Intention Recognition of Spatial Non-Cooperative Targets Based on HMM

To address the challenge of recognizing the behavioral intentions of non-cooperative spacecraft targets in space, this paper proposes a dynamic time-series analysis method based on the hidden Markov model (HMM). This approach enables intention recognition without reliance on external prior knowledge (e.g., target control models). The model learns the evolution patterns of typical target behaviors during the training phase and performs intention inference based on observation sequences during the testing phase. By comprehensively considering the characteristic features of typical non-cooperative behaviors, a behavioral sample dataset is constructed using the Monte Carlo shooting method. Three-dimensional observation sequences—comprising target distance, horizontal entry angle, and relative velocity—are defined, and a left-to-right HMM structure is designed to characterize the four-stage evolution of three types of intentions: hovering, rendezvous, and fly-around. Model parameters are learned via maximum likelihood estimation, and the forward algorithm calculates the log-likelihood of observation sequences to achieve accurate intention recognition. Numerical simulations validate the effectiveness of the proposed method.