Road pavement damage poses significant risks to driving comfort and safety. It can also contribute to traffic accidents, ultimately leading to losses in life and property. Accurate detection of road pavement damage is, therefore, crucial for maintaining road safety and upkeep. This paper introduces a novel methodology for detecting road pavement damage, which is based on an enhanced YOLOv8 network. We implemented three pivotal enhancements: first, we substituted the backbone network with GhostNetv2, significantly reducing the model’s parameter count while bolstering its feature extraction capabilities. Second, we integrated the efficient multi-scale attention (EMA) mechanism, which strengthened the model’s generalization ability, thereby indirectly enhancing the robustness of its feature extraction. Last, we transitioned from Complete Intersection over Union (CIoU) loss to SCYLLA Intersection over Union (SIoU) loss, aiming to elevate the flexibility and precision of the bounding box loss function. Additionally, we curated the RDD2022 data set to create the RDDYGES data set, which served as the foundation for training, testing, and validating our model. The culmination of these strategies led to the development of the YOLOv8s-GES model, specifically tailored for the detection of road pavement damage. Through evaluation, YOLOv8s-GES achieved a precision of 83.6%, recall of 70.5%, an F1 score of 76.5%, and mAP of 73.7%, with a model size of 7.29M. A comparative analysis revealed the superiority of YOLOv8s-GES over existing models, demonstrating its balanced trade-off between model size, detection speed, and detection accuracy, thereby establishing a strong foundation for further research in this domain.


    Zugriff

    Download

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    Deep Learning Network Model for Road Pavement Damage Detection Based on YOLOv8s-GES


    Weitere Titelangaben:

    Transportation Research Record: Journal of the Transportation Research Board


    Beteiligte:
    Wang, Niannian (Autor:in) / Gao, Ziqiao (Autor:in) / Fang, Hongyuan (Autor:in) / Hu, Haobang (Autor:in) / Li, Bin (Autor:in) / Wang, Cuixia (Autor:in)


    Erscheinungsdatum :

    02.02.2025




    Medientyp :

    Aufsatz (Zeitschrift)


    Format :

    Elektronische Ressource


    Sprache :

    Englisch



    Track Fastener Detection Method Based on Improved Yolov8s Algorithm

    Zhu, Jianpeng / Zhang, Haigang / Wu, Lei et al. | IEEE | 2024


    Detection of Pavement Maintenance Treatments using Deep-Learning Network

    Gao, Lu / Yu, Yao / Hao Ren, Yi et al. | Transportation Research Record | 2021



    Deep-Learning-Based Intelligent PotholeEye+ Detection Pavement Distress Detection System

    Gorintla, Shobana / Kumar, B. Anil / Chanadana, B. Sai et al. | IEEE | 2022


    Automated Pavement Crack Damage Detection Using Deep Multiscale Convolutional Features

    Weidong Song / Guohui Jia / Hong Zhu et al. | DOAJ | 2020

    Freier Zugriff