This paper proposes a knowledge-and-data-driven graph neural network-based collaboration learning model for reliable aircraft recognition in a heterogeneous radar network. The aircraft recognizability analysis shows that: (1) the semantic feature of an aircraft is motion patterns driven by the kinetic characteristics, and (2) the grammatical features contained in the radar cross-section (RCS) signals present spatial-temporal-frequency (STF) diversity decided by both the electromagnetic radiation shape and motion pattern of the aircraft. Then a STF graph attention convolutional network (STFGACN) is developed to distill semantic features from the RCS signals received by the heterogeneous radar network. Extensive experiment results verify that the STFGACN outperforms the baseline methods in terms of detection accuracy, and ablation experiments are carried out to further show that the expansion of the information dimension can gain considerable benefits to perform robustly in the low signal-to-noise ratio region.