Smart Waste Material Classification System

An end‑to‑end Computer Vision project for automatic recyclable material classification into:

cardboard · glass · metal · paper · plastic · trash + unknown handling

This project combines CNN feature extraction (ResNet) with Machine Learning classifiers (SVM & KNN) to classify waste in real‑time or from image datasets.

---

🧠 System Pipeline

1️⃣ Data Augmentation – balances dataset using rotation, flip, brightness, zoom & noise 2️⃣ CNN Feature Extraction – ResNet extracts meaningful feature vectors 3️⃣ Feature Dataset Saving – stores cnn_features.npy & cnn_labels.npy 4️⃣ Model Training

• KNN → GridSearch tuning + adaptive confidence threshold
• SVM → Scaling + PCA + RBF kernel 5️⃣ Deployment
• Real‑time Webcam Classification
• Batch Folder Classification

---

📁 Project Structure

project/
│
├── Preprocessing/
│ ├── Run_Preprocessing.py # Augmentation + Feature extraction pipeline
│ ├── DataAugmentation.py # Dataset balancing using augmentations
│ └── FeatureExtraction.py # ResNet CNN feature extraction
│
├── TrainModels/
│ ├── KNN.py # Trains and saves KNN classifier
│ └── SVM.py # Trains and saves SVM classifier
│
├── test.py                    # Folder batch prediction tool
├── camera_app.py              # Real‑time + folder classification app
│
├── data/             # Train + augmented images
│   ├── cnn_features.npy
│   └── cnn_labels.npy
│
└── models/
    ├── KNN/
    │   ├── knn_model.pkl
    │   ├── knn_scaler.pkl
    │   ├── knn_label_encoder.pkl
    │   └── knn_threshold.pkl
    └── SVM/
        ├── svm_model.pkl
        ├── svm_scaler.pkl
        ├── svm_pca.pkl
        └── svm_label_encoder.pkl

---

⚙️ Requirements

• Python 3.8+
• PyTorch + TorchVision
• OpenCV
• NumPy
• scikit‑learn
• joblib
• tqdm

Install dependencies:

pip install torch torchvision opencv-python numpy scikit-learn joblib tqdm pillow

---

🚀 Usage

1️⃣ Build Dataset (Augmentation + Feature Extraction)

python Run_Preprocessing.py

This will:

• Balance dataset
• Extract CNN features
• Save feature dataset

---

2️⃣ Train Models

Train KNN Model

python KNN.py

Train SVM Model

python SVM.py

Models will be saved inside /models/.

---

🎥 Real‑Time Classification (Webcam)

python camera_app.py

Select:

1 → Real‑time camera mode
2 → Predict from folder

Press q to exit camera mode.

---

📂 Predict From Folder

Run:

python camera_app.py

Choose option 2 and enter folder path.

or directly using test tool:

python test.py

---

🔍 Model Logic

KNN

• GridSearch Hyperparameter Tuning
• Confidence threshold tuning
• Unknown rejection handling

SVM

• Data Standardization
• PCA Dimensionality Reduction
• RBF Kernel
• Probability output enabled

---

🧪 Classes

cardboard
glass
metal
paper
plastic
trash

Objects below confidence threshold are labeled as unknown.

---

✅ Key Features

• Robust preprocessing & augmentation
• CNN powered feature extraction
• Real‑time intelligent classification