👁️ Computer Vision Group Project - Waste Classification System

📌 Overview

This project builds an image classification system to classify waste into three categories: Plastic, Paper, and Organic. It compares traditional machine learning and deep learning approaches.

---

👥 Group Members

• SEE CHWAN KAI (242UT2449P)
• KHO WEI CONG (242UT2449Z)
• TEO JING AN (242UT24490)
• Kerk Yi (1211203133)

---

🧠 Models Implemented

1. HOG + SVM

• Feature extraction using HOG
• Classification using SVM
• Tested with and without preprocessing

2. ResNet50

• Pretrained on ImageNet
• Transfer learning with frozen base layers

3. VGG16

• Pretrained CNN model
• Custom classification head

---

⚙️ Dataset

• Input folder: /content/drive/MyDrive/Dataset
• Classes:
  • Plastic
  • Paper
  • Organic

Data Split

• Train: 70%
• Validation: 15%
• Test: 15%

Output structure:

Dataset_Split/
  ├── train/
  ├── val/
  └── test/

---

🧪 Preprocessing

Image Enhancement

• CLAHE
• Gaussian Blur
• Sharpening

Deep Learning Preprocessing

• Normalization (0–1)
• ImageNet preprocessing

---

📏 Evaluation Metrics

• Accuracy
• Precision
• Recall
• F1-Score

---

📊 Results

• Deep learning models (ResNet50, VGG16) outperform HOG + SVM
• Preprocessing improves traditional model performance
• Transfer learning provides strong classification accuracy

---

🚀 How to Run

1. Install dependencies

pip install numpy pandas matplotlib opencv-python tensorflow scikit-learn scikit-image

2. Prepare dataset

Place dataset in:

/content/drive/MyDrive/Dataset

3. Run notebook

jupyter notebook

Open the notebook and run all cells.

---

🔍 Key Insights

• Transfer learning models perform best
• Preprocessing is important for traditional methods
• Deep learning models are more robust

---

🔮 Future Improvements

• Add more waste categories
• Apply data augmentation
• Fine-tune pretrained models
• Use advanced architectures like EfficientNet

---

📚 Conclusion

This project shows that deep learning models significantly outperform traditional methods in image classification tasks, especially when using transfer learning.