《计算机应用》唯一官方网站 ›› 2022, Vol. 42 ›› Issue (1): 265-272.DOI: 10.11772/j.issn.1001-9081.2021010206

• 前沿与综合应用 • 上一篇    

基于卷积神经网络的细长路面病害检测方法

许慧青1,2, 陈斌2(), 王敬飞3,4, 陈志毅3,4, 覃健5   

  1. 1.中国科学院 成都计算机应用研究所, 成都 610041
    2.中国科学院大学, 北京 101408
    3.广东华路交通科技有限公司, 广州 510420
    4.广东交科检测有限公司, 广州 510550
    5.中科院广州电子技术有限公司, 广州 510070
  • 收稿日期:2021-02-03 修回日期:2021-04-25 接受日期:2021-04-28 发布日期:2022-01-11 出版日期:2022-01-10
  • 通讯作者: 陈斌
  • 作者简介:许慧青(1989—),男,河南汝南人,博士研究生,主要研究方向:目标检测、语义分割、路面病害智能检测
    陈斌(1970—),男,四川广汉人,研究员,博士,主要研究方向:机器视觉、深度学习
    王敬飞(1980—),男,河南漯河人,高级工程师,硕士,主要研究方向:快速道路检测和评估
    陈志毅(1986—),男,广东茂名人,工程师,主要研究方向:快速路面检测
    覃健(1992—),男,广西贺州人,工程师,主要研究方向:工业自动化。

Elongated pavement distress detection method based on convolutional neural network

Huiqing XU1,2, Bin CHEN2(), Jingfei WANG3,4, Zhiyi CHEN3,4, Jian QIN5   

  1. 1.Chengdu Institute of Computer Applications,Chinese Academy of Sciences,Chengdu Sichuan 610041,China
    2.University of Chinese Academy of Sciences,Beijing 101408,China
    3.Guangdong Hualu Transport Technology Company Limited,Guangzhou Guangdong 510420,China
    4.Guangdong Jiaoke Testing Company Limited,Guangzhou Guangdong 510550,China
    5.Guangzhou Electronic Technology Company Limited,Chinese Academy of Sciences,Guangzhou Guangdong 510070,China
  • Received:2021-02-03 Revised:2021-04-25 Accepted:2021-04-28 Online:2022-01-11 Published:2022-01-10
  • Contact: Bin CHEN
  • About author:XU Huiqing, born in 1989, Ph. D. candidate. His research interests include target detection, semantic segmentation, intelligent detection of pavement distress.
    CHEN Bin, born in 1970, Ph. D., research fellow. His research interests include machine vision, deep learning.
    WANG Jingfei, born in 1980, M. S., senior engineer. His research interests include rapid pavement detection and evaluation.
    CHEN Zhiyi, born in 1986, engineer. His research interests include rapid pavement detection.
    QIN Jian, born in 1992, engineer. His research interests include industrial automation.

摘要:

针对细长路面病害人工检测耗时长和当前检测方法精度不足的问题,依据病害的弱语义特性和异常几何属性,提出了能够精准定位和分类出病害的二阶段细长路面病害检测方法Epd RCNN。首先,针对细长路面病害的弱语义特性,提出了一种复用低层特征并反复融合不同阶段特征的骨干网络;其次,在训练过程中,使用一种符合病害几何属性分布的锚框机制来生成高质量的正样本供网络训练;然后,在单一高分辨率特征图上预测病害包围框,并针对该特征图使用并行级联空洞卷积模块来提升其多尺度特征表达能力;最后,针对形状各异的候选区域,使用由可变形感兴趣区域池化(RoI Pooling)和空间注意力模块组成的候选区域特征改良模块来提取符合病害几何属性的候选区域特征。实验结果表明,所提方法在光照充足图像上的平均准确率均值(mAP)为0.907,在存在光照问题图像上的mAP为0.891,综合mAP为0.899, 表明该方法具有良好的检测性能和对光照的鲁棒性。

关键词: 细长路面病害, 卷积神经网络, 包围框, 几何属性, 并行级联空洞卷积, 候选区域特征, 空间注意力

Abstract:

Focusing on the problems of the large time consumption of manual detection and the insufficient precision of the current detection methods of elongated pavement distress, a two-stage elongated pavement distress detection method, named Epd RCNN (Elongated pavement distress Region-based Convolutional Neural Network), which could accurately locate and classify the distress was proposed according to the weak semantic characteristics and abnormal geometric properties of the distress. Firstly, for the weak semantic characteristics of elongated pavement distress, a backbone network that reused low-level features and repeatedly fused the features of different stages was proposed. Secondly, in the training process, the high-quality positive samples for network training were generated by the anchor box mechanism conforming to the geometric property distribution of the distress. Then, the distress bounding boxes were predicted on a single high-resolution feature map, and a parallel cascaded dilated convolution module was used to this feature map to improve its multi-scale feature representation ability. Finally, for different shapes of region proposals, the region proposal features conforming to the distress geometric properties were extracted by the proposal feature improvement module composed of deformable Region of Interest Pooling (RoI Pooling) and spatial attention module. Experimental results show that the proposed method has the mean Average Precision (mAP) of 0.907 on images with sufficient illumination, the mAP of 0.891 on images with illumination problems and the comprehensive mAP of 0.899, indicating that the proposed method has good detection performance and robustness to illumination.

Key words: elongated pavement distress, Convolutional Neural Network (CNN), bounding box, geometric property, parallel cascaded dilated convolution, region proposal feature, spatial attention

中图分类号: