Multi-function radar has gained widespread application due to its flexible working modes， agile waveform characteristics， and ability to perform multiple tasks in parallel. However， these capabilities also pose significant challenges to radar intelligence reconnaissance and countermeasures. Recognizing the working modes of multi-function radar serves as a fundamental step for subsequent threat assessment， adaptive countermeasures， and guided attacks， directly influencing the specificity and effectiveness of radar countermeasures. In this paper， a novel method for recognizing the working modes of multi-function radar， leveraging relational graph convolutional networks （RGCNs） is proposed. By analyzing the various working modes of multi-function radar， the method enables parallel data processing and addresses the interactions between different working modes and their characteristic parameters.