《计算机应用》唯一官方网站 ›› 2024, Vol. 44 ›› Issue (7): 2264-2270.DOI: 10.11772/j.issn.1001-9081.2023070956

• 多媒体计算与计算机仿真 • 上一篇    下一篇

基于加强特征提取的道路病害检测算法

龙伍丹1, 彭博1(), 胡节1, 申颖1,2, 丁丹妮3   

  1. 1.西南交通大学 计算机与人工智能学院, 成都 611756
    2.可持续城市交通智能化教育部工程研究中心(西南交通大学), 成都 611756
    3.成都信息工程大学 计算机学院, 成都 610225
  • 收稿日期:2023-07-17 修回日期:2023-09-10 接受日期:2023-09-20 发布日期:2023-10-26 出版日期:2024-07-10
  • 通讯作者: 彭博
  • 作者简介:龙伍丹(1998—),女,重庆人,硕士研究生,主要研究方向:深度学习、目标检测;
    胡节(1979—),女,四川成都人,副教授,博士,CCF会员,主要研究方向:数据挖掘,知识发现;
    申颖(1999—),女,四川资阳人,硕士研究生,主要研究方向:深度学习、目标检测;
    丁丹妮(1999—),女,湖北松滋人,硕士研究生,主要研究方向:图像处理、生物视觉。
    第一联系人:彭博(1980—),女,四川成都人,教授,博士,CCF会员,主要研究方向:计算机视觉、模式识别;
  • 基金资助:
    四川省自然科学基金资助项目(2022NSFSC0502);四川省科技计划项目(2023YFG0354);四川省科技创新苗子工程培育项目(MZGC20230077)

Road damage detection algorithm based on enhanced feature extraction

Wudan LONG1, Bo PENG1(), Jie HU1, Ying SHEN1,2, Danni DING3   

  1. 1.School of Computing and Artificial Intelligence,Southwest Jiaotong University,Chengdu Sichuan 611756,China
    2.Engineering Research Center of Ministry of Education for Sustainable Urban Transportation Intelligence (Southwest Jiaotong University),Chengdu Sichuan 611756,China
    3.School of Computer Science,Chengdu University of Information Technology,Chengdu Sichuan 610225,China
  • Received:2023-07-17 Revised:2023-09-10 Accepted:2023-09-20 Online:2023-10-26 Published:2024-07-10
  • Contact: Bo PENG
  • About author:LONG Wudan, born in 1998, M. S. candidate. Her research interests include deep learning, object detection.
    HU Jie, born in 1979, Ph. D., associate professor. Her research interests include data mining, knowledge discovery.
    SHEN Ying, born in 1999, M. S. candidate. Her research interests include deep learning, object detection.
    DING Danni, born in 1999, M. S. candidate. Her research interests include image processing, biological vision.
    First author contact:PENG Bo, born in 1980, Ph. D., professor. Her research interests include computer vision, pattern recognition.
  • Supported by:
    Natural Science Foundation of Sichuan Province(2022NSFSC0502);Science and Technology Plan Project of Sichuan Province(2023YFG0354);Cultivation Project of Sichuan Scientific and Technological Innovation Seedlings Engineering(MZGC20230077)

摘要:

针对道路病害区域小、类别数量不均衡导致检测困难的问题,提出基于YOLOv7-tiny的道路病害检测算法RDD-YOLO。首先,采用K-means++算法得到拟合目标尺寸更好的锚框。其次,在小目标检测支路上使用量化感知重参数化模块(QARepVGG),增强浅层特征提取,同时构建加强注意力模块(AM-CBAM)嵌入颈部的3个输入,抑制复杂背景干扰。然后,设计特征融合模块(Res-RFB),模拟人眼扩大感受野融合多尺度信息,提高表征能力;另外,构造轻量级解耦头(S-DeHead)提高小目标检测精确率。最后,采用归一化Wasserstein距离度量(NWD)优化小目标定位过程,并缓解样本不均衡问题。实验结果表明,与YOLOv7-tiny相比,RDD-YOLO算法在仅增加0.71×106参数量和1.7 GFLOPs计算量的成本下,mAP50提高6.19个百分点,F1-Score提高5.31个百分点,并且检测速度达到135.26 frame/s,满足道路养护工作中对检测精度和速度的需求。

关键词: 道路病害检测, 加强特征提取, YOLOv7-tiny, 小目标, 类别数量不平衡

Abstract:

In response to the challenge posed by the difficulty in detecting small road damage areas and the uneven distribution of damage categories, a road damage detection algorithm termed RDD-YOLO was introduced based on the YOLOv7-tiny architecture. Firstly, the K-means++ algorithm was employed to determine anchor boxes better conforming to object dimensions. Subsequently, a Quantization Aware RepVGG (QARepVGG) module was utilized within the auxiliary detection branch, thereby enhancing the extraction of shallow features. Concurrently, an Addition and Multiplication Convolutional Block Attention Module (AM-CBAM) was embedded into the three inputs of the neck, effectively suppressing disturbances arising from intricate background. Furthermore, the feature fusion module Res-RFB (Resblock with Receptive Field Block) was devised to emulate the expansion of receptive field in human visual perception, consequently fusing information across multiple scales and thereby amplifying representational aptitude. Additionally, a lightweight Small Decoupled Head (S-DeHead) was introduced to elevate the precision of detecting small objects. Ultimately, the process of localizing small objects was optimized through the application of the Normalized Wasserstein Distance (NWD) metric, which in turn mitigated the challenge of imbalanced samples. Experimental results show that RDD-YOLO algorithm achieves a notable 6.19 percentage points enhancement in mAP50, a 5.31 percentage points elevation in F1-Score and the detection velocity of 135.26 frame/s by only increasing 0.71×106 parameters and 1.7 GFLOPs, which can meet the requirements for both accuracy and speed in road maintenance.

Key words: road damage detection, enhanced feature extraction, YOLOv7-tiny, small object, category number unbalance

中图分类号: