In practical applications, frequency hopping (FH) signals play a crucial role in UAV communication, radar, navigation, and other fields due to their high confidentiality and low probability of interception. Therefore, the effective classification and identification of FH signals hold paramount importance in ensuring national security and enhancing military combat effectiveness. In this paper, we propose a novel method for modulation recognition of FH signals using Graph Convolutional Networks (GCN). Initially, we extract distinct features from the FH signals and construct adjacency matrix. Subsequently, we design a GCN to further extract signal features and accomplish modulation recognition. The experimental results demonstrate the effectiveness of our approach, achieving a remarkable recognition rate of 81.8% at a signal-to-noise ratio of −10dB. This performance represents a significant improvement compared to the current mainstream time-frequency transformation method.