YOLOv9: A High-Performance Deep Learning Approach for Asphalt Pavement Distresses Detection in Roadway Images
Roads are crucial infrastructure for connectivity and transportation. Various asphalt pavement distresses pose safety risks to road users and vehicles. Early detection of these distresses is crucial for road safety and efficiency. This study proposes employing the deep learning algorithm YOLOv9, the...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Universitas Riau
2025-06-01
|
| Series: | International Journal of Electrical, Energy and Power System Engineering |
| Subjects: | |
| Online Access: | https://ijeepse.id/journal/index.php/ijeepse/article/view/239 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Roads are crucial infrastructure for connectivity and transportation. Various asphalt pavement distresses pose safety risks to road users and vehicles. Early detection of these distresses is crucial for road safety and efficiency. This study proposes employing the deep learning algorithm YOLOv9, the latest You Only Look Once (YOLO) version, for identifying asphalt pavement distresses. YOLOv9 excels in efficiency, accuracy, and computational speed compared to its predecessors. The trained model exhibited outstanding performance in identifying different types of pavement distresses, especially potholes, rutting, and alligator cracking, achieving a precision of 1.0 (at a confidence threshold of 0.969), a recall of 0.9 (at a threshold of 0.0), and a mAP of 0.796. It also demonstrated swift image processing at 2.7 ms per image. While the model struggles with accurately identifying longitudinal cracking at higher confidence levels, its effectiveness in visualizing distress detection results on different pavement types is commendable. This YOLOv9-based system has promising potential for integration into road maintenance and monitoring protocols, enabling prompt identification and rectification of road issues to enhance safety and convenience for road users. Further enhancements are necessary to refine longitudinal cracking detection and evaluate the model across a broader spectrum of road conditions. |
|---|---|
| ISSN: | 2654-4644 |