基于红外热图像增强的市政道路护栏立柱焊接节点缺陷深度检测

Deep detection of welding node defects in municipal road guardrail columns based on infrared thermal image enhancement

  • 摘要: 市政道路护栏立柱焊接缺陷深度不一,常规红外检测方法会因为热扩散效应,使不同缺陷程度的焊接节点温差信号趋同,导致检测结果错误,产生结构安全性的误判,从而给市政道路的安全运行带来潜在风险。因此通过红外热图像完成市政道路护栏立柱焊接节点缺陷深度检测。通过多尺度引导滤波和局部显著图权重融合的方法对原始焊接红外热图像基础层与细节层分别进行增强处理,降低热扩散导致的模糊效应,突出缺陷区域的温度梯度特征,使微缺陷可视化。通过增强后的图像映射为加权图,并利用基于高斯拟合的亮度模型和显著性约束项突出缺陷区域,采用最大流/最小割算法求解图割问题,得到缺陷区域。由此,基于区域生长准则,根据缺陷区域的平均灰度和当前生长点的灰度进行区域生长,通过比较缺陷深度与护栏立柱的厚度,准确地捕捉缺陷的真实边界,完成缺陷深度的识别。实验结果表明,所提方法分割处理后得到的结果与实际焊接节点缺陷区域重叠程度接近99%,且在对500张红外热图像进行缺陷深度检测过程中,准确率高达99%。说明所提方法可以精准高效的完成市政道路护栏立柱焊接节点缺陷深度检测,保证市政道路的安全。

     

    Abstract: The depth of welding defects on the guardrail columns of municipal roads varies. Conventional infrared detection methods can cause the temperature difference signals of welding nodes with different degrees of defects to converge due to thermal diffusion effects, resulting in incorrect detection results and misjudgment of structural safety, which poses potential risks to the safe operation of municipal roads. Therefore, the depth detection of welding node defects in municipal road guardrail columns is completed through infrared thermal imaging. By using multi-scale guided filtering and local saliency map weight fusion, the base layer and detail layer of the original welding infrared thermal image are respectively enhanced to reduce the blurring effect caused by thermal diffusion, highlight the temperature gradient characteristics of the defect area, and visualize micro defects. By mapping the enhanced image to a weighted image and using a Gaussian fitting based brightness model and saliency constraint to highlight the defect area, the maximum flow/minimum cut algorithm is used to solve the image cutting problem and obtain the defect area. Therefore, based on the region growth criterion, region growth is carried out according to the average gray level of the defect area and the gray level of the current growth point. By comparing the defect depth with the thickness of the guardrail column, the true boundary of the defect is accurately captured, and the recognition of the defect depth is completed. The experimental results show that the overlap between the segmentation results obtained by the proposed method and the actual welding node defect areas is close to 99%, and the accuracy of defect depth detection on 500 red external thermal images is as high as 99%. The proposed method can accurately and efficiently complete the de

     

/

返回文章
返回