PCB Defect Detection

A YOLOv8-based Computer Vision solution for automated detection and classification of defects on Printed Circuit Boards (PCBs), with a desktop application built with Tauri.

Features

• Detection: Locate defects with bounding boxes
• Classification: Identify 6 types of PCB defects
• Real-time: Fast inference with YOLOv8
• Desktop App: Cross-platform application with Tauri
• Export: ONNX format for deployment

Defect Classes

ID	Defect	Description
0	missing_hole	Missing drill hole
1	mouse_bite	Irregular edge
2	open_circuit	Broken trace
3	short	Short circuit
4	spur	Copper protrusion
5	spurious_copper	Unwanted copper

Résultats

Métrique	Valeur	Description
Précision de détection	95.5%	Le modèle localise correctement 95.5% des défauts
Précision stricte	51.8%	Précision avec des critères plus stricts
Fiabilité	95.4%	95.4% des détections sont de vrais défauts
Taux de détection	91.0%	Le modèle trouve 91% de tous les défauts

Résultats après 100 époques d'entraînement sur Kaggle avec YOLOv8s.

Quick Start

Training on Kaggle (Recommended)

Prerequisites:

1. Add dataset akhatova/pcb-defects via "+ Add Input"
2. Enable GPU T4 in notebook settings

!pip install ultralytics -q
!rm -rf /kaggle/working/pcb-defect-detector
!git clone https://github.com/alainpaluku/pcb-defect-detector.git
%cd /kaggle/working/pcb-defect-detector
!python run_kaggle.py

Local Training

git clone https://github.com/alainpaluku/pcb-defect-detector.git
cd pcb-defect-detector
pip install -r requirements.txt

# Train
python main.py train --epochs 100

# Detect
python main.py detect path/to/image.jpg --save

# Export to ONNX
python main.py export --model output/pcb_model.pt --format onnx

Desktop Application

A cross-platform desktop app built with Tauri 2 and React for real-time PCB inspection.

Features

• Drag & drop image upload
• Real-time defect detection with bounding boxes
• Classification with confidence scores
• Dark/Light mode
• History of inspections

Build from Source

Prerequisites:

• Node.js 18+
• Rust 1.70+
• System dependencies (see below)

Fedora:

sudo dnf install webkit2gtk4.1-devel javascriptcoregtk4.1-devel gtk3-devel glib2-devel openssl-devel libsoup-devel librsvg2-devel

Ubuntu/Debian:

sudo apt install libwebkit2gtk-4.1-dev libgtk-3-dev libayatana-appindicator3-dev librsvg2-dev

Build:

cd desktop-app
npm install
npm run tauri build

The built application will be in:

• Linux: src-tauri/target/release/bundle/deb/ or rpm/
• Windows: src-tauri/target/release/bundle/msi/
• macOS: src-tauri/target/release/bundle/dmg/

Using the Trained Model

1. Train on Kaggle and download best.pt or pcb_model.onnx
2. Place the ONNX model in desktop-app/public/model/pcb_detector.onnx
3. Build and run the app

Project Structure

pcb-defect-detector/
├── src/                    # Python training code
│   ├── config.py           # Configuration
│   ├── data_ingestion.py   # Data loading & conversion
│   ├── model.py            # YOLOv8 model wrapper
│   ├── detector.py         # Inference interface
│   ├── trainer.py          # Training pipeline
│   └── utils.py            # Utilities
├── desktop-app/            # Tauri desktop application
│   ├── src/                # React frontend
│   ├── src-tauri/          # Rust backend
│   └── public/model/       # ONNX model location
├── run_kaggle.py           # Kaggle runner
└── requirements.txt

Model Architecture

Property	Value
Base Model	YOLOv8s
Input Shape	(1, 3, 640, 640)
Output	Bounding boxes + 6 classes
ONNX Size	~22 MB

Dataset

PCB Defects - Akhatova

• 1386 images with XML annotations
• 6 defect classes
• VOC format bounding boxes

Author

Alain Paluku - @alainpaluku

License

MIT License