一种条带池化道路裂缝自动提取方法

An automatic extraction method of Strip Pool road cracks

针对现有路面裂缝自动化提取精度低和效率差的问题,该文提出一种结合条带池化改进U-Net网络的道路裂缝自动提取方法。该方法以U-Net网络为基础,将编码器与残差模块、空洞卷积相结合,增加网络深度扩大感受野,丰富裂缝提取信息、有效抑制噪声;使用注意力机制将编码与解码过程相连接,提升裂缝提取效率和准确率;以条带池化模块替换池化层,解决传统裂缝分割方法对条状特征提取精度差的问题。以CFD数据集为例,将该文方法与U-Net等其他4种提取方法进行对比分析。结果表明,结合条带池化改进U-Net网络的道路裂缝自动提取方法提取的裂缝完整,计算时间短,在指标F1上有明显提升。

Aiming at the low accuracy and poor efficiency of existing road crack automatic extraction,this paper proposes an automatic extraction method of road cracks combined with Strip Pooling improved U-Net network.Based on the U-Net network,First,the encoder is combined with the residual module and the hole convolution to increase the depth of the network to expand the receptive field,enrich the crack extraction information,and effectively suppress noise.Then,use the attention mechanism to connect the encoding and decoding processes to improve the efficiency and accuracy of crack extraction.Finally,the Strip Pooling module is used to replace the pooling layer,which solves the problem of poor accuracy of strip feature extraction in traditional fracture segmentation methods.Taking the CFD data set as an example,this method is compared with other four extraction methods including U-Net.The results show that the cracks extracted by automatic extraction of road cracks based on Strip Pooling U-Net network are complete,the calculation time is short,and the index F1 is significantly improved.