Answers are grounded in this paper and cite their sources.
Detecting dim small targets in infrared image sequences remains challenging because point-like targets are easily confused with structured clutter, while platform motion degrades inter-frame correspondence. This paper presents a fixed-pixel spatiotemporal saliency framework with candidate-gated adaptive background modeling. Robust quantile normalization and quality-controlled subpixel registration are first employed to reduce radiometric variation and reject unreliable frame alignment. A multi-scale spatial point response and a locally normalized positive residual are combined to generate a soft target likelihood, which adaptively controls the background learning rate and reduces the risk of weak-target absorption. At temporally aligned pixel locations, temporal residual energy, stability, and run-length continuity are accumulated over accepted frames. Spatial and temporal evidence is then integrated through weighted geometric fusion, followed by robust local constant false-alarm rate detection, multi-frame confirmation, and non-maximum suppression. Experiments on six infrared sequences, including three self-collected UAV sequences and three public ATR sequences, show that the proposed method achieves the highest signal-to-clutter ratio gain (SCRG) on four sequences and ranks among the top three methods on all six. Its mean SCRG is 7.4603, which is 38.34% higher than that of the strongest baseline. ROC comparisons further demonstrate favorable performance in the low-false-alarm region. These results indicate that jointly exploiting spatial target compactness and temporally consistent residual responses improves dim small-target detection in challenging infrared image sequences.
No takes yet. Share an insight, caveat, or question.
Xin Li (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: