UAV-Based Structural Health Monitoring Using a Two-Stage CNN Model with Lighthouse Localization in GNSS-Denied Environments
This study presents a UAV-based Structural Health Monitoring (SHM) system that combines Lighthouse localization with a two-stage CNN architecture—AlexNet for crack classification and YOLOv4 for segmentation—to enable reliable crack detection and spatial mapping in GNSS-denied environments. This stu...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Ital Publication
2025-06-01
|
| Series: | HighTech and Innovation Journal |
| Subjects: | |
| Online Access: | https://hightechjournal.org/index.php/HIJ/article/view/1100 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This study presents a UAV-based Structural Health Monitoring (SHM) system that combines Lighthouse localization with a two-stage CNN architecture—AlexNet for crack classification and YOLOv4 for segmentation—to enable reliable crack detection and spatial mapping in GNSS-denied environments. This study explores the effectiveness of this combination as a practical and computationally efficient solution for indoor SHM tasks. The UAV was deployed within a 1.5 m × 1.2 m × 1.2 m test volume to inspect synthetic cracks derived from Özgenel’s dataset, as well as a real-world wall crack. Two experiments were conducted: evaluating UAV localization accuracy and assessing the system’s ability to detect cracks and provide corresponding pose data. The system achieved a 1–2 cm margin of error in pose estimation, alongside 100% precision, 83.33% recall, and 91.89% accuracy in crack detection. This level of localization accuracy supports stable autonomous UAV flight and ensures that cracks are detected and spatially localized with minimal deviation. Beyond classification and segmentation, the system returns pose data tied to each detected crack, allowing users to identify defect locations precisely and use this information to guide inspection or maintenance tasks. Future work includes expanding the dataset, generalization, and evaluating scalability via multi-base station setups.
|
|---|---|
| ISSN: | 2723-9535 |