🟢 PCB Defect Detection using YOLOv5

Automated detection and localization of PCB manufacturing defects using YOLOv5 and the Akhatova PCB Defects dataset

📌 Overview

Printed Circuit Boards (PCBs) are the backbone of modern electronics — even tiny manufacturing defects can lead to malfunction, costly repairs, or total product failure. Manual inspection is time-consuming and error-prone.

This project implements an automated PCB defect detection system using YOLOv5, a state-of-the-art real-time object detection framework. The model is trained to locate and classify multiple PCB defect types with high accuracy, enabling rapid and reliable inspection in manufacturing environments.

🎯 Key Highlights

• ⚡ Fast & accurate detection using YOLOv5
• ✅ Trained on the Akhatova PCB Defects Dataset
• 📌 Detects common PCB defects: missing hole, mouse bite, open circuit, short, spurious copper, etc.
• 🗂️ Modular pipeline for training and inference
• 🔍 Precise bounding boxes for defect localization
• 🧩 Easy to retrain or extend for other defect types

⚙️ Tech Stack

• Python 3.x
• YOLOv5 (Ultralytics)
• PyTorch
• OpenCV
• NumPy, Pandas, Matplotlib

📂 Project Structure

pcb-defect-detection/ ├── dataset/ # PCB images & labels (YOLO format) ├── runs/ # YOLOv5 training runs & results ├── weights/ # Trained model weights (.pt) ├── train.py # Training script ├── detect.py # Inference script for testing ├── dataset.yaml # YOLOv5 dataset configuration ├── requirements.txt # Project dependencies └── README.md # Project documentation (this file)

🚀 Getting Started

1️⃣ Clone the repository

bash git clone https://github.com/ChiragSingh01/pcb-defect-detection.git cd pcb-defect-detection

2️⃣ Install dependencies

bash pip install -r requirements.txt or manually:

bash pip install torch torchvision torchaudio pip install opencv-python pip install matplotlib pip install pandas

3️⃣ Download the dataset

Download the PCB Defects Dataset and extract it to the dataset/ folder. Ensure your folder structure looks like this:

bash dataset/ ├── images/ │ ├── train/ │ ├── val/ │ ├── test/ ├── labels/ │ ├── train/ │ ├── val/ │ ├── test/

4️⃣ Configure the dataset

Make sure your dataset.yaml file is correctly set up. Example:

```yaml train: dataset/images/train val: dataset/images/val

nc: 6 names: ['missinghole', 'mousebite', 'opencircuit', 'short', 'spuriouscopper', 'other_defects'] ```

5️⃣ Train YOLOv5

bash python train.py --img 640 --batch 16 --epochs 50 --data dataset.yaml --weights yolov5s.pt Adjust image size, batch size, or starting weights as needed.

6️⃣ Run detection

After training, run detection on new images: bash python detect.py --weights runs/train/exp/weights/best.pt --source dataset/images/test Detected images with bounding boxes will be saved in runs/detect/.

Model performance:

• ✔️ mAP: ~95% (update with your actual test results!)
• ✔️ Fast detection speed on standard hardware
• ✔️ Multiple defect categories detected with high precision

📌 Use Cases

• 📏 Automated optical inspection (AOI) in PCB production lines
• ⚡ Real-time defect detection and sorting
• 🧩 Extendable to other electronics manufacturing tasks

🔍 Future Improvements

• 📈 Improve accuracy with more diverse training data & augmentations
• 🗂️ Add support for real-time video stream inspection
• 🧩 Integrate with industrial edge devices (Jetson Nano, Raspberry Pi, etc.)
• 📝 Generate automatic defect reports for quality control

📚 References

• YOLOv5 by Ultralytics
• Akhatova PCB Defects Dataset

👨‍💻 Author

Chirag

⭐ Support

If you find this project useful, please ⭐ star this repo and share it — it helps more developers discover it!