EdgePCB YOLOv8 Defect Detection Model
Overview
This repository contains a trained YOLOv8-based object detection model for automated detection of defects in Printed Circuit Boards (PCBs). The model is part of the EdgePCB-AI project, designed for real-time industrial inspection using edge devices such as the NVIDIA Jetson Orin.
The model is optimized for detecting multiple defect types in PCB images with high accuracy and real-time performance.
Model Details
- Model Architecture: YOLOv8 (Ultralytics)
- Task: Object Detection
- Classes: 6 PCB defect categories
- Input Size: 640×640
- Framework: PyTorch
- Optimization: TensorRT (FP16 / INT8 for edge deployment)
Classes
The model detects the following PCB defects:
- Missing Hole
- Mouse Bite
- Open Circuit
- Short Circuit
- Spur
- Spurious Copper
Training Details
- Dataset Size: ~28,000 images
- Annotation Format: YOLO
- Training Platform: Google Colab (NVIDIA Tesla T4 GPU)
- Augmentations: Mosaic, flipping, scaling, brightness adjustments
- Epochs: ~30–80
Performance
The model achieves high detection accuracy while maintaining real-time inference capability.
Deployment
The model is deployed on:
- Edge Device: NVIDIA Jetson Orin 8GB
- Inference Engine: TensorRT
- Framework: NVIDIA DeepStream
- Performance: ~25–30 FPS
Usage
Install Dependencies
pip install ultralytics opencv-python
Inference Example
from ultralytics import YOLO
model = YOLO("best.pt")
results = model("image.jpg", show=True)
Applications
- Automated PCB inspection
- Industrial quality control
- Edge AI-based manufacturing systems
- Real-time defect detection pipelines
Limitations
- Performance may degrade on extremely small or low-contrast defects
- Sensitive to drastic lighting variations
- Requires diverse dataset for better generalization
License
Specify your license (MIT / Apache 2.0 / Custom)
Author
Tanish Jain B.Tech Electronics & Telecommunications Engineering VIIT Pune
Citation
If you use this model in your work, please cite:
EdgePCB-AI: Real-Time PCB Defect Detection Using Embedded Vision