The cooperative combat is a significant aspect of air combat game, and accurately identifying the cooperative tactical intent of opponent’s formation is essential to accelerating the OODA loop. This paper proposes a novel framework for cooperative intent recognition in the cooperative air combat game scenarios, which is based on self-supervised learning and utilizes the multimodal information obtained from multi-sensor. By taking different modal information with consistent timestamp as positive sample pairs, and employing the modal contrast loss to aggregate multimodal and temporal information from sensor signals into account, the robust representations of characteristics for each timestamp are extracted from unlabeled multimodal time series based on temporal continuity and inter-modal correlation. Finally, a downstream classifier is used to output results for the cooperative tactical intent of opponent. In order to verify the effectiveness of the proposed method, experiments were conducted using air combat game simulation dataset of dual-aircraft formation air combat game. The experimental results demonstrate that the superiority of our method is exhibited in terms of accuracy and generalization via the comparison to several traditional supervised learning methods and general self-supervised methods without requiring a large amount of manually labelled data.