Page 63 - 无损检测2025年第四期
P. 63

土木工程中的无损检测技术应用
              土木工程中的无损检测技术应用

              DOI:10.11973/wsjc240413



                   装配式钢结构建筑抗侧力支架缺陷超像素级

                                              Gabor 识别方法




                                                           刘晏长
                                         (北京艺苑风景园林工程有限公司,北京 100076)

                       摘  要:为提升装配式钢结构建筑的安全性与可靠性,设计了一种装配式钢结构建筑抗侧力支
                   架缺陷超像素级Gabor识别方法。首先对抗侧力支架缺陷图像实施归一化处理,通过图像旋转、错
                   切变换、镜像翻转处理扩充数据集。然后利用由多模态特征抽取机制、鲁棒像素相似度评估以及像
                   素至超像素软映射策略组成的基于多任务学习的图像超像素分割方法实施抗侧力支架缺陷图像超
                   像素分割。最后通过二维Gabor滤波器对生成的超像素实施Gabor滤波,提取超像素特征,依据提
                   取的Gabor局部相位特征与Gabor局部方向特征,通过支持向量回归(SVR)实现抗侧力支架缺陷
                   识别。试验结果表明,设计方法能够识别各种抗侧力支架缺陷,对于所有尺寸的缺陷,设计方法的
                   假阳性率均较低,对于精细缺陷的识别假阳性率仅为0. 025。
                       关键词:装配式钢结构建筑;抗侧力支架缺陷;图像超像素分割;像素至超像素软映射;Gabor
                   识别;支持向量回归

                       中图分类号:TP391;TG115.28      文献标志码:A    文章编号:1000-6656(2025)04-0033-06

                  Superpixel level Gabor identification method for defects in anti lateral force brackets of
                                           prefabricated steel structure buildings


                                                         LIU Yanchang
                                 (Beijing Yiyuan Landscape Engineering Co., Ltd., Beijing 100076,China)

                      Abstract: In order to improve the safety and reliability of prefabricated steel structure buildings, a superpixel level
                   Gabor  identification  method  for  the  defects  of  anti-lateral  force  brackets  in  prefabricated  steel  structure  buildings  was
                   designed.  The images of defects in the anti-lateral force bracket were normalized. The dataset was expanded through
                   image rotation, cropping transformation, and mirror flipping processing. A multi-task learning based image superpixel
                   segmentation method consisting of multimodal feature extraction mechanism, robust pixel similarity evaluation, and pixel
                   to superpixel soft mapping strategy was used to implement anti lateral force stent defect image superpixel segmentation. For
                   generating superpixels, their Gabor filtering was implemented through a two-dimensional Gabor filter to extract superpixel
                   features. Based on the extracted Gabor local phase features and Gabor local directional features, SVR was used to identify
                   defects in anti-lateral force brackets. The experimental test results showed that the design method could identify various
                   lateral force resistant bracket defects. For all sizes of defects, the false positive rate of the design method was low, and the
                   false positive rate for identifying fine defects was only 0.025.
                      Key  words:  prefabricated  steel  structure  building;  defect  in  lateral  force  resistant  bracket;  image  superpixel
                   segmentation; pixel to superpixel soft mapping; Gabor recognition; SVR

                                                                     在装配式钢结构建筑领域,随着工业化进程的
                 收稿日期:2024-08-27
                                                                加快,对结构安全性的要求日益提升 。抗侧力支
                                                                                                  [1]
                 作者简介:刘晏长(1973—),男,本科,工程师,主要研究方向为
                                                                架作为关键承重与稳定构件,其性能直接关系到整
              土木建筑工程施工和管理
                                                                                           [2]
                 通信作者:刘晏长,qzpudom@163. com                      体建筑的抗震能力和安全水平 。然而,实际工程中,
                                                                                                          33
                                                                                         2025 年 第 47 卷 第 4 期
                                                                                                  无损检测
   58   59   60   61   62   63   64   65   66   67   68