基于深度特征金字塔的路面病害检测  

Pavement Disease Detection Based on Depth Feature Pyramids

作　　者：张伟 刘宁钟[1] 寇金桥 ZHANG Wei;LIU Ning-zhong;KOU Jin-qiao(School of Computer Science and Technology,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China;Ark Key Laboratory,Beijing Institute of Computer Technology and Application,Beijing 100854,China)

机构地区：[1]南京航空航天大学计算机科学与技术学院,江苏南京211106 [2]北京计算机技术及应用研究所方舟重点实验室,北京100854

出　　处：《计算机技术与发展》2022年第12期173-178,共6页Computer Technology and Development

基　　金：中央高校基本科研业务费专项资金(3082020NZ2020017)。

摘　　要：道路为人们的生活和工作提供了方便,路面作为道路最重要的组成部分,直接影响了道路的使用,但由于车辆行驶和风吹日晒,路面病害层出不穷。利用目标检测技术对路面病害进行快速检测,可以及时筛选出病害,降低日常人工检查的负担,提高养护效率。然而,路面病害特征比较细微,随着神经网络深度的不断增加和下采样,细节信息损失比较多。通过将通道注意力集成到特征金字塔网络,可以从通道和空间两个维度上提高网络对路面病害的表征能力,同时提出了一种新的路面病害特征提取器,使得网络更关注低层次特征。实验部分,将改进后的特征金字塔分别应用在Road Damage Dataset 2018数据集和自制的沥青路面病害数据集上,并与其他经典的目标检测模型进行了比较,实验结果证明了基于改进后的特征金字塔的模型在路面病害检测上的有效性。The road provides convenience for people’s life and work,and the pavement,as the most important component of the road,directly affects the use of the road.Due to vehicle movement and wind and sun,pavement diseases are endless.Rapid detection of pavement diseases using object detection technology allows timely screening of diseases,reduces the burden of daily manual inspection,and improves maintenance efficiency.However,the pavement disease features are relatively subtle,and more detailed information is lost as the depth of the neural network continues to increase and downsample.By integrating channel attention into the feature pyramid network,the network’s ability to characterize pavement distress can be improved in both channel and spatial dimensions,and a new pavement disease feature extractor is proposed to make the network more focused on low-level features.In the experimental part,the improved feature pyramid is applied to the Road Damage Dataset 2018 dataset and the homemade asphalt pavement disease dataset,respectively,and compared with other classical object detection models.The experimental results show the effectiveness of the model based on the improved feature pyramid for pavement disease detection.

关 键 词：路面病害 目标检测 特征金字塔 通道注意力 特征提取器 

分 类 号：TP31[自动化与计算机技术—计算机软件与理论]