Battlefield Intention Recognition Based on Multivariate Time Series Representation and Few-Shot Learning

In battlefield situation assessment, the rapidity and accuracy of intention recognition is always a difficult problem to solve. Considering the multi-dimensional and sequential nature of battlefield situation, propose a novel battlefield intention recognition universal framework based on multivariate time series representation and few-shot learning. Under the few-shot learning mode, the combination of enemy targets and our targets is encoded as input vector, the feature vector is obtained by using Multivariate Time Series representation, and the intention result is generated by Sequence Generation network. Experiments show that MST+SG network can recognize battlefield intention accurately and in real time, which verifies the robustness of few-shot learning in battlefield intention recognition.