基于深度学习的X射线胶片数字化与缺陷检测算法

Algorithm of X-Ray Film Digitization and Defect Detection Based on Depth Learning

X射线胶片数字化和焊缝缺陷自动检测对提高航天大型零件生产加工质量和检测效率具有重要意义。在某些特定场景中,X射线检测无法采用数字式接收器,将X射线胶片转化为数字图像是缺陷识别的前提,但现有方法难以实现X射线胶片的高保真数字化,此外,大型零件的焊缝缺陷具有小目标特点,人工判读和传统图像检测算法难以保证识别精度。本文提出了一种基于深度学习的X射线胶片缺陷检测算法,首先基于全卷积神经网络在X射线胶片上不同曝光时间的图像中自动选择曝光时间最佳的数字图像,然后设计了基于轻量级MoGaA网络的缺陷检测网络,实现了数字化图像中的小目标缺陷检测。数字化和检测结果表明,该算法对于焊缝缺陷检测的准确率可达96%,取得了良好的检测效果。

The digitization of X-ray film and automatic detection of weld defects are of great significance for improving the production and processing quality and detection efficiency of large aerospace parts.In some specific scenes,the digital receiver cannot be used for X-ray detection,and the transformation of X-rayfilm into digital image is the premise of defect recognition.However,it is difficult to realize the highfidelity digitization of X-rayfilm by existing methods.In this paper,an algorithm of X-rayfilm defect detection based on depth learning is proposed.Firstly,based on full convolution neural network,the digital image with best exposure time on the X-rayfilm is automatically selected from the images with diffierent exposure time.Then a defect detection network based on lightweight MoGaA network is designed to detect small target defects in X-ray digital images.The digitization and detection results show that the accuracy of this algorithm for weld defect detection can reach 96%,and good detection effect is obtained.