🔧 PCB Defect Detection and Classification System

📌 Overview

This project presents an automated PCB Defect Detection and Classification System that combines image processing and deep learning to identify, localize, and classify defects in Printed Circuit Boards (PCBs).

The system compares a defect-free template PCB with a test PCB, extracts defect regions, and classifies them using a trained EfficientNet model. A Streamlit web application allows users to upload images, visualize results in real time, and download outputs.

---

🎯 Objectives

The system aims to:

• Detect and localize PCB defects using image subtraction and contour analysis
• Classify defects into predefined categories using a CNN model
• Provide a user-friendly web interface for real-time inference
• Generate annotated images and downloadable logs
• Deliver a complete, deployable end-to-end solution

---

📂 Dataset

• Dataset Used: DeepPCB Dataset

• Contains paired images of:

  • Template PCB (defect-free)
  • Test PCB (with defects)

---

🧱 Project Structure

PCB_Project/
│
├── Milestone1/
│   ├── data/
│   ├── outputs/
│   └── scripts/
│
├── Milestone2/
│   ├── data/
│   ├── models/
│   ├── results/
│   └── training scripts
│
├── Milestone3/
│   ├── app.py
│   ├── inference.py
│   ├── utils.py
│   ├── models/
│   └── requirements.txt
│
└── README.md

---

✅ Milestone Summary

Milestone 1 — Image Processing

• Image alignment and preprocessing
• Template-based image subtraction
• Thresholding and morphological filtering
• Defect mask generation
• Contour detection and ROI extraction

Milestone 2 — Model Training & Evaluation

• Trained EfficientNet-B0 model in PyTorch

• Image size: 128×128

• Optimizer: Adam

• Loss: Cross-Entropy / Focal Loss

• Achieved ~97.9% test accuracy

• Generated:

  • Accuracy plot
  • Loss plot
  • Confusion matrix
  • Classification report

Milestone 3 — Web UI + Backend Integration

• Streamlit-based web application

• Upload template and test images

• Real-time defect detection

• Annotated PCB output with bounding boxes and labels

• Modular backend with:

  • utils.py → image processing
  • inference.py → model prediction

Milestone 4 — Finalization & Deployment

• Downloadable:

  • Annotated image
  • Prediction log (CSV)

• Fully functional Streamlit app

• GitHub repository organized

• Ready for presentation and demo

---

🧠 Defect Classes

The model classifies defects into six categories:

• Missing Hole
• Mouse Bite
• Open Circuit
• Short
• Spur
• Spurious Copper

---

🚀 How to Run Locally

1️⃣ Clone the repository

git clone https://github.com/balaganeshbalireddi/Pcb_Project.git
cd Pcb_Project/Milestone3

2️⃣ Install dependencies

pip install -r requirements.txt

3️⃣ Run the Streamlit app

streamlit run app.py

Then open in your browser:

http://localhost:8501

---

🌐 Deployment

The app is designed to be deployed on Streamlit Cloud. Use Milestone3/app.py as the entry point and ensure requirements.txt includes:

streamlit
torch
torchvision
opencv-python-headless
numpy
pandas
pillow

---

📈 Performance

• Test Accuracy: 97.91%
• Fast inference: typically < 3 seconds per image pair
• Low false positives and false negatives

---

📌 Future Enhancements

• Add defect severity scoring
• Support batch uploads
• Real-time camera integration
• Edge deployment on embedded systems