Intelligent waste classification system using vision transformers

Abstract

The issue of waste management is a growing concern, ranging from agricultural fields to industries and villages to cities. Our project aims to contribute to this issue by solving the problem of segregation of waste by using the latest advances in computer vision. The Transformers for Image Recognition at Scale, which will give highly efficient results in less time compared to the models based on Convolutional Neural Networks which loses a lot of valuable information and ignores the relationship between part of images and as a whole. The Self-Attention in vision transformers gives them the capability to understand the connection between inputs, where operations can be processed in parallel on multiple GPUs and CPUs, which cannot be achieved in the case of Convolutional Neural Networks. This project allows us to build an effective system that can classify waste in real-time without human intervention in large scale industries and also in a regular household.