基于改进YOLOv8的道路表面缺陷检测  

Road Surface Defect Detection Based on Improved YOLOv8

作　　者：尹晶 胡永祥[1] YIN Jing;HU Yong-xiang(Hunan University of Technology,Zhuzhou 412007,China)

机构地区：[1]湖南工业大学,株洲412007

出　　处：《价值工程》2024年第9期140-143,共4页Value Engineering

摘　　要：道路损伤检测是道路安全及道路损伤及时修复的重要前提。针对现有的道路损伤检测算法存在的精度较低,计算量大等问题,提出了一种基于YOLOv8改进的轻量型道路损伤检测算法。首先,将骨干网络中的C2f模块替换为C2f_FasterNext模块,增强有效特征复用的同时降低计算复杂度;其次,然后在骨干网络末端和颈部网络中引入坐标注意力机制(Coordinate Attention,CA),将位置信息嵌入到通道注意力中,强化特征提取能力,并抑制无关特征的干扰。在开源道路损害数据集RDD20(Road Damage Detection20)上的实验结果表明:所提方法的平均F1得分为0.61,每秒检测帧数(FPS)为88,模型大小为45.5MB,改进算法与原算法相比m AP50分别提高了2%和3.7%,算法检测速度达到88FPS,能够实时准确检测道路损伤目标。通过与其他主流目标检测算法比较,验证了该方法的有效性和优越性。Road damage detection is an important prerequisite for road safety and timely repair of road damage.A lightweight road damage detection algorithm based on YOLOv8 improvement is proposed to address the problems of low accuracy and large computational load in existing road damage detection algorithms.Firstly,the C2f module in the backbone network is replaced with the C2f-FasterNext module to enhance effective feature reuse while reducing computational complexity.Secondly,coordinate attention(CA)mechanism is introduced into the backbone network end and neck network to embed position information into channel attention,enhance feature extraction ability,and suppress interference from irrelevant features.The experimental results on the open-source road damage dataset RDD20(Road Damage Detection 20)show that the proposed method has an average F1 score of 0.61,a detection frame rate per second(FPS)of 88,and a model size of 45.5 MB.Compared with the original algorithm,the improved algorithm has increased mAP50 by 2%and 3.7%,respectively.The detection speed of the algorithm reaches 88 FPS,and it can accurately detect road damage targets in real-time.Compared with other mainstream object detection algorithms,the effectiveness and superiority of this method have been verified.

关 键 词：目标检测 道路损伤 YOLOv8 注意力机制 深度学习 

分 类 号：U418[交通运输工程—道路与铁道工程]