Waste Classification with VGG16

This project aims to classify waste images into two categories: Organic and Recycle waste using a pre-trained VGG16 model fine-tuned for the task.

Table of Contents

• Introduction
• Problem Statement
• Model Overview
• Dataset
• Results

Introduction

Waste Classification is a computer vision project that leverages a VGG16-based deep learning model to classify waste images. The project focuses on distinguishing between Organic and Recycle waste, making it a valuable tool for waste management and environmental initiatives.

Problem Statement

The waste classification problem is a critical aspect of waste management. Properly identifying and classifying waste materials can help in recycling and disposal processes, reducing environmental impact.

Model Overview

This project utilizes a fine-tuned VGG16-based CNN model to classify waste materials. The model architecture includes custom dense layers with batch normalization, ReLU activation, and dropout layers to improve performance. Here's a brief overview of the model architecture:

• VGG16-based feature extraction layer (pre-trained on ImageNet)
• Dropout layer with a 20% dropout rate
• Flattening layer
• Batch normalization layers
• Dense layers with 1024, 512, 256, 512, and 512 filters, respectively
• Activation layers (ReLU) The final output layer is a single neuron with a sigmoid activation function for binary classification. You can access the last Traiend Weight on Drive

Dataset

Problem: Waste management challenges, including landfill overflow and pollution. Approach: Analyzed waste components, segregated into Organic and Recyclable using IoT and ML. Implementation: Dataset split - 85% training (22,564 images) and 15% testing (2,513 images). You can access the complete dataset on Kaggle.

Results

The model achieved impressive accuracy in classifying waste images. Training Performance

• Loss: 0.1450
• AUC (Area Under the Curve): 0.9865

Validation Performance

• Loss: 0.3357
• AUC (Area Under the Curve): 0.9517

Test Performance

• Loss: 0.2976
• AUC (Area Under the Curve): 0.9559

Interpretation

• The training results show a low loss and high AUC, indicating that the model learned well from the training data.
• The validation performance, while slightly lower than training, still shows good model generalization.
• The test results, with a loss of 0.2976 and AUC of 0.9559, demonstrate the model's ability to perform well on new, unseen data.

These results suggest that the model is effective in classifying waste materials, reducing landfill waste and potential environmental issues.