Early surface crack detection and localization in structures: an artificial intelligence approach

• Published in: Discover applied sciences Vol. 7; no. 9; pp. 949 - 25
• Main Authors: Yogi, Biswarup, Das, Sourav Kumar, Modak, Soham, Biswas, Aritra, Roy, Satyabrata
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.09.2025; Springer Nature B.V; Springer
• Subjects: Accuracy; Applied and Technical Physics; Artificial intelligence; Artificial neural networks; Automation; Chemistry/Food Science; Classification; CNN; Cracks; Datasets; Deep learning; Drones; Earth Sciences; Earthquakes; Efficiency; Engineering; Environment; High rise buildings; Human error; Image quality; Inspection; Inspections; Localization; Materials Science; Methods; Neural networks; Object detection; Object recognition; Preventive maintenance; Recall; Roof crack detection; Roofing; Structure protection; Surface cracks; Unmanned aerial vehicles; Urban structure; YOLOv5; Deep learning; CNN; Urban structure; YOLOv5; Roof crack detection; Structure protection; Object detection; Real-time localization
• ISSN: 3004-9261; 2523-3963; 3004-9261; 2523-3971
• DOI: 10.1007/s42452-025-07562-5

Detecting cracks on the roof is crucial for protecting buildings. Traditional inspection methods are slow, risky and require a lot of manual effort. These problems are worse for tall or sloped roofs. This paper presents an automatic method for detecting and locating roof cracks. It uses Convolutional Neural Networks (CNNs) and YOLOv5. CNNs help in extracting features from the images to identify cracks. YOLOv5 is used to quickly and accurately locate cracks. The system uses high-quality images taken by drones. These drone images make the process faster, safer, and more complete. The model was evaluated using standard metrics, including Accuracy, precision, recall, and F1 score. It achieved 91.30% accuracy, 95.05% precision, 96.05% recall, and 95.84% F1 score. The results were also checked using the Receiver Operating Characteristic Curve (ROC curves), histograms, and confusion matrices. This method is more effective than traditional inspection methods. Improves Accuracy, saves time, and increases safety in roof maintenance.