华南理工大学学报(自然科学版) ›› 2024, Vol. 52 ›› Issue (10): 146-158.doi: 10.12141/j.issn.1000-565X.230350

所属专题: 2024年图像处理

• 图像处理 • 上一篇    

交通基础设施裂缝病害图像增广方法

蒋盛川1(), 钟山2(), 吴荻非2, 刘成龙2   

  1. 1.上海理工大学 交通系统工程系,上海 200093
    2.同济大学 道路与交通工程教育部重点实验室,上海 201804
  • 收稿日期:2023-05-24 出版日期:2024-10-25 发布日期:2024-01-26
  • 通信作者: 钟山 E-mail:jsc@usst.edu.cn;zhongshanbetter@tongji.edu.cn
  • 作者简介:蒋盛川(1987—),女,博士,副教授,主要从事交通基础设施智慧运维管理研究。E-mail: jsc@usst.edu.cn
  • 基金资助:
    国家自然科学基金资助项目(52202390);上海市科委启明星项目(22QB1405100)

Augmentation Method of Transportation Infrastructure Crack Images

JIANG Shengchuan1(), ZHONG Shan2(), WU Difei2, LIU Chenglong2   

  1. 1.Department of Transportation System Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China
    2.Key Laboratory of Road and Traffic Engineering of the Ministry of Education,Tongji University,Shanghai 201804,China
  • Received:2023-05-24 Online:2024-10-25 Published:2024-01-26
  • Contact: ZHONG Shan E-mail:jsc@usst.edu.cn;zhongshanbetter@tongji.edu.cn
  • Supported by:
    the National Natural Science Foundation of China(52202390);the Star Project of Shanghai Science and Technology Commission(22QB1405100)

摘要:

基于深度学习的交通基础设施裂缝检测模型依赖于大规模的数据进行训练。针对特定交通设施场景下多种形态的裂缝样本图像难以获取的问题,文中提出了一种基于Pix2PixHD模型的交通基础设施裂缝病害图像增广方法。首先,基于收集的少量裂缝图像数据,利用Pix2PixHD模型建立裂缝病害的真实图像和标注标签之间的空间映射关系;然后,编辑标签域中的对象,通过迁移其他数据集标签、人工编辑、图像形态学膨胀操作、随机叠加操作等生成表征多种形态的裂缝轮廓;最后,将编辑后的标签域通过Pix2PixHD模型转换回图像域,以实现交通基础设施裂缝数据集的自适应增广。文中考虑了主要的交通基础设施材料和结构(沥青路面、隧道衬砌和混凝土材料结构),选取GAPS384、Tunnel200和DeepCrack数据集进行实验。结果表明:通过扩增数据集训练的U-Net模型,检测精度更高,更易于跳出局部最优解;相比于DCGAN模型,文中方法能够控制裂缝的形态且保证骨架的连续性,可以有针对地提高原有裂缝数据集的形态多样性,使检测模型在特定交通基础设施场景下具有更好的泛化能力。

关键词: 交通基础设施, 图像生成, 裂缝检测, 生成对抗网络

Abstract:

The crack detection model for transportation infrastructure based on deep learning relies on large-scale data for training. To address the problem of limited availability of diverse crack samples in specific transportation facility scenarios, this paper proposed a transportation infrastructure crack image augmentation method based on Pix2PixHD model. Initially, the Pix2PixHD model was used to establish a spatial mapping relationship between real crack images and annotated labels based on a small amount of collected crack image data. Subsequently, the objects in the label domain were edited to generate crack contours representing various forms, using methods such as label transfer from other datasets, manual editing, morphological dilation operations, and random superimposition operations. Finally, the edited label domain was transformed back to the image domain using the Pix2PixHD model, so as to achieve an adaptive augmentation of the transportation infrastructure crack dataset. This paper considered major materials and structures in transportation infrastructure (asphalt pavement, tunnel linings, and concrete structures) and conducted experiments using the GAPS384, Tunnel200, and DeepCrack datasets. Results demonstrate that the U-Net model trained on the augmented dataset achieves higher detection accuracy and is more likely to avoid local optima. Compared to the DCGAN model, the proposed method effectively controls crack morphology while maintaining the continuity of the crack skeleton, thereby enhancing the morphological diversity of the original crack dataset and improving the generalization capability of the detection model in specific transportation infrastructure scenarios.

Key words: transportation infrastructure, image generation, crack detection, generative adversarial network

中图分类号: