首页|改进U-Net的无人机航拍路面破损检测方法

改进U-Net的无人机航拍路面破损检测方法

扫码查看
为了探究无人机航拍视角下细小裂缝难以检测和检测出现断裂的问题,基于U-Net架构提出了ASE-Net网络.首先,使用改进的VGG-16作为编码器,以便提取破损的特征信息;其次在最小尺度网络层引入多尺度特征融合(MSFF)模块、通道增强条形池化(CESP)模块;最后在解码阶段添加ECA_X注意力模块.实验结果表明,本文模型在自建的无人机航拍路面破损数据集上的mIoU、mPA、mPrecision分别达到0.820 9、0.930 2、0.865 1,相比于基线U-Net分别提高15.97%、12.72%、11.02%.最后,在开源数据集Crack500中验证了本文模型相比于其他主流模型具有更强的性能和泛化能力.模型能实现对路面细小裂缝、坑槽、修补的准确检测,有效解决了裂缝检测的断裂问题,可提升大尺寸航拍图像路面破损检测效果.
Improved pavement damage detection method of UAV based on U-Net
In order to explore the difficulties in detecting fine cracks and the occurrence of breaks in the aerial view of UAVs,a network called ASE-Net is proposed based on the U-Net architecture.First,an improved VGG-16 is used as the encoder to extract the broken feature information.Second,multi-scale feature fusion block(MSFF)module and channel enhanced strip pooling(CESP)module are introduced at the minimum scale network layer.Finally,the ECA_X attention module is added to the decoding stage.The experimental results indicate that the model presented in this paper achieves a mIoU of 0.820 9,a mPA of 0.930 2,and a mPrecision of 0.865 1 on the self-constructed UAV aerial pavement breakage dataset.These results represent improvements of 15.97%,12.72%,and 11.02%over the baseline U-Net,respectively.Ultimately,the model in this work has been demonstrated to exhibit better performance and generalization ability than other standard models utilizing the open-source dataset Crack500.The model can realize accurate detection of small cracks,potholes,and repairs on the road surface,effectively solving the fracture problem of crack detection,and enhancing the effect of pavement damage detection in large-size aerial images.

unmanned aerial vehicledamage detectionsemantic segmentationstrip poolingmulti-scale featureU-Net

韩建峰、张静、宋丽丽、陶永昭

展开 >

内蒙古工业大学信息工程学院 呼和浩特 010080

内蒙古自治区感知技术与智能系统重点实验室 呼和浩特 010051

无人机 破损检测 语义分割 条形池化 多尺度特征 U-Net

2024

电子测量技术
北京无线电技术研究所

电子测量技术

CSTPCD北大核心
影响因子:1.166
ISSN:1002-7300
年,卷(期):2024.47(19)