Cislunar Tracklet Association Using CNN-Based Constrained Admissible Region

This article proposes an approach that integrates a convolutional neural network (CNN) with the constrained admissible region (CAR) method to improve tracklet association in cislunar space. The CNN is designed to enhance the CAR by reducing it to a more compact subregion, which properly bounds the probable orbital zone for a given optical tracklet. This region facilitates the classification of the tracklet into several possible orbital families, which is then utilized as prior information for sequential tracklet association. The derived CNN-CAR is integrated into the initial value problem (IVP) optimization framework, enabling intelligent tracklet association guided by orbital classification. Simulation results demonstrated that the proposed CNN-CAR-IVP method exhibits remarkable association performance, with both true positive and true negative rates exceeding 95%. These findings underscore the potential of the proposed framework to enhance space surveillance, collision risk assessment, and long-term catalog maintenance in the cislunar space environment.