DETECTION OF SURFACE DEFECTS IN PRINTED CIRCUIT BOARDS USING YOLOv8

An end-to-end YOLOv8-based system for detecting six types of PCB surface defects. It includes dataset conversion, automated train/val split, model training, and real-time inference, offering a fast and accurate solution for PCB quality inspection.

---

📋 Table of Contents

• Overview
• Dataset
• Project Structure
• Model Details
• Results
• License
• Acknowledgements

---

🚀 Overview

This project automates the detection of six common PCB surface defects using the YOLOv8 object detection model.
It includes:

• Data extraction and formatting
• XML to YOLO annotation conversion
• Dataset splitting (train/val)
• YOLOv8 model training
• Inference and result visualization

---

📦 Dataset

• Format: Pascal VOC (XML annotations + images)
• Size: ~1386 images
• Classes:
  • Missing Hole
  • Mouse Bite
  • Open Circuit
  • Short
  • Spur
  • Spurious Copper

---

🏗️ Project Structure

.
├── pcbarchive.zip         # Original dataset zip
├── dataset/               # Processed dataset in YOLO format
│   ├── images/
│   │   ├── train/
│   │   └── val/
│   ├── labels/
│   │   ├── train/
│   │   └── val/
├── pcb.yaml               # YOLOv8 configuration file
├── runs/                  # Training runs and model weights
├── test.jpg               # Sample test image
├── result.jpg             # Output image after detection
└── main.py                # Full project script

---

🧠 Model Details

• Model Used: YOLOv8n (Nano Version)
• Epochs: 20
• Image Size: 640x640
• Optimization: Automatically handles missing labels and skipped annotations.

---

📊 Results

• High accuracy in detecting all six types of defects.
• Real-time detection enabled.
• Lightweight model suitable for deployment on edge devices.

---

📄 License

• TBA

---

🙏 Acknowledgements

• Ultralytics YOLOv8
• Kaggle PCB Dataset

---

👨‍💻 Author

• Name: ABHINAV K R
• Contact: abhinavnowkr@gmail.com
• LinkedIn: abhinavkravi