钢筋混凝土桥梁表面裂缝智能识别及评价技术研究

Research on Intelligent Identification and Evaluation Technology of Cracks on the Surface of Reinforced Concrete Bridge

为实现钢筋混凝土桥梁损伤的准确定位和快速修复,以常见的裂缝损伤为研究对象,采用自适应匀光和直方图均衡算法增强裂缝目标特征信息,进行图像预处理。通过优化YOLO-v5和ATT-UNet算法,分别完成裂缝图像细观尺度目标快速智能识别及像素分割。采用改进的最短距离法定量分析不同形态裂缝的细节特征尺寸,构建深度学习模型评价体系。通过智能识别模型预测和实测真实数集对比实验,得到训练后模型的精确率、召回率、调和平均F1值和交并比IoU分别为89.47%、96.23%、92.73%和86.44%,裂缝长度和宽度识别计算误差分别控制在±3.2mm和±0.35mm范围内,从而验证了钢筋混凝土桥梁表面裂缝智能识别及评价技术的有效性、可靠性与泛化性。

In order to realize the accurate location and rapid repair of the damage of the reinforced concrete bridge,taking the common crack damage as the research object,the self-adaptive light diffusion and histogram equalization algorithm are used to enhance the crack target feature information,and the image preprocessing is carried out.By optimizing the YOLO-v5 and ATT-UNet algorithms,the rapid intelligent recognition and pixel segmentation of mesoscale targets in crack images are completed respectively.The improved shortest distance method is used to quantitatively analyze the detailed feature size of different morphological cracks,and the evaluation system of deep learning model is constructed.Through the intelligent recognition model prediction and the comparison experiments of the measured real number set,the precision ratio,recall rate,harmonic average F1 value and merging ratio IoU of the trained model are 89.47%,96.23%,92.73% and 86.44% respectively,and the error of crack length and width recognition calculation is controlled in the range of ±3.2mm and ±0.35mm respectively,which verifies the effectiveness,reliability and generalization of the intelligent identification and evaluation technology of cracks on the surface of the reinforced concrete bridge.