A Lightweight YOLO26 Pavement Defect Detection Method for Complex Pavement Environments

DOI: https://doi.org/10.62517/jbdc.202601219

Author(s)

Borong Zhu1, Kaiying Tian1, Qihao Xu1, Yixuan Wang1, Dingyu Wu2, Tianyuan Liu1,*

Affiliation(s)

1School of Artificial Intelligence and Big Data, Henan University of Technology, Zhengzhou, Henan, China 2School of Mechanical and Electrical Engineering, Henan University of Technology, Zhengzhou, Henan, China *Corresponding Author

Abstract

Timely identification of pavement deterioration plays a vital role in upholding traffic safety standards and optimizing the efficiency of road maintenance operations. Current approaches often struggle with extracting discriminative features for small-scale defects embedded in cluttered backgrounds, while surface textures and cast shadows frequently trigger both missed detections and spurious alarms. To overcome these limitations, a lightweight road damage detection framework built upon YOLO26 is presented in this work, specifically tailored for challenging pavement conditions. Taking the recently released YOLO26n model from Ultralytics as the foundation, and while strictly constraining model parameters and underlying computational expenditure, an Efficient Multi-Scale Attention (EMA) mechanism is appended and integrated at the terminal stage of the backbone network to further enhance detection accuracy. Through adaptive orthogonal and cross-spatial calibration of channel-wise feature responses, this mechanism effectively suppresses irrelevant background clutter and reinforces the representational strength of critical damage signatures. Evaluation on the China partition of the publicly available RDD2022 dataset reveals that the proposed E-YOLO26 model attains a Precision of 78.0% and an mAP@0.5 of 74.0%, accompanied by an exceptionally low computational burden of merely 5.9 GFLOPs. These figures correspond to gains of 2.6 and 0.1 percentage points relative to the baseline YOLO26 model, respectively. The findings of this study furnish an optimized computational trade-off solution along with practical engineering support for automated detection on embedded devices deployed in real-world road inspection scenarios.

Keywords

Road Damage; YOLO26; Attention Mechanism; Deep Learning; Feature Enhancement; RDD2022

References

[1] Zhang J Q, Yang X, Wang W, et al. Automated guided vehicles and autonomous mobile robots for recognition and tracking in civil engineering. Automation in Construction, 2023, 146: 104699. [2] Luo Z, Jiang Y, Li W C. Road Defect Detection Model Based on Improved YOLO11n. Microelectronics & Computer, 2025, 42(11): 25-36. [3] Liu D D, Xue J J, Zhao X K, et al. YOLOv8 Road Crack Detection Method Based on Frequency-Aware Feature Fusion. Journal of Spatiotemporal Information, 2025, 32(06): 675-684. [4] Feng C H, Xu H Y, Tao J Q, et al. SSBS-YOLO: Optimization of Small Target Pavement Damage Detection in YOLOv8 Based on Lightweight Road Inspection. Journal of Zhejiang Normal University (Natural Sciences), 2026:1-12[2026-05-10]. [5] Yavuz B. Scale-Dependent Performance Analysis of YOLO26 and YOLOv11 for PPE Detection. Electronics, 2026, 15(6): 1146. [6] Su W G, Wang J X. Research on Road Crack Recognition Model Based on YOLO v3 Deep Learning Algorithm. Journal of China & Foreign Highway, 2023, 43(2): 58-63. [7] Yuan H, Li Q, Wang Y. Road Crack Detection Based on YOLO-CD. Science Technology and Engineering, 2025, 25(9):3888-3895 [8] Li G. Road Defect Identification and Location Method Based on an Improved ML-YOLO Algorithm. Sensors (Basel, Switzerland), 2026, 24(21):6783. [9] Liu Y Y, Zhu K, Gu Z H, et al. Lightweight Pavement Crack Detection Method with Adaptive Features. Optics and Precision Engineering, 2026, 34, (2):336-351. [10] Ning Z P, Wang H, Li S L, et al.YOLOv7-RDD: A lightweight efficient pavement distress detection model. IEEE Transactions on Intelligent Transportation Systems, 2024, 25(7):6994-7003. [11] Xuan Y G, Yu C B, Jiang Q C, et al. Road Crack and Pothole Detection Algorithm Based on Improved YOLOv7. Science Technology and Engineering, 2024, 24(17): 7205-7213. [12] Arya D, Maeda H, Ghosh S K, et al. RDD2022: a multi-national image dataset for automatic road damage detection. Geoscience Data Journal, 2024, 11(4):846-862.