基于多层Hopfield神经网络的X射线焊缝气泡检测

NEW ALGORITHM FOR DETECTING AIR BUBBLES IN STEEL PIPE WELDING OF X-RAY BASED ON HOPFIELD NEURAL NETWORK

提出利用Hopfield神经网络来分割X射线焊缝图像以判断焊缝是否存在气泡,将焊缝图像的分割问题转化为一个优化问题进行处理。针对焊缝图像噪声大、气泡出现位置随机的特点,构造Hopfield神经网络的能量函数。通过试验计算,确定能量函数系数的选取原则。在此基础上,提出基于神经网络的X射线焊缝图像分割算法,算法结合中值滤波和神经网络以便有效地去除噪声和检测气泡。对某实际生产线的焊缝图像进行处理的结果表明,中值滤波结合多层Hopfield神经网络可以准确地检测到焊缝中的气泡。

In order to detect the air bubbles in welding gap, the multi-layer Hopfield neural network is presented to segment welding X-ray image. The image segmentation is posed as an optimization problem. The energy function is constructed to meet the characteristics of welding X-ray image such as great noise and random positions of air bubbles. The principle of selecting coefficient is given through some experiments. A new algorithm for segmenting welding X-ray image is also put forward based on multi-layer Hopfield neural network. The algorithm is combined with median filtering and neural network to wipe off noise and find air bubbles effectively. As an application, the algorithm successfully segments some real industrial welding X-ray images.