Metal Scrap Classification and Detection using Deep Learning

📌 Overview

This project uses Convolutional Neural Networks (CNNs) to classify metallic scrap images into four categories — Aluminium, Copper, Steel, and Brass — based on shape, color, and other visual features. With proper preprocessing, data augmentation, and training strategies, the model achieved 96.5% accuracy on real-world test data.

🛠️ Technologies Used

• Python 3.x
• TensorFlow / Keras
• Matplotlib, NumPy

---

🧪 Performance

Class	Precision	Recall	F1-Score
Aluminium	0.97	0.96	0.96
Copper	0.95	0.94	0.95
Steel	0.96	0.97	0.96
Brass	0.95	0.94	0.95

• Overall Accuracy: 96.51%
• Evaluated using Confusion Matrix and Real-World Test Images

---

🌍 Real-World Impact

• Automates scrap sorting in industrial scrapyards
• Enables vision-based inspection in recycling plants
• Contributes to AI-driven circular economy solutions

---

🚀 Future Developments

• 🔍 Train and fine-tune YOLOv8 for object detection with bounding boxes
• 🧠 Integrate attention mechanisms to handle cluttered and overlapping scrap regions
• 🎥 Enable real-time detection from video streams
• 🧪 Use a hybrid dataset (synthetic + real) to improve generalizability