims to implement automatic waste segregation using deep learning network architectures
- Combines datasets from their own endeavors and Trash Net to train a convolutional neural network for image classification
- Hardware system in the form of a dustbin is built to segregate different types of waste into separate compartments
- Eliminates the need for human intervention in waste segregation, saving valuable time and introducing automation in waste management
- Municipal solid waste is a significant source of renewable energy, benefiting the government, society, and industrialists
- Best validation accuracy achieved through fine-tuning of ResNet18 Network was 87.8%
- Comparative analysis shows promising results with an accuracy above 87% for multiple deep CNN models
- Further improvements can be made by expanding the dataset and fine-tuning the model properly
- Reliable and accurate waste classification system can be utilized in recycling processes if implemented in real-life scenarios

## Summary

The research focuses on analyzing possibilities for automatic waste sorting and collecting to facilitate the recycling process. Current waste management approaches are not efficient and require human intervention. The project aims to fill this gap by implementing automatic waste segregation using deep learning network architectures for waste classification. The study combines datasets from their own endeavors and Trash Net to train a convolutional neural network for image classification. A hardware system in the form of a dustbin is built to segregate different types of waste into separate compartments. By eliminating the need for human intervention in waste segregation, this study saves valuable time and introduces automation in waste management. Municipal solid waste is a significant source of renewable energy, making this situation beneficial for the government, society, and industrialists. Through fine-tuning of the ResNet18 Network, the best validation accuracy achieved was 87.8%. The comparative analysis of multiple deep CNN models shows promising results with an accuracy above 87%. Further improvements can be made by expanding the dataset and fine-tuning the model properly. The reliable and accurate waste classification system can be utilized in recycling processes if implemented in real-life scenarios. The researchers express their appreciation to their supervisor for his support, motivation, and guidance throughout the project.