A feature fusion enhanced multiscale CNN with attention mechanism for spot-welding surface appearance recognition

As the most important welding method in the welding process of automobile body-in-whites, resistance spot welding still relies on manual inspection of appearance quality to eliminate unqualified products, which leads to a low detection efficiency and a high error rate. In response to this problem, the modular design method is adopted in this paper and a multiscale convolution assemble block is proposed firstly, which can effectively distinguish the differences in similar welding spots. Then a dual-use attention block is designed to calibrate the spatial and channel information of welding spot feature maps, so that the model can pay more attention to the valid welding spot features. Several such multiscale blocks with attention mechanism are stacked to generate the main body of the convolutional neural network (CNN) for welding spot appearance recognition. Finally, the de-pooling and feature fusion strategies are used to further improve the computational efficiency and precision, and the model is named AcmNet. The results of classification experiments demonstrated that the proposed strategies are efficient and the accuracy of AcmNet is 95.2%, which is higher than the existing models and very suitable for the body-in-white production lines.