文献类型：期刊文献

中文题名：基于FCN的路面裂缝分割算法

英文题名：Pavement crack segmentation algorithm based on FCN

作者：韩静园[1];王育坚[1];谭卫雄[1];李深圳[1]

第一作者：韩静园

机构：[1]北京联合大学智慧城市学院,北京100101

第一机构：北京联合大学智慧城市学院

年份：2022

卷号：41

期号：6

起止页码：146-149

中文期刊名：传感器与微系统

外文期刊名：Transducer and Microsystem Technologies

收录：CSTPCD;;北大核心:【北大核心2020】；CSCD:【CSCD_E2021_2022】；

基金：国家自然科学基金资助项目(61872042);北京联合大学研究生科研创新资助项目(YZ2020K001)。

语种：中文

中文关键词：全卷积神经网络;裂缝;挤压和激励;图像分割

外文关键词：fully convolutional networks(FCN);crack;squeeze and excitation;image segmentation

摘要：为了准确地对路面裂缝进行分割,提高复杂背景下路面裂缝分割的效果,将全卷积神经网络(FCN)用于路面裂缝分割。考虑FCN无法对裂缝边缘、图案和形状等特征分配权重,提出一种基于FCN的路面裂缝分割改进算法。通过分析特征通道之间的关系,在FCN的池化层加入挤压和激励模块。利用权重对原始特征在通道上进行重标定,以提升重要特征和抑制无用特征,自适应地为裂缝边缘、图案和形状等特征分配权重。使用真实的路面裂缝数据集对改进模型进行训练,完成了模型的测试和验证。实验结果表明:改进模型的路面分割效果良好,分割结果的误检率、漏检率和DSC综合评价指标等都有较好的改善。
In order to segment pavement cracks accurately and improve the effect of pavement crack segmentation under complex backgrounds,fully convolutional neural networks(FCN)are used for pavement crack segmentation.Considering that FCN cannot assign weights to features such as crack edges,patterns and shapes,an improved algorithm for pavement crack segmentation based on FCN is proposed.By analyzing the relationship between the feature channels,an extrusion and excitation module is added to the pooling layer of the FCN network.Use weights to re-calibrate the original features on the channel to enhance important features and suppress useless features,and adaptively assign weights to features such as crack edges,patterns and shapes.The real road crack dataset is used to train the improved model,and the test and verification of the model are completed.The experimental results show that the improved model has a good pavement segmentation effect,and the misdetection rate,missed detection rate and DSC comprehensive evaluation index of the segmentation results are improved.