摘要
由于航拍公路裂缝数据缺乏并且裂缝图像存在目标小、分布复杂的特点,导致语义分割模型在航拍公路裂缝检测中效果差,影响模型在实际场景的应用,为此提出基于改进DeeplabV3+的公路裂缝检测方法。构建语义分割模型,选定DeeplabV3+模型并作如下优化:由于低级特征包含更多裂缝细节信息,增加了提取低级特征的路径,从ASPP模块输出的特征为高级特征,高级特征包含更多语义信息,将两者信息进行融合能保证模型不丢失裂缝的细节信息;在网络中嵌入SCSE注意力模块抑制对其他无关信息的响应,改善模型在裂缝数据集检测效果差的问题。实验结果表明,改进DeeplabV3+算法可以有效解决模型对小目标裂缝分割时效果差的问题,模型的检测精度提高了2.59%,具有较强的应用价值,可以为实际公路裂缝检测提供参考。
Due to the lack of aerial highway crack data and the characteristics of small targets and complex distribution of crack images,resulting in the poor effect of semantic segmentation model in aerial highway crack detection,and affecting its application in actual scenes,a highway crack detection method based on improved DeeplabV3+is proposed.The semantic segmentation model is built,and DeeplabV3+model is seletced and optimized as follows:the path of extracting low-level feature is increased due to the cracks low-level features include more detail information,the feature output from ASPP module is advanced features,which includes more semantic information,the fusion of the two information can ensure that the model does not lose the detail information of cracks.The SCSE attention module is embedded into the network to suppress the response to other irrelevant information and improve the poor detection effect of the model in the crack dataset.The experimental results show that improved DeeplabV3+can solve the problem of poor segmentation effect of the model for small target cracks,and the detection accuracy of the model is improved by 2.59%,which has strong application value and can provide reference for practical highway crack detection.
作者
张瑞燕
ZHANG Ruiyan(School of Engineering,Ocean University of China,Qingdao 266100,China)
出处
《现代电子技术》
2023年第3期100-104,共5页
Modern Electronics Technique
基金
中央高校基本科研业务费(20170105)。
关键词
道路裂缝
裂缝检测
语义分割
多尺度特征融合
注意力机制
深度学习
网络模型改进
智能检测
road crack
crack detection
semantic segmentation
multiscale feature fusion
attentional mechanism
deep learning
network model improvement
intelligent detection
