【摘要】 为提高Deeplabv3+模型对桥梁裂缝的分割精度与检测效率，提出一种多分支卷积块和Deeplabv3+改进的桥梁裂缝分割算法。在Deeplabv3+模型中加入RFB多分支卷积模块，将Deeplabv3+中的backbone替换为Mobilenetv2,用深度分离卷积替换算法中所有普通卷积，增加一次底层特征融合。将改进模型与主流的图像检测模型如PSPNet、U-Net在相同数据集下进行实验对比，结果表明，改进后的Deeplabv3+模型对桥梁裂缝的检测具有较好的效果，检测精度可达90.15%,较原始模型提高了4.07%。改进后的模型分割精度和速度有明显提高，对于完成裂缝检测任务具有实际应用价值。更多还原

【Abstract】 To improve the accuracy and efficiency of segmentation of bridge cracks by Deeplabv3+ model, an improved bridge crack segmentation algorithm based on multi-branch convolutional block and Deeplabv3+ is proposed. The RFB(receptive field block) multi-branch convolutional block is added to the Deeplabv3+ model, the backbone in Deeplabv3+ is replaced with Mobilenetv2, all ordinary convolutions in the algorithm are replaced with depth-separated convolutions, and the one underlying feature fusion is added. The improved model is compared experimentally with mainstream image detection models such as PSPNet and U-Net under the same data set. The results show that the improved Deeplabv3+ algorithm has a better effect on the detection of bridge cracks, and the detection accuracy can reach 90.15%, which is 4.07% better than the original algorithm. The improved model segmentation accuracy and speed are significantly improved and have practical application value for crack detection tasks.更多还原