The infrared anti-interference technique of missiles under the background of complex air combat is one of the core technologies of infrared air-to-air missiles. Aiming at the fact that traditional static Bayesian networks cannot express the dynamic relationship of feature variables in sequence images in time series, this paper proposes an anti-jamming recognition algorithm for a space-time correlation inference network that conforms to the process of human visual inference and recognition. First, the proposed space-time association reasoning network takes into account the feature space constraint relationship, introduces prior knowledge of the time constraints of feature variables, and establishes a target reasoning network recognition model that expresses the characteristic spatiotemporal relationship, thereby enhancing the stability of sequence image target recognition. Second, a sample set is built through simulation data, offline training and learning the space-time correlation inference network structure and feature jump probability parameters, to determine the probabilistic inference network to identify the offline model. Finally, based on the test data, the model is combined with the inference identification network model to perform probabilistic inference to achieve recognition and classification of targets. The experimental results show that the anti-jamming recognition rate based on the spatiotemporal correlation inference network reaches 94% under the condition of the interference of the infrared decoy, which is 3% higher than the static Bayesian network anti-jamming recognition algorithm, which effectively improves the stability of target recognition.