HGMMC: A Space Target Detection Algorithm Based on Hierarchical Gaussian Mixture Model Clustering

Reducing the image segmentation threshold is a crucial approach for detecting space targets with a low signal-to-noise ratio (SNR). However, this reduction introduces a significant amount of noise points and stars, which severely hampers the detection of space targets. In addition, space targets located at the detectable edge of brightness may be lost in certain frames, increasing the challenge of trajectory association. This article proposes an algorithm based on hierarchical Gaussian mixture model clustering (HGMMC) for detecting space target trajectories amid a starry background with low SNRs. The algorithm initially employs a fast local contrast measure (LCM) method to extract potential target points and eliminates star background interference by leveraging the principle of angular distance invariance of stars. Subsequently, it introduces a subtrajectory model that incorporates both static and dynamic features of space targets, with these subtrajectory features serving as the sample dataset. Employing a hierarchical clustering approach, it adaptively determines the number of classification clusters. The target trajectory is then determined using HGMMC. Ultimately, it utilizes the connected components method of the connectivity matrix to merge similar trajectories. The algorithm fully integrates feature information from space targets and effectively filters out noise points’ interference while correlating trajectories in scenes with numerous noise points and single-frame target loss, accurately identifying multitarget trajectories within the field of view.