基于深度学习的钢筋混凝土桥梁掉块露筋病害识别

Identification of Spalled Concrete and Exposed Reinforcement in Reinforced Concrete Bridge Based on Deep Learning

为对桥梁结构表面混凝土掉块、露筋进行定量分析,将深度学习算法与传统图像处理技术相结合,采用深度学习算法DeepLabV3+模型对桥梁结构表面病害进行语义分割,将各病害图像语义分割结果转化为2幅二值化图像,然后统计二值化图像中连通区域的数量、面积等参数,最后根据像素标定值得出掉块、露筋缺陷区域的实际物理尺寸。将基于深度学习的病害识别算法用于无人机桥梁检测系统中,对某混凝土桥梁表面病害进行检测,试验结果表明该算法病害识别准确率高,检测误差小,可以满足实际工程应用。

The deep learning algorithm and the conventional image processing technique are used in combination to conduct quantitative analysis of spalled concrete and exposed reinforcement in bridges.First,deep learning algorithm method DeepLabV3+is employed to do semantic image segmentation on the surface damages of the bridge;then,the semantic segmentation results of each deterioration image are converted to two pieces of binary images;furthermore,parameters,such as the number and area of connected regions in the binary images are counted;at last,the actual physical dimensions of spalled concrete and exposed reinforcement are obtained according to the pixel calibration values.The damage identification method based on deep learning has been used in the UVA bridge detection system to detect the surface damages of a concrete bridge.The results of the detection prove that the damage detection accuracy of the method is high,with minimal error,which can meet the demands of actual bridge projects.