ReClassify: Trash Classification with EfficientNet + SVM

An automated trash classification system using EfficientNetB0 and a fine-tuned Support Vector Machine (SVM).
ReClassify identifies recyclable materials across six categories to assist with smarter, cleaner waste sorting.

---

Model Overview

• Feature Extractor: EfficientNetB0 (pretrained on ImageNet)
• Classifier: Support Vector Machine (RBF kernel)
• Tuning: GridSearchCV for C, gamma, kernel
• Accuracy: 91% on TrashNet dataset

---

Dataset

ReClassify uses the TrashNet dataset, containing:

• 2,500+ labeled images
• 6 classes:
  cardboard, glass, metal, paper, plastic, trash
• Images resized to 224×224 and preprocessed for EfficientNet

---

How It Works

1. Load and preprocess images
2. Extract feature vectors with EfficientNetB0
3. Normalize features using StandardScaler
4. Train an SVM classifier with the best hyperparameters
5. Evaluate accuracy, F1-score, and generate predictions

---

Results

Class	Precision	Recall	F1-score
Cardboard	0.96	0.94	0.95
Glass	0.91	0.92	0.91
Metal	0.84	0.91	0.87
Paper	0.90	0.97	0.93
Plastic	0.94	0.85	0.89
Trash	0.89	0.67	0.76

• Overall Accuracy: 91%
• Macro F1 Score: 0.89
• Weighted F1 Score: 0.90

---

Dependencies

Install all required packages:

pip install -r requirements.txt

Main packages:

• tensorflow / keras
• scikit-learn
• opencv-python
• matplotlib
• numpy

---

Run the classifier:

python main.py

This script will:

• Load images from the TrashNet/ dataset
• Extract features using EfficientNetB0
• Train and evaluate the SVM model
• Display classification report and prediction samples

---

Project Structure

├── TrashNet/                # Dataset (6 folders by class)
├── main.py                  # Main training & evaluation script
├── preprocess.py            # Image loading/preprocessing helper
├── requirements.txt
└── README.md

---

Future Improvements

• Add image augmentation to boost low-sample classes like trash
• Try ResNet or EfficientNetV2 for deeper feature extraction
• Deploy as a web app using Streamlit or Flask
• Implement real-time webcam classification

---

Author

Edwin Rivera
GitHub
LinkedIn