基于空洞分层注意力胶囊网络的X射线焊缝缺陷识别方法

X-ray Weld Defect Identification Based on Dilated Hierarchical Attention Capsule Network

由于X射线焊缝图像的复杂多样性,使得很多传统基于X射线焊缝缺陷检测方法的准确性不高,泛化能力较差。提出一种基于空洞分层注意力胶囊网络(DHACNet)的X射线焊缝缺陷识别方法。DHACNet由卷积模块、空洞分层注意力和胶囊网络(CapsNet)组成。卷积模块用来提取图像的卷积特征,空洞分层注意力用来提取多尺度显著性特征,CapsNet利用胶囊层和动态路由算法替代卷积神经网络(CNN)中的池化操作和全连接操作。DHACNet具有强大多尺度特征提取能力,能够克服CNN只关注图像局部特征和池化操作导致图像部分信息丢失等不足。在构建的X射线焊缝缺陷图像集上进行识别试验,识别准确率为96%以上,与传统方法进行比较,结果表明,该方法有效可行,能够为X射线焊缝缺陷识别系统提供技术支持。

Due to the complexity and diversity of X-ray images,many traditional methods based on X-ray weld defect detection have poor accuracy and generalization.An X-ray weld defect identification method based on dilated hierarchical attention capsule network(DHACNet)is proposed.DHACNet consists of convolution layer,dilated hierarchical convolution,attention mechanism,and capsule network(CapsNet).Convolution layer is used to extract the convolutional features of images,dilated hierarchical convolution and attention layer are used to extract multi-scale significance features.CapsNet uses capsule layer and dynamic routing algorithm to replace the pooling operation and full connection operation in convolutional neural network(CNN).DHACNet has powerful multi-scale feature extraction capability,which can overcome the drawbacks of CNN,such as focusing only on the local features of the image and the loss of partial image information caused by pooling operation.The experiments are carried out on the constructed X-ray weld defect image set,and its recognition accuracy reached over 96%.Compared with the traditional method the results show that the improved method is effective,and can provide technical support for the X-ray weld defect recognition system.