Abstract: This paper proposed a Graph-in-Graph Convolutional Network (G-GCN) model based on ultrasound-guided waves for the rapid detection and localisation of structural defects in composite laminates. G-GCN was constructed an advanced representation map of the spatio-temporal characteristics of guided wave signals. The local map was constructed by local global transformation to represent the relationship information in a single guided wave signal. Then, the global map was constructed based on the local map to represent the relationship information between multiple guided wave signals. The global map input map was used to convolve neural network model training and learning, and the corresponding composite defect prediction was output, It realized fast and accurate defect detection and location with very few sensors. Finally, an experimental platform for nondestructive testing of ultrasonic guided wave composite materials was built to verify the advancement and reliability of G-GCN.