Answers are grounded in this paper and cite their sources.
Dynamic anomaly detection of space targets is critical for space situational awareness. With the range-Doppler projection mechanism, inverse synthetic aperture radar (ISAR) image sequences possess significant potential for interpreting abnormal dynamics of noncooperative targets. Previous studies on anomaly detection have primarily focused on low-frequency bands, neglecting the richer pixel-level details available in terahertz (THz) ISAR images. This article proposes an anomaly detection method utilizing THz ISAR image sequences, featuring a spatial-entropy dual-branch temporal classification network with Image Spatial Feature Extraction (ISFE) and Boundary-Guided Feature Fusion (BGFF) modules. The ISFE module employs a convolutional neural network (CNN) to automatically extract spatial features from ISAR image sequences. The BGFF module captures statistical features including image entropy and contrast for each frame, and integrates them into CNN feature maps in a structured spatial manner. Ultimately, the long short-term memory temporal modeling is applied to generate sequence classification results. Experimental results on the THz ISAR dataset demonstrate the superior performance of the proposed architecture.
No takes yet. Share an insight, caveat, or question.
Hu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: