Assessment of YOLO-Based Deep Learning Models for Pavement Crack Segmentation and Crack Extent Estimation

Shafabakhsh, Gholamali; Jabbari, Sina

doi:10.22075/jrce.2026.2706

Assessment of YOLO-Based Deep Learning Models for Pavement Crack Segmentation and Crack Extent Estimation

Articles in Press

Document Type : Regular Paper

Authors

Gholamali Shafabakhsh ¹

Sina Jabbari ²

¹ Professor, Faculty of Civil Engineering, Semnan University, Semnan, Iran

² Ph.D. Candidate, Faculty of Civil Engineering, Semnan University, Semnan, Iran

10.22075/jrce.2026.2706

Abstract

Pavement crack segmentation is a critical task for improving road safety, supporting pavement management systems, and prioritizing maintenance actions in limited-data conditions. This study conducts a comprehensive evaluation of six YOLO-based instance segmentation models, comprising YOLO26 and YOLO11 variants (n, s, m), for pixel-level pavement crack segmentation using the CRACK500 dataset. The dataset, containing 500 pavement crack images with one crack class, was prepared in YOLO segmentation format and split into 350 training images, 75 validation images, and 75 independent test images. All models were trained under a unified experimental configuration. Performance was assessed using Mask mAP@50-95, Mask mAP@50, Precision, Recall, and F1-score. Results showed that YOLO26m-seg achieved the highest validation Mask mAP@50-95 of 0.2972 and was selected for testing. On the test set, it obtained a Mask mAP@50 of 0.5003 and a Mask mAP@50-95 of 0.1809. Increasing the input size to 640 × 640 improved Mask mAP@50-95 to 0.2554. Statistical validation using the Friedman test confirmed significant overall differences among models (χ² = 11.578, p = 0.0411), although Kendall’s W = 0.0322 indicated a small effect size. Finally, predicted masks were converted into Crack Area Ratio values to classify crack extent into low, moderate, and high severity levels. The innovation lies in providing a statistically supported YOLO benchmark and linking segmentation outputs to automated severity screening.

Graphical Abstract

Assessment of YOLO-Based Deep Learning Models for Pavement Crack Segmentation and Crack Extent Estimation

Highlights

Six YOLO26/YOLO11 models benchmarked for pavement crack instance segmentation.
YOLO26m-seg-640 achieved the best mask-level crack segmentation accuracy.
Friedman and Wilcoxon tests with bootstrap CIs support conservative model ranking.
Crack Area Ratio (CAR) introduced as a low-cost crack severity screening index.
Higher input resolution (640 vs 448) improved boundary-sensitive mask accuracy.

Keywords

Pavement crack detection

Insstance segmentation

YOLO26

YOLO11

Road maintenance

Subjects

Pavement Evaluation and Rehabilitation

[1] Shahin MY. Pavement Management for Airports, Roads, and Parking Lots. Boston, MA: Springer US; 2005. https://doi.org/10.1007/b101538.

[2] ASTM International. Practice for Roads and Parking Lots Pavement Condition Index Surveys 2020. https://doi.org/10.1520/D6433-20.

[3] Koch C, Georgieva K, Kasireddy V, Akinci B, Fieguth P. A review on computer vision based defect detection and condition assessment of concrete and asphalt civil infrastructure. Adv Eng Informatics 2015;29:196–210. https://doi.org/10.1016/j.aei.2015.01.008.

[4] Sony S, Laventure S, Sadhu A. A literature review of next-generation smart sensing technology in structural health monitoring. Struct Control Heal Monit 2019;26:e2321. https://doi.org/10.1002/stc.2321.

[5] Dong C-Z, Catbas FN. A review of computer vision–based structural health monitoring at local and global levels. Struct Heal Monit 2021;20:692–743. https://doi.org/10.1177/1475921720935585.

[6] Maeda H, Sekimoto Y, Seto T, Kashiyama T, Omata H. Road Damage Detection and Classification Using Deep Neural Networks with Smartphone Images. Comput Civ Infrastruct Eng 2018;33:1127–41. https://doi.org/10.1111/mice.12387.

[7] Arya D, Maeda H, Ghosh SK, Toshniwal D, Mraz A, Kashiyama T, et al. Deep learning-based road damage detection and classification for multiple countries. Autom Constr 2021;132:103935. https://doi.org/10.1016/j.autcon.2021.103935.

[8] Cha Y, Choi W, Büyüköztürk O. Deep Learning‐Based Crack Damage Detection Using Convolutional Neural Networks. Comput Civ Infrastruct Eng 2017;32:361–78. https://doi.org/10.1111/mice.12263.

[9] Zou Q, Zhang Z, Li Q, Qi X, Wang Q, Wang S. DeepCrack: Learning Hierarchical Convolutional Features for Crack Detection. IEEE Trans Image Process 2019;28:1498–512. https://doi.org/10.1109/TIP.2018.2878966.

[10] Yang F, Zhang L, Yu S, Prokhorov D, Mei X, Ling H. Feature Pyramid and Hierarchical Boosting Network for Pavement Crack Detection. IEEE Trans Intell Transp Syst 2020;21:1525–35. https://doi.org/10.1109/TITS.2019.2910595.

[11] Fan Z, Wu Y, Lu J, Li W. Automatic Pavement Crack Detection Based on Structured Prediction with the Convolutional Neural Network. ArXiv Prepr 2018. https://doi.org/10.48550/arXiv.1802.02208.

[12] Lau SLH, Chong EKP, Yang X, Wang X. Automated Pavement Crack Segmentation Using U-Net-Based Convolutional Neural Network. IEEE Access 2020;8:114892–9. https://doi.org/10.1109/ACCESS.2020.3003638.

[13] Ronneberger O, Fischer P, Brox T. U-Net: Convolutional Networks for Biomedical Image Segmentation. Lect. Notes Comput. Sci. (including Subser. Lect. Notes Artif. Intell. Lect. Notes Bioinformatics), vol. 9351, 2015, p. 234–41. https://doi.org/10.1007/978-3-319-24574-4_28.

[14] He K, Gkioxari G, Dollar P, Girshick R. Mask R-CNN. 2017 IEEE Int. Conf. Comput. Vis., IEEE; 2017, p. 2980–8. https://doi.org/10.1109/ICCV.2017.322.

[15] Long J, Shelhamer E, Darrell T. Fully convolutional networks for semantic segmentation. 2015 IEEE Conf. Comput. Vis. Pattern Recognit., vol. 07-12- June, IEEE; 2015, p. 3431–40. https://doi.org/10.1109/CVPR.2015.7298965.

[16] Shi Y, Cui L, Qi Z, Meng F, Chen Z. Automatic Road Crack Detection Using Random Structured Forests. IEEE Trans Intell Transp Syst 2016;17:3434–45. https://doi.org/10.1109/TITS.2016.2552248.

[17] Zhang L, Yang F, Daniel Zhang Y, Zhu YJ. Road crack detection using deep convolutional neural network. 2016 IEEE Int. Conf. Image Process., vol. 2016- Augus, IEEE; 2016, p. 3708–12. https://doi.org/10.1109/ICIP.2016.7533052.

[18] Redmon J, Divvala S, Girshick R, Farhadi A. You Only Look Once: Unified, Real-Time Object Detection. 2016 IEEE Conf. Comput. Vis. Pattern Recognit., vol. 2016- Decem, IEEE; 2016, p. 779–88. https://doi.org/10.1109/CVPR.2016.91.

[19] Redmon J, Farhadi A. YOLO9000: Better, Faster, Stronger. 2017 IEEE Conf. Comput. Vis. Pattern Recognit., vol. 2017- Janua, IEEE; 2017, p. 6517–25. https://doi.org/10.1109/CVPR.2017.690.

[20] Redmon J, Farhadi A. YOLOv3: An Incremental Improvement. ArXiv Prepr 2018. https://doi.org/10.48550/arXiv.1804.02767.

[21] Bochkovskiy A, Wang C-Y, Liao H-YM. YOLOv4: Optimal Speed and Accuracy of Object Detection. ArXiv Prepr ArXiv200410934 2020. https://doi.org/10.48550/arXiv.2004.10934.

[22] Wang C-Y, Bochkovskiy A, Liao H-YM. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors. 2023 IEEE/CVF Conf Comput Vis Pattern Recognit 2022;2023-June:7464–75. https://doi.org/10.1109/CVPR52729.2023.00721.

[23] Jocher G, Chaurasia A, Qiu J. Ultralytics YOLO. Ultralytics 2023.

[24] Ultralytics. Ultralytics YOLO11 Documentation. Ultralytics Docs 2024.

[25] Jocher G, Qiu J, Liu M, Lyu S, Akyon FC, Kalfaoglu ME. Ultralytics YOLO26: Unified Real-Time End-to-End Vision Models. ArXiv Prepr ArXiv260603748 2026. https://doi.org/10.48550/arXiv.2606.03748.

[26] Fan L, Tang S, Ariffin MKA b. M, Ismail MIS, Wang X. Impact of Image Preprocessing and Crack Type Distribution on YOLOv8-Based Road Crack Detection. Sensors 2025;25:2180. https://doi.org/10.3390/s25072180.

[27] Mohamed ME, Esmat AA, Hamdy AM, El Sheshtawy AM. Assessment of Deep Learning Models for Pavement Distress Detection in High-Resolution UAV Imagery. J Rehabil Civ Eng 2026;14:1–23. https://doi.org/10.22075/jrce.2025.2415.

[28] Fakharian P, Naderpour H. Damage Severity Quantification Using Wavelet Packet Transform and Peak Picking Method. Pract Period Struct Des Constr 2022;27:04021063. https://doi.org/10.1061/(ASCE)SC.1943-5576.0000639.

[29] Soleimani Nezhad S, Khademian F, Naderpour H, Kalantari SM, Fakharian P. Signal processing-based damage detection of steel braced frame subjected to consequent excitations. Innov Infrastruct Solut 2024;9:454. https://doi.org/10.1007/s41062-024-01762-5.

[30] Nouri Y, Shahabian F, Shariatmadar H, Entezami A. Structural Damage Detection in the Wooden Bridge Using the Fourier Decomposition, Time Series Modeling and Machine Learning Methods. J Soft Comput Civ Eng 2024;8:83–101. https://doi.org/10.22115/scce.2023.401971.1669.

[31] Lin T-Y, Maire M, Belongie S, Bourdev L, Girshick R, Hays J, et al. Microsoft COCO: Common Objects in Context. Lect Notes Comput Sci (Including Subser Lect Notes Artif Intell Lect Notes Bioinformatics) 2015;8693 LNCS:740–55. https://doi.org/10.1007/978-3-319-10602-1_48.

[32] Dice LR. Measures of the Amount of Ecologic Association Between Species. Ecology 1945;26:297–302. https://doi.org/10.2307/1932409.

[33] Jaccard P. Étude comparative de la distribution florale dans une portion des Alpes et des Jura. Bull La Société Vaudoise Des Sci Nat 1901;37:547–79.

[34] Friedman M. The Use of Ranks to Avoid the Assumption of Normality Implicit in the Analysis of Variance. J Am Stat Assoc 1937;32:675. https://doi.org/10.2307/2279372.

[35] Wilcoxon F. Individual Comparisons by Ranking Methods. Biometrics Bull 1945;1:80. https://doi.org/10.2307/3001968.

[36] Rezatofighi H, Tsoi N, Gwak J, Sadeghian A, Reid I, Savarese S. Generalized Intersection Over Union: A Metric and a Loss for Bounding Box Regression. 2019 IEEE/CVF Conf. Comput. Vis. Pattern Recognit., vol. 2019- June, IEEE; 2019, p. 658–66. https://doi.org/10.1109/CVPR.2019.00075.

[37] Powers DMW. Evaluation: from precision, recall and F-measure to ROC, informedness, markedness and correlation. J Mach Learn Technol 2020;2:37–63. https://doi.org/2010.16061.

[38] Everingham M, Van Gool L, Williams CKI, Winn J, Zisserman A. The Pascal Visual Object Classes (VOC) Challenge. Int J Comput Vis 2010;88:303–38. https://doi.org/10.1007/s11263-009-0275-4.

[39] Holm S. A Simple Sequentially Rejective Multiple Test Procedure. Scand J Stat 1979;6:65–70. https://doi.org/10.2307/4615733.

[40] Efron B, Tibshirani RJ. An Introduction to the Bootstrap. Chapman and Hall/CRC; 1994. https://doi.org/10.1201/9780429246593.

[41] Ha J, Kim D, Kim M. Assessing severity of road cracks using deep learning-based segmentation and detection. J Supercomput 2022;78:17721–35. https://doi.org/10.1007/s11227-022-04560-x.

[42] Owor NJ, Du H, Daud A, Aboah A, Adu-Gyamfi Y. Image2PCI -- A Multitask Learning Framework for Estimating Pavement Condition Indices Directly from Images. ArXiv Prepr ArXiv231008538 2023. https://doi.org/10.48550/arXiv.2310.08538.