Weld Feature Extraction Based on Fully Convolutional Networks

Based on the feature learning ability of deep convolutional neural networks, a weld feature extraction method based on fully convolutional networks is proposed. In this method, the fully convolutional networks is used to predict the pixels containing the feature information of the weld, and the edge feature information of weld is supplemented by the fusion of low-level and high-level feature information. The results show that the method can get the weld position accurately under the interference of strong arc and soot particles, and has the advantages of strong anti-interference ability and accurate recognition.