A dataset of building surface defects collected by UAVs for machine learning-based detection.

Area of Science:

• Civil Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Traditional infrastructure inspection is labor-intensive and inefficient.
• Lack of large, high-quality datasets hinders deep learning for infrastructure defect detection.

Purpose of the Study:

• To create a comprehensive, diverse, and publicly accessible UAV-based dataset for building surface defect analysis.
• To facilitate advancements in automated visual assessment and multi-task learning for infrastructure maintenance.

Main Methods:

• Collected 14,471 high-resolution UAV images of building surfaces.
• Captured data across six structural types and five defect categories (cracks, abscission, leakage, corrosion, bulging).
• Annotated images with bounding boxes and divided into training, validation, and testing sets.

Main Results:

• Developed a large-scale, diverse dataset of building surface defects.
• Dataset covers various conditions, including different environments and illumination.
• Provides a benchmark for defect detection, segmentation, and automated assessment.

Conclusions:

• The created dataset addresses the need for high-quality data in deep learning for infrastructure inspection.
• It serves as a valuable resource for developing and evaluating automated visual assessment tools.
• Enables multi-task research for improved civil infrastructure maintenance and management.