【摘要】 针对施工道路影像中因背景信息复杂导致道路提取错分、漏分及边缘粗糙的问题，提出了一种融合多元特征的E-TransUNet模型施工道路要素提取方法。E-TransUNet模型通过设计多元特征增强模块对图像特征信息进行增强；在模型下采样中融入空洞空间金字塔池化(atrous spatial pyramid pooling, ASPP)模块，增强网络对道路影像多尺度特征的提取能力；跳跃连接部分加入卷积注意力(convolutional block attention module, CBAM)模块，从不同维度上捕获道路特征之间的相关性；最后组合采用Dice loss和CE loss作为损失函数解决样本数量不均衡问题。结果表明，该方法对施工道路要素的提取在OA、MIoU和MPA指标分别达到了93.30%、80.37%和91.19%,相比其他网络U-Net、DeeplabV3+、Swin-Unet、HRNet和SegFormer提取效果更好，为施工道路提供了准确的要素提取方法。更多还原

【Abstract】 In order to solve the problems of missegmentation, missing segmentation and rough edges in construction road image due to complex background information, this paper proposes an E-TransUNet model method of construction road element extraction by integrating multiple features. The E-TransUNet model enhances the image feature information by designing multiple feature enhancement module. The ASPP(atrous spatial pyramid pooling) module is integrated into the model subsampling to enhance the network’s ability to extract multi-scale features of road images. The CBAM(convolutional block attention module)module is added to the skip connection part to capture the correlation between road features in different dimensions. Finally, Dice loss and CE loss are used as loss functions to solve the problem of unbalanced sample quantity. The results show that the method proposed in this paper achieves 93.30%,80.37% and 91.19% in OA,MIoU and MPA indexes, respectively, which is better than that of U-Net, DeeplabV3+,Swin-Unet, HRNet and SegFormer. It provides an accurate extraction method for construction road.更多还原