基于计算机视觉的公路边坡裂缝监测方法

Computer Vision-based Method for Monitoring Road Slope Cracks

为及时发现裂缝问题并减少潜在的危害,提出一种公路边坡裂缝的自动监测方法。以公路边坡裂缝为研究对象,针对裂缝图像形态不规则、周边环境干扰大的特点,设计了公路边坡裂缝分割网络(slope cracks segmentation network,SCSNet)和公路边坡裂缝几何参数计算方法。该网络首先采用编码器逐渐捕获更高层次的语义特征,其次采用解码器通过逐渐恢复空间信息并结合跳跃连接融合不同尺度间的信息,然后利用通道注意力机制,学习每个通道间的特征,增强裂缝的特征表达。另外,提出参数计算方法,基于连通域分析得到裂缝连通域,并计算裂缝长度、宽度、面积几何参数。实验结果表明:裂缝分割网络的平均交并比达87.86%,该网络能较好地提取公路边坡裂缝特征;裂缝几何参数计算方法能较准确地测算裂缝的当前状态。

An automatic monitoring method for highway slope cracks was proposed,aiming to timely detect crack problems and reduce potential hazards.Taking highway slope cracks as the research object,for the characteristics of irregularity of crack image pattern and large interference of surrounding environment,a highway slope cracks segmentation network(SCSNet)and a highway slope crack geometric parameter calculation method were designed.An encoder was used to gradually capture higher-level semantic features,and a decoder was used to fuse information between different scales by gradually recovering spatial information and combining jump connections.Then,for the case of complex road slope crack images and other situations,a channel attention mechanism was used to learn the features between channels and enhance the feature representation of cracks.A method based on the connectivity domain analysis was proposed to obtain the crack connectivity domain and calculate the crack length,width and area geometric parameters.Results show that the average intersection ratio of the segmentation network reaches 87.86%,which can extract the highway slope crack features better,and the current state of the cracks can be measured more accurately using the crack geometric parameter calculation method.