A Mask R-CNN Algorithm for Automated Segmentation of Asphalt Road Cracks

Bakhytzhan Kulambayev, Olzhas Olzhayev

2025

Full text Open access: Gold Metadata (CSL-JSON)
DOI 10.1016/j.procs.2025.08.257
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

Automated segmentation of pavement distress is critical for efficient maintenance and asset management of road networks. This study presents a Mask R-CNN–based framework for pixel-level detection and delineation of four distress categories longitudinal cracks, transverse cracks, alligator cracks, and potholes using the publicly available RDD2020 dataset. The proposed architecture integrates a ResNet-101 backbone for multi-scale feature extraction, ROI Align for high-fidelity region pooling, and specialized mask and bounding-box heads for simultaneous classification and regression. The model was trained for 100 epochs with extensive data augmentation and achieved a mean Intersection over Union (mIoU) of 0.78 and a mean Average Precision (mAP) of 0.85 across all classes on a held-out test split. Notably, pothole segmentation yielded an IoU above 0.88, while fragmented alligator cracks attained an IoU of 0.69, reflecting the challenges of fine-grained crack morphology. Rapid convergence of training and validation curves indicates strong generalization with minimal overfitting, even under varying lighting, surface textures, and environmental conditions. The end-to-end mask prediction enables precise quantification of damage extents, facilitating automated road health mapping and prioritization of repairs. Future enhancements will explore attention mechanisms for micro-crack sensitivity, semi-supervised learning to leverage unlabeled imagery, and fusion with three-dimensional sensing modalities for improved robustness in complex scenarios.