改进YOLOv8的道路凹陷检测算法
Improved road depression detection algorithm for YOLOv8
张旭中 1李波 2贝绍轶 1林棻 3殷国栋4
作者信息
- 1. 江苏理工学院汽车与交通工程学院,江苏常州 213001
- 2. 江苏理工学院汽车与交通工程学院,江苏常州 213001;清华大学苏州汽车研究院,江苏苏州 215200
- 3. 南京航空航天大学,南京 210016
- 4. 东南大学机械工程学院,南京 210096
- 折叠
摘要
针对现有的道路凹陷检测算法中检测速度慢,很难应用于汽车车载移动设备的问题,提出一种改进YOLOv8的轻量型道路凹陷检测算法YOLOv8-CAG.将YOLOv8的主干网络第二层之后的普通卷积替换成Ghost Conv,通过低廉的线性变换,有效减少了模型的参数量.在neck中的C2f模块中引入CA注意力机制,在降低整体模型参数量和浮点运算量的同时,强化特征提取能力,减少无关特征的影响.在YOLOv8中运用C2f-GS模块,减少网络结构的复杂性,进一步提升检测精度.实验结果表明:在道路凹陷的数据集上,改进算法与原算法相比,检测精度提高了1％,模型参数量与计算量分别下降了 16％和11％,并通过与其他算法的性能比较,验证了改进算法的实用性.
Abstract
To address the slow detection speed in the existing road depression detection algorithms,which are hardly applicable on vehicle-mounted mobile devices,this paper proposes a lightweight road depression detection algorithm,YOLOv8-CAG.First,the ordinary convolution after the second layer of the backbone network of YOLOv8 is replaced with the Ghost Conv,which effectively reduces the parameter number of the model by an inexpensive linear transformation.Second,the CA attention mechanism is introduced into the C2f module in the neck,reducing the number of parameters and floating-point operations of the overall model,offsetting the impacts of irrelevant features while strengthening the feature extraction capability.Finally,the C2f-GS module is utilized in YOLOv8 to decrease the complexity of the network structure and further improve the detection accuracy.The practicality of our improved algorithm is verified by comparing its performance with other algorithms.Experimental results show,on the dataset of road depression,our algorithm improves the performance by 1％compared with that of the original one.And the number of model parameters and the computation are down by 16％and 11％respectively.
关键词
道路凹陷检测/YOLOv8/Ghost卷积/注意力机制/C2f-GS模块Key words
road depression detection/YOLOv8/Ghost Conv/attention mechanism/C2f-GS module引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据