%0 Journal Article
%T 基于YOLO算法的路面缺陷识别方法研究
Research on Road Defect Detection Method Based on YOLO Algorithm
%A 傅蒙恩
%A 张婷
%A 王元兴
%J Computer Science and Application
%P 117-125
%@ 2161-881X
%D 2025
%I Hans Publishing
%R 10.12677/csa.2025.151012
%X 本文针对路面缺陷检测方法在精度、效率及小目标检测能力方面的不足，提出了一种基于YOLO11的改进算法。通过设计EIEStem模块替代骨干网络的前两个卷积层，融合边缘特征和空间信息，有效提升了模型对小缺陷的检测能力。实验结果表明，改进算法在mAP0.5、推理速度(FPS)等指标上均展现显著优势：在mAP0.5上达到0.854，比原YOLO11提高1.5%，相比Faster-RCNN提升12.7%；推理速度达到113.3 FPS，相较YOLO11提高了约15.8%，远超Faster-RCNN的21 FPS。此外，计算量仅小幅增加，GFLOPs从6.3增至6.4，验证了改进算法的高效性与优越性。
This paper addresses the limitations of road defect detection methods in terms of accuracy, efficiency, and small object detection capabilities. An improved algorithm based on YOLO11 is proposed by introducing the EIEStem module to replace the first two convolutional layers of the backbone network, integrating edge features and spatial information to enhance the detection capability for small defects. Experimental results demonstrate that the improved algorithm achieves significant advantages in key metrics such as mAP0.5 and FPS: achieving an mAP0.5 of 0.854, which is a 1.5% improvement over the original YOLO11 and a 12.7% increase compared to Faster-RCNN. It achieves an inference speed of 113.3 FPS, approximately 15.8% faster than YOLO11 and far exceeding the 21 FPS of Faster-RCNN. Additionally, the computational cost only increased slightly, with GFLOPs rising from 6.3 to 6.4, confirming the efficiency and superiority of the improved algorithm.
%K YOLO11，
%K 路面缺陷检测，
%K 目标检测，
%K 深度学习
YOLO11
%K Road Defect Detection
%K Object Detection
%K Deep Learning
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=106155