Enhanced Data Augmentation and Hierarchical LSTM for UAV Intent Prediction

In modern aerial warfare, situational awareness plays a critical role in enhancing combat effectiveness, and intent recognition in air combat is the core element for achieving this objective. Addressing the limitations of traditional air combat intent recognition methods in complex operational environments, this paper proposes a novel multitask intent recognition method based on feature enhancement. This method first enhances the correlation between feature space and sample space through contrastive learning, effectively addressing the challenge of imbalanced aerial target datasets. Subsequently, a hierarchical variable-length Long Short-Term Memory (LSTM) model is employed to extract and process information across different time scales, leading to a more comprehensive and accurate understanding of target behavior. Experimental results indicate that our method demonstrates significantly superior performance in complex air combat environments. It notably enhances recognition stability, accuracy, and speed, showcasing strong potential for practical applications.