Lightweight Intent Recognition Method Based on Diffusion Model

To address the challenges posed by imbalanced and limited battlefield data, which typically results in complex models prone to overfitting during training, we introduce a novel diffusion model grounded in Wasserstein distance (WDiffusion) tailored for the multi-categorical and multivariate characteristics inherent to intent recognition data. Subsequently, we propose a streamlined tactical intent recognition framework predicated on gate recurrent unit (GRU), designed to enhance model responsiveness, and train it on an innovative dataset. Comparative experimental analyses corroborate that the synthetic data generated via WDiffusion significantly outperforms other prevalent generation models. Furthermore, the WDiffusion-GRU model achieves a recognition accuracy of 97.09%, surpassing current aerial target intent recognition models by more than 1.07%. This advancement maintains high recognition precision while substantially curtailing model parameters, thereby amplifying the agility and reliability of battlefield commanders’ decision-making processes.