.png){kind=icon}
.png){kind=icon}
Return
Lightweight real-time road defect detection algorithm integrating multi-coordinate aggregation attention and shared convolution
.png){kind=icon}
PRE
AI
DOI：10.1007/s11554-025-01747-2.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Prolonged road use leads to surface defects that, if undetected, degrade road life and pose safety risks. Conventional detection methods are slow and costly. To address this, LAR-YOLO (Lightweight Aggregate Re-param-YOLO), a lightweight model based on YOLOv8, was developed in this paper, featuring the RGCSPELAN (Re-param Ghost CSP ELAN) module to minimize model size and enhance detection speed. It includes the AMCA (Aggregate Multiple Coordinate Attention) mechanism for more accurate feature extraction and an Attention-Enhanced Screening Pyramid Network for improved feature representation, reduced feature loss, and better detection outcomes. Additionally, a Lightweight Shared Convolutional Detection Head (LSCD) was developed. Experimental results showed LAR-YOLO improved mAP50 (Mean Average Precision) by 4.8% over YOLOv8 on the RDD2022 dataset and reduced parameters and computational needs by 46.40% and 41.54%, respectively, achieving 212 FPS for real-time detection. It also outperformed other models on the VOC2007 and NEU-DET datasets, proving its practical value.
Keywords:
RGCSPELAN
AMCA
Surface defects
YOLOv8
Journal
IF:
3
Papers:
369
Citations:
2.0K