Segregating_Household_Waste

(The code repository for my Masters project)

Owing to the drastic increase in solid waste and garbage generated by humans, waste management, segregation and recycling have become problems of scale. This warrants an automated solution, rather than relying on manual workforce, which will help in increasing life quality and give more control over the full waste generation chain from the source to the dump.

Project Proposal:

The project proposal submitted to Dublin City University is linked here: PDF file

Libraries/Packages used:

• numpy
• pandas
• matplotlib
• seaborn
• PIL
• tensorflow-keras
• scikit-learn
• os
• glob
• shutils
• flask

Data Sources:

For Version-1 (v001): Used the Waste Classification data from Kaggle. This dataset contains 22500 images of Organic and Recyclable items.

Class Distribution for Waste Classification dataset - Version 1

For Version-2 (v002): Used the trashnet dataset from GitHub. The dataset spans six classes: glass, paper, cardboard, plastic, metal, and trash. Currently, the dataset consists of 2527 images (501 'glass', 594 'paper', 403 'cardboard', 482 'plastic', 410 'metal', 137 'trash').

Class Distribution for trashnet dataset - Version 2

For Version-3 (v003): Used the same trashnet dataset as in Version-2 and manually removed the 137 images belonging to the 'trash' category. Now, the dataset has 2390 images (501 'glass', 594 'paper', 403 'cardboard', 482 'plastic', 410 'metal').

Class Distribution for trashnet dataset - Version 3

Final Results:

V003 Resnet Model - Accuracy and Loss VS Number of Epochs - Experiment 5

V003 Resnet Model - Confusion Matrix - Experiment 5

V003 Resnet Model - Classification Report - Experiment 5

V003 Resnet Model - Testing using custom images - Experiment 5

Flask App Deployment - Results - Experiment 6

To-be-done:

• Attaching the Thesis report after final submission and results.