Machine learning evaluation demonstrates robust asphalt deterioration detection using grayscale YOLOv8s models, indicating strong potential for real-time road maintenance planning.
Key Points
To evaluate the performance of YOLOv8 object-recognition models in detecting and classifying multiple types of asphalt pavement deterioration under low-color conditions.
Trained and tested YOLOv8 algorithm variants across seven asphalt distress classes: Crack, Patch-Crack, Pothole, Patch-Pothole, Net, Patch-Net, and Manhole.
Processed pavement imagery in grayscale to test model detection capabilities under low-color-information scenarios.
YOLOv8s emerged as the best-performing architecture, reaching an mAP@50 of 0.963 and an mAP@50-95 of 0.780.
Class-specific evaluation showed Patch-Crack achieved exceptionally high recall with the fewest false negatives, whereas the severely under-represented Patch-Net class produced zero true positives.