基于Mask R-CNN模型的铁路隧道衬砌机制砂混凝土裂缝视觉检测

Visual detection of sand concrete cracks in railway tunnel lining mechanism based on Mask R-CNN model

铁路隧道衬砌结构裂缝图像具有复杂的灰度分布和变化特征,局部和全局的多特征信息会干扰跟踪方向和边界跟踪参数,模型可扩展性受限,检测准确率较低。为此,提出基于Mask R-CNN模型的铁路隧道衬砌机制砂混凝土裂缝视觉检测方法。首先输入分段线性变换后的砂混凝土裂缝图像,抽取阈值,生成连通域标识,再以像素点为背景点,在Mask R-CNN模型中,同时检测裂缝区域的位置和标记像素级的边缘掩膜,判定裂缝边界起点与裂缝宽度;然后进行累加视觉检测方法设计,按照裂缝的几何特征以及排序结果,求解裂缝长度,获得完整的裂缝轮廓。试验结果表明,所提方法可以较为完整地检测所有关键位置,裂缝参数信息检测准确率较高;迭代次数升高后,检测结果受到的影响较小,可扩展性得到了改善,可适应任务需求,具有较好的应用价值。

The crack image of railway tunnel lining structure has complex grayscale distribution and variation characteristics,and local and global multi feature information interferes with tracking direction and boundary tracking parameters.The model’s scalability is limited,and the detection accuracy is low.Therefore,a visual detection of sand concrete cracks in railway tunnel lining mechanism based on Mask R-CNN model was proposed.Firstly,after inputting the segmented linear transformation of the sand concrete crack image,a threshold was extracted to generate a connected domain identifier.Pixel points were used as background points in the Mask R-CNN model to simultaneously detect the position of the crack area and mark pixel level edge masks,the starting point and width of the crack boundary was determined,and a cumulative visual detection method was designed.Based on the geometric characteristics and sorting results of the cracks,the crack length was calculated,and the complete crack contour was obtained.The experimental results showed that after using the method proposed in this paper,all key positions can be detected completely;After increasing the number of iterations,the detection results were less affected,indicating that its scalability had been improved and can adapt to task requirements;and therefore,it has good application value.