Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DAPONet: A Dual Attention and Partially Overparameterized Network for Real-Time Road Damage Detection

Weichao Pan, Jiaju Kang, Xu Wang, Zhihao Chen, Yiyuan Ge

TL;DR

DAPONet is proposed, a model incorporating three key modules: a dual attention mechanism combining global and local attention, a multi-scale partial over-parameterization module, and an efficient downsampling module, to enhance real-time road damage detection using street view image data (SVRDD).

Abstract

Current road damage detection methods, relying on manual inspections or sensor-mounted vehicles, are inefficient, limited in coverage, and often inaccurate, especially for minor damages, leading to delays and safety hazards. To address these issues and enhance real-time road damage detection using street view image data (SVRDD), we propose DAPONet, a model incorporating three key modules: a dual attention mechanism combining global and local attention, a multi-scale partial over-parameterization module, and an efficient downsampling module. DAPONet achieves a mAP50 of 70.1% on the SVRDD dataset, outperforming YOLOv10n by 10.4%, while reducing parameters to 1.6M and FLOPs to 1.7G, representing reductions of 41% and 80%, respectively. On the MS COCO2017 val dataset, DAPONet achieves an mAP50-95 of 33.4%, 0.8% higher than EfficientDet-D1, with a 74% reduction in both parameters and FLOPs.

DAPONet: A Dual Attention and Partially Overparameterized Network for Real-Time Road Damage Detection

TL;DR

DAPONet is proposed, a model incorporating three key modules: a dual attention mechanism combining global and local attention, a multi-scale partial over-parameterization module, and an efficient downsampling module, to enhance real-time road damage detection using street view image data (SVRDD).

Abstract

Current road damage detection methods, relying on manual inspections or sensor-mounted vehicles, are inefficient, limited in coverage, and often inaccurate, especially for minor damages, leading to delays and safety hazards. To address these issues and enhance real-time road damage detection using street view image data (SVRDD), we propose DAPONet, a model incorporating three key modules: a dual attention mechanism combining global and local attention, a multi-scale partial over-parameterization module, and an efficient downsampling module. DAPONet achieves a mAP50 of 70.1% on the SVRDD dataset, outperforming YOLOv10n by 10.4%, while reducing parameters to 1.6M and FLOPs to 1.7G, representing reductions of 41% and 80%, respectively. On the MS COCO2017 val dataset, DAPONet achieves an mAP50-95 of 33.4%, 0.8% higher than EfficientDet-D1, with a 74% reduction in both parameters and FLOPs.
Paper Structure (16 sections, 2 figures, 3 tables)

This paper contains 16 sections, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Overview
  4. Global Localization Context Attention (GLCA) Module
  5. Cross Stage Partial Depthwise Over-parameterized Attention (CPDA) Module
  6. Mix Convolutional Downsampling (MCD) Module
  7. Experimental details
  8. Datasets
  9. Experimental environment
  10. Evaluation metrics
  11. Experimental results and discussion and analysis
  12. Comparative experiments
  13. Generalized object detection experiments
  14. Ablation study
  15. Error analysis
  16. ...and 1 more sections

Figures (2)

  • Figure 1: Shows the overall framework. In GLCA Module, GN is Group Normalization, CLR is Conv-Layer Normal-ReLU, and CB is Conv-Batch Normalization. In the PB Block of CPDA Module, a quarter of the input feature map is divided into DOConv channels, and the rest is directly concat with the feature map after DOConv operation.
  • Figure 2: Experimental models recognize visual results on the SVRDD dataset. Different models vary in detecting road damage. YOLOv5n struggles with minor cracks, resulting in higher miss rates. YOLOv8n improves accuracy for transverse cracks and manhole covers but still misses subtle damages. YOLOv9t prioritizes speed but loses precision in detecting finer details. YOLOv10n is effective for larger cracks but has more false negatives for smaller damages. DAPONet outperforms the other models, accurately detecting a broad range of damages, including fine cracks and manhole covers, with high confidence and fewer errors, demonstrating its robustness and precision in various road damage scenarios.