To improve the time consumption and low accuracy of traditional manual methods of laser cladding crack detection, an automatic identification method combined with the attention model is proposed to identify and detect laser cladding cracks. The semantic segmentation network of laser cladding cracks based on the U-net network cannot sufficiently extract small local features. By adding the convolutional block attention model (CBAM) layer to extract the feature space and feature channel weight information, we can label the microscopic cracks of the laser cladding zone without any time difference in the pixel level. Experimental results show that the deep learning model combined with the CBAM can improve the accuracy of cladding crack identification and detection by 2.7 percentages. The network fused with the CBAM achieves an accuracy of 79.8% on the cladding area crack test set. Both the labeling accuracy and speed of the deep learning model exceed those of manual labeling, providing an effective method for identifying laser cladding cracks.