Heritage cracks dataset: Superficial damage

Area of Science:

• * Civil Engineering and Computer Science
• * Cultural Heritage Preservation
• * Artificial Intelligence and Machine Learning

Background:

• * Heritage buildings face deterioration from various environmental and material factors.
• * Accurate crack detection is crucial for preventive conservation and structural health monitoring.
• * Existing datasets may not fully represent real-world conditions for heritage structures.

Purpose of the Study:

• * Introduce the Heritage Cracks dataset for AI-driven cultural heritage inspection.
• * Support the development of object detection and computer vision models for damage assessment.
• * Facilitate research on non-invasive crack detection in historic infrastructure.

Main Methods:

• * Collected 1,716 high-resolution images of ignimbrite masonry and lime-based plaster cracks.
• * Acquired 686 images in field conditions using smartphone cameras.
• * Generated 1,030 images via data augmentation and manually annotated all images.

Main Results:

• * The dataset includes preprocessed, resized (640x640 pixels), and annotated images.
• * Images document cracks from weathering, degradation, environmental exposure, and other causes.
• * Dataset is structured for direct use in deep learning object detection models (e.g., YOLO).

Conclusions:

• * The Heritage Cracks dataset aids in developing and evaluating AI tools for heritage building inspection.
• * It enables advancements in non-invasive crack detection, preventive conservation, and structural health monitoring.
• * Valuable resource for researchers and specialists working with vulnerable historic infrastructure.