摘要
为提高我国高速路面资产巡检效率,文章提出了一种有效改进SSD算法用于高速路面资产检测。首先,在单发多盒探测器(Single Shot MutiBox Detectior,SSD)算法的基础上引入MobileNetV3-Small网络与VGG16网络构成M&V(MobileNetV3-Small&VGG16)同步特征提取结构;其次,针对SSD中间4层特征图提出相邻三层融合机制(Adjacent Three Layer Fusion,ATLF),增强中低层特征图的语义信息。实验结果表明,所提算法在中国交通数据集上具有良好的检测精度,相较于SSD算法mAP提高了13%,在自定义资产数据集上也有良好的表现,mAP达到91.81%。
In order to enhance the efficiency of inspection of highway pavement assets in my country,an effective improved SSD algorithm is raised for expressway asset detection.Firstly,the MobileNetV3-Small network and the VGG16 are introduced to form an MobileNetV3-Small&VGG16(M&V)synchronous feature extraction network,secondly,a fusion mechanism is proposed for the middle four-layer feature map Adjacent Three Layer Fusion(ATLF).The experimental results proof that the M&V-ATLF SSD algorithm has excellent perception accuracy on CCTSDB,which is 13%higher than the mAP of the SSD algorithm.Performance of the improved algorithm is great on the custom dataset and the mAP is 91.81%.
作者
魏鑫宇
王池社
WEI Xinyu;WANG Chishe(School of Computer Science and Engineering,Anhui University of Science and Technology,Huainan Anhui 232001,China;School of Network and Communication Engineering,Jinling Institute of Technology,Nanjing Jiangsu 211169,China)
出处
《信息与电脑》
2022年第4期101-104,共4页
Information & Computer
基金
交通运输部2020年重点项目“基于5G主网的智慧高速全息感知及车路协同控制研究与场景应用”(项目编号:2020-ZD3-029)。
