摘要
路面的坑洞破损严重影响着驾驶舒适性与安全性,因此路面坑洞的检测与修补是一项重要的路面维护任务。本文基于改进YOLOv3-tiny目标检测神经网络模型,实现了路面坑洞的有效检测。其中,改进的模型用一些稠密块替换原模型中的大部分卷积层,使模型的深度大幅增加,而模型的参数和运行时所需的GPU内存显著降低。使用建立的路面坑洞破损数据集训练模型,以mAP作为评价模型的指标。同原YOLOv3-tiny和YOLOv4-tiny模型相比,检测速度相当,并且取得了最高mAP值。
Pavement potholes seriously affect the driving comfort and the safety.The detection and repair of the pavement potholes is particularly important in the road management.Based on the improved YOLOv3-tiny target detection neural network,this paper realized the effective detection of pavement potholes.Among them,the im-proved model replaces most of the convolutional layers in the original model with some dense blocks,which greatly increases the depth of the model,while the parameters of the model and the required GPU memory for runtime are significantly reduced.Use the established pavement pothole dataset to train models,and mean average precision is used as an indicator to evaluate the quality of the model.Compared with the original YOLOv3-tiny and YOLOv4-tiny,the detection speed is equivalent,and the improved model has achieved the highest mAP value.
作者
翟帅
冯永慧
罗宏煜
肖思航
田丽玲
吴少伟
ZHAI Shuai;FENG Yonghui;LUO Hongyu;XIAO Sihang;TIAN Liling;WU Shaowei
出处
《计量与测试技术》
2021年第9期45-49,共5页
Metrology & Measurement Technique