Anomalous individuals searching framework for space debris detection in single optical astronomical image

Abstract

With the growing concern over space debris, effective space surveillance is significant to prevent potential collisions. Traditional approaches often design specific algorithms for distinct surveillance tasks, thereby neglecting the shared aspects across these tasks. Such methods frequently exhibit limited robustness, particularly when confronted with external interference or shifts in scene dynamics. This study introduces a space debris detection framework within single astronomical image, conceptualizing the task as an anomalous individuals searching challenge. The framework is structured around three core modules: individuals extraction, feature extraction and anomaly detection. Utilizing this framework, a versatile methodology is designed, which has been rigorously tested across two primary observational contexts. According to the similarity between ideal and actual imaging, our method begins by extracting sources within a normalized correlation space. It then compiles a comprehensive feature matrix for each source, encompassing motion, intensity, and morphological attributes. By exploiting the inherent low-rank characteristics and sparsity of the feature matrix, we identify foundational feature vectors for stars. Anomalous sources are subsequently identified via the Mahalanobis distance, facilitating the identification of targets. The method is validated through both simulated and actual observed datasets, with 97.7% average detection accuracy, outperforms than eight classical methods across various scenarios. Given the modular nature of the framework, each component can be refined to accommodate more complex situations. Moreover, the uniform anomaly scores generated offer valuable confidence for subsequent tracking algorithms, which underscore the potential of the framework in advancing practical space surveillance endeavors.