A Deep Learning Approach for Surface Crack Classification and Segmentation in Unmanned Aerial Vehicle Assisted...

Area of Science:

• Civil Engineering
• Computer Vision
• Robotics

Background:

• Surface crack detection is crucial for infrastructure health monitoring.
• Traditional inspection methods are time-consuming and labor-intensive.
• Accessing confined spaces in infrastructure poses significant challenges.

Purpose of the Study:

• To develop a rapid and reliable system for surface crack detection in infrastructure.
• To evaluate the effectiveness of a novel convolutional neural network (CNN) architecture, CrackClassCNN, for crack classification.
• To assess the performance of the Segment Anything Model (SAM) for crack segmentation.

Main Methods:

• Deployment of two unmanned aerial vehicles (UAVs) for simultaneous image acquisition.
• Utilizing a binary classification CNN (CrackClassCNN) for crack identification in images.
• Employing the Segment Anything Model (SAM) for segmenting detected crack areas.
• Benchmarking CrackClassCNN against state-of-the-art CNN architectures and SAM against manual annotations.

Main Results:

• The novel CrackClassCNN achieved a classification accuracy of 95.02%.
• The Segment Anything Model (SAM) demonstrated strong performance in crack segmentation with a mean IoU of 0.778 and an F1 score of 0.735.
• The integrated UAV and CNN system proved highly effective for efficient infrastructure inspection.

Conclusions:

• The developed UAV platform and CrackClassCNN offer a transformative solution for rapid and reliable surface crack detection.
• The system is particularly suitable for inspecting infrastructure in confined spaces.
• The combination of advanced UAV technology and AI-driven image analysis significantly enhances infrastructure health surveys.