ID: 23FE10CSE00078

ScrapNet – Waste Classification System

AI-powered waste image classification using EfficientNet-B0 and PyTorch to automate waste segregation for smarter recycling.

Open Live Demo View GitHub Repo

Problem Statement

Manual waste segregation is slow and inconsistent, leading to mixed waste and lower recycling efficiency. ScrapNet classifies waste images into categories to enable faster, scalable, and consistent sorting.

Literature Review / Market Research

Reviewed transfer-learning CNNs (ResNet/EfficientNet), waste classification pipelines, and lightweight deployment patterns for ML demos.

Research Gap / Innovation

Many solutions are heavy or not reproducible. ScrapNet focuses on a compact EfficientNet-B0 pipeline, portable artifact handling, and a clean Streamlit demo + GitHub Pages presentation.

System Methodology

Dataset / Input

Waste images across 7 classes (cardboard, compost, glass, metal, paper, plastic, trash). Images resized to 224×224 and normalized.

Model / Architecture

EfficientNet-B0 pretrained on ImageNet, classifier head replaced for 7-class prediction, fine-tuned using cross-entropy loss.

Live Execution

VIEW DEMO

RESULTS & ANALYSIS

BEST VALIDATION ACCURACY

~82%

The model generalizes reasonably well, but some classes overlap (notably plastic ↔ trash). Check the confusion matrix + per-class F1 for the detailed story.

Counts Matrix Per-class F1

Academic Credits

Project Guide

Mahesh Jangid

Student

Bhavya Bansal

23FE10CSE00078