Weld Image Recognition based on Deep Learning

In order to improve the automatic operation level of welding robot, which is very important for weld recognition, this paper proposes a method based on deep learning to identify and locate target welds for the determination of weld types and weld positions in images. Through the idea of classification and then segmentation, the three weld types used in this paper can be accurately segmented. Firstly, MobileNetV3, a lightweight network with a special bneck structure, is used to classify the three images, and SeGAN neural network is used to segment the weld images to obtain the results. In this paper, a few sample images are used for training, and then a higher accuracy is achieved by expanding the sample. The experimental results show that the accuracy of the classification results reaches 99.39%, which is 3.74% higher than that of VGG, and the accuracy of the positioning results can reach 95%, which proves the effectiveness of the method and has important significance in industrial automation welding.