【摘要】 针对传统路面裂缝检测准确度不高、受噪声信息干扰等问题，提出一种改进的Deeplabv3+网络的路面裂缝语义分割方法。上述算法在Deeplabv3+网络的基础上，使用密集连接的方式重构网络ASPP模块，通过跳跃连接共享信息获取更大的感受野。并针对背景和裂缝区域所占像素比例相差较大的特点，设置权重让神经网络更加关注裂缝特征。最后对检测结果中出现的边缘断裂，采用最大连通域方法实现裂缝的精细提取。在公共裂缝数据集CFD和CRACK500上实验表明，所提算法的MIOU分别为85.7%和87.2%,与其它语义分割算法相比，新方法能够准确完整地实现路面裂缝提取。更多还原

【Abstract】 Aiming at the problems of low accuracy of traditional pavement crack detection and interference by noise information, an improved Deeplabv3+network semantic segmentation method for pavement cracks is proposed. Based on the Deeplabv3+network, the algorithm uses dense connection to reconstruct the network ASPP module, and obtains a larger receptive field by sharing information through hopping connections. In view of the large difference between the background and the proportion of pixels in the crack area, the weights are set to make the neural network pay more attention to the crack characteristics. Finally, for the edge fractures that appear in the detection results, the maximum connected domain method is used to realize the fine extraction of the fractures. Experiments on the public crack data set CFD and CRACK500 show that the MIOU of the proposed algorithm is 85.7% and 87.2%, respectively. Compared with other segmentation algorithms, the proposed method can accurately and completely extract road cracks.更多还原