Aiming at the identification and classification requirements of welding defects, the electromagnetic eddy current detection technology method is used to establish a welding data acquisition system, the continuous wavelet transform (CWT) is used as a preprocessing technology to reveal rich electromagnetic information, and a network model of CBAM-inception-res is proposed. For classification, the inception multi-scale convolution is integrated into the front end of the network, the receptive fields of different sizes are increased, and the features of different scales are spliced and merged to improve the classification accuracy. The verification of this method shows that the accuracy of discrimination can reach 99.88%.