🧠 Surface Defect Detection on Hot-Rolled Steel Strips

This project is a Streamlit web application for detecting surface defects in hot-rolled steel strips using a Convolutional Neural Network (CNN). It provides features such as user login, prediction history with SQLite, and PDF report export functionality.

---

📁 Project Structure

Steel_Defect_Streamlit_final/
│
├── app.py                      # Main Streamlit app
├── model.py                    # CNN model definition
├── database.py                 # SQLite DB functions (users, history)
├── auth.py                     # User authentication logic
├── pdf_export.py               # PDF report generation utility
├── utils.py                    # Preprocessing & helper functions
├── requirements.txt            # Required Python libraries
├── users.db                    # SQLite database for users and history
├── data/
│   └── sample_images/          # Test images for prediction

---

🚀 Features

• 🔍 Surface Defect Detection using deep learning (PyTorch CNN)
• 🔐 User Authentication (Sign up and log in)
• 🧠 Real-time Predictions on uploaded images
• 🗂️ Prediction History stored per user in SQLite
• 📄 Export to PDF: Generate downloadable reports

---

🛠️ Installation

1. Clone or Extract the Project

If you're using the zip file:

unzip Steel_Defect_Streamlit.zip
cd Steel_Defect_Streamlit

2. Set up Virtual Environment (Optional)

python -m venv venv
source venv/bin/activate      # On Windows: venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

---

▶️ Running the App

Use the following command to launch the Streamlit web application:

streamlit run app.py

The app will open in your browser at http://localhost:8501.

---

🖼️ How to Use

1. Login or Sign Up with your credentials.
2. Upload an image OR zip file of a steel strip.
3. The model will detect and classify the defect (if any).
4. You can view your prediction history from the sidebar.
5. Export results as a PDF for record-keeping.

---

🧪 Model Info

• CNN implemented with PyTorch
• Input: Steel surface image (PNG/JPG/JPEG/BMP)
• Output: Defect class label with prediction probability

---

📦 Dependencies

Major dependencies include:

• tensorflow
• protobuf
• numpy
• pandas
• opencv-python
• scikit-learn
• matplotlib
• plotly
• Pillow
• streamlit
• validators
• gitpython
• altair
• typing-extensions
• requests
• watchdog
• blinker
• tornado
• fpdf

See requirements.txt for the complete list.

---

🧰 Notes

• The model weights should be loaded within model.py. Ensure model.pth is placed correctly if loading external weights.
• The users.db SQLite database stores user credentials and prediction history.

---

🧑‍💻 Authors

Developed by Shesh Kanade Harsh Prajapati Yashkumar Desai Rutvik Sindhav For academic purposes and AI-based defect detection research.

---

📄 License

This project is licensed for educational and research use only.