A lightweight YOLO11n seg framework for real time surface crack detection with segmentation.

Area of Science:

• Civil Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Superficial crack recognition is crucial for infrastructure safety and longevity.
• Traditional methods rely on manual inspection and classical image processing.
• Deep learning offers automated solutions for crack detection.

Purpose of the Study:

• To propose a deep learning framework for automated surface crack detection.
• To evaluate the YOLO11n-seg architecture for crack recognition tasks.
• To establish a benchmark for edge-deployable crack recognition systems.

Main Methods:

• Utilized the Crack-Seg dataset for training and evaluation.
• Implemented the YOLO11n-seg deep learning architecture.
• Assessed model performance using precision, Box mAP@50, and Mask mAP@50 metrics.

Main Results:

• The YOLO11n-seg model achieved a Precision of 78.8% and Box mAP@50 of 76.2%.
• The lightweight model (2.8 million parameters) demonstrated a Mask mAP@50 of 58.7%.
• Achieved an ultra-fast inference rate of 3.6ms per image on Tesla T4.

Conclusions:

• The YOLO11n-seg architecture provides strong performance for surface crack detection.
• This lightweight model offers acceptable segmentation performance with lower computational cost.
• The findings establish a new benchmark for efficient, edge-deployable crack recognition.