Waste Image Classifier with TensorFlow

This repository contains a Computer Vision project to classify waste images into "Recyclable" or "Household Waste." It employs transfer learning with MobileNetV2 and a custom-built pipeline for preprocessing, training, evaluation, and deployment.

---

Table of Contents

1. Overview
2. Dataset
3. Exploratory Data Analysis (EDA)
4. Model Development
5. Training and Evaluation
6. Deployment
7. Usage
8. Files in Repository
9. Future Work

---

Overview

This project aims to automate waste sorting by classifying images into two categories:

• Recyclable
• Household Waste

Key highlights:

• Comprehensive EDA for dataset understanding.
• Transfer learning with MobileNetV2.
• Deployment using Streamlit for user-friendly interaction.

---

Dataset

The Recycling dataset consists of images for machine learning classification, that were divided into two categories:

• Recyclable: Aluminum cans, glass bottles, etc.
• Household Waste: Food waste, plastic straws, etc.

Dataset Organization

The dataset was split into:

• Training: 70%
• Validation: 15%
• Test: 15%

Images were preprocessed with unique names to avoid duplication and ensure consistency.

---

Exploratory Data Analysis (EDA)

• Conducted using custom scripts in eda_utils.py.
• Key analyses include:
  • Image size and format.
  • Dataset distribution by categories.

---

Model Development

Baseline Model

A custom Convolutional Neural Network (CNN) was built to establish a baseline for classification performance.

Transfer Learning with MobileNetV2

• Architecture: MobileNetV2 pre-trained on ImageNet.

---

Training and Evaluation

The models were trained and evaluated with:

• Metrics: Accuracy, loss, precision, recall, and F1-score.
• Regularization: learning rate scheduling.
• Callbacks: Early stopping and model checkpoints.

---

Deployment

The model is deployed using Streamlit:

• Interface: Upload an image and classify it as "Recyclable" or "Household Waste."
• Backend: Uses a saved Keras model (best_mobilenet_model.keras) for inference.

How to Run

# Install dependencies
pip install -r requirements.txt

# Run the app
streamlit run app.py

---

Usage

Training

Run the notebook MyTemplate.ipynb to train and evaluate models:

• Edit paths to match your dataset structure.
• Experiment with different augmentation and transfer learning techniques.

Prediction

Upload an image to the Streamlit app and receive a prediction with confidence.

---

Files in Repository

File	Description
requirements.txt	List of Python dependencies required for the project..
RecyclingClassifier.ipynb	Notebook for experimenting with the extended dataset.
eda_utils.py	Utility functions for EDA and dataset analysis.
utils.py	Helper functions for organizing images and assigning unique names.
app.py	Streamlit application for deploying the classification model.
best_mobilenet_model.keras	Pre-trained MobileNetV2 model saved after fine-tuning.
RecyclingClassification.pdf	Presentation detailing the project details and conclusions.

---

Future Work

• Model Optimization: Test alternative architectures like ResNet or EfficientNet.
• Augmentation: Applied transformations like rotation, zoom, and flipping to improve generalization.
• Fine-tuning: Enabled training on specific layers to adapt MobileNetV2 to the dataset.
• Data Expansion: Incorporate more waste categories and diverse images.
• Deployment: Extend deployment to cloud platforms (e.g., AWS, Google Cloud).