基于深度学习的燃气PE管道焊缝缺陷检测

为了解决人工与传统数字图像处理方法进行燃气PE管道焊缝缺陷识别时面临的效率低、漏检率高、评片效果不佳等问题,提出了基于深度学习算法的燃气PE管道焊缝缺陷智能检测方法,实现从输入燃气PE管道焊缝DR检测图像到输出缺陷种类及其测量值的精细化测量。首先,在宏观区域层面采用YOLOv5网络预提取缺陷区域,减少与缺陷相似的非目标区域的干扰,并设计了融合坐标注意力机制(CA)与加权双向特征金字塔网络(BiFPN)的CA-BiFPN模块,以提高对小目标缺陷检测能力,其最终的缺陷识别定位平均精确度为95.1%。然后,在微观边界层面采用语义分割算法Deeplabv3+,实现像素级别的缺陷分割,缺陷分割平均像素准确率为91.25%、平均交并比值为85.52%。最后,在几何特征层面采用最小外接矩形法计算其实际尺寸大小,其平均相对误差为5.47%。结果表明该检测方法可实现燃气PE管缺陷高效率、高精度、智能化检测。

Gas PE Pipeline Weld Defect Detection Based on Deep Learning

In order to solve the problems of low efficiency, high missed detection rate and poor evaluation effect when manual and traditional digital image processing methods are used to identify gas PE pipeline weld defects, an intelligent detection method for gas PE pipeline weld defects based on deep learning algorithms is proposed. The fine measurement from inputting gas PE pipeline weld DR Detection image to outputting defect types and their measured values were realized. Firstly, YOLOv5 network was used to pre-extract the defect region at the macro region level to reduce the interference of non-target regions similar to the defect. And a CA-BIFPN module integrating coordinate attention mechanism (CA) and weighted bidirectional Feature Pyramid network (BiFPN) was designed to improve the detection ability of small target defects. The final average accuracy of defect identification and location was 92.5%. Then, at the micro boundary level, Deeplabv3+, a deep semantic segmentation network, was used to achieve pixel-level defect segmentation. The average pixel accuracy of defect segmentation was 91.25%, and the average intersection-union ratio was 85.52%. Finally, the minimum enclosing rectangle method was used to calculate the actual size at the geometric feature level, and the average relative error was 5.48%. The results show that the method can achieve high efficiency, high precision and intelligent detection of gas PE pipe defects.