Infrared small target detection (IRSTD) supports remote-sensing surveillance and early warning, but infrared targets often occupy only a few pixels, exhibit weak appearance cues, and resemble thermal clutter. Vision-language detectors such as Grounding DINO provide a flexible prompt-driven detection interface, but direct transfer to IRSTD faces three mismatches: a single prompt cannot describe target appearance diversity, generic decoder queries are not calibrated for weak scale-sensitive targets, and IoU-style localization losses are unstable for few-pixel boxes. We propose PromptScaleDINO as a lightweight adaptation framework for Grounding DINO. It introduces an Anchor Prompt Bank (APB) to enrich the text input while keeping supervision connected to a stable anchor phrase, a Scale-Aware Query Refinement Network (SQRN) to refine selected queries according to predicted box scale and semantic confidence, and a geometry-aware localization objective that combines Normalized Wasserstein Distance (NWD) with Generalized Intersection over Union (GIoU). Under a unified detection protocol, PromptScaleDINO achieves its largest gains on the challenging real-scene IRSTD-1k benchmark while maintaining near-ceiling performance on SIRST and NUDT-SIRST. On IRSTD-1k, it reaches 88.32% F1 and 87.40% mAP@0.5, improving the baseline by 2.56 and 4.00 percentage points, respectively. These gains mainly reflect improved weak-target recovery at comparable precision; accordingly, we position the framework as a target-sensitivity and localization adaptation rather than an explicit false-alarm suppression method.

Infrared stripe noise, mainly caused by detector nonuniformity and readout inconsistencies, is a common structured degradation in infrared imaging systems, exhibiting pronounced directional bias, strong column-wise persistence, and distinctive frequency-domain characteristics. These properties make generic restoration networks prone to a trade-off between stripe suppression and detail preservation: conventional two-dimensional attention often allocates modeling capacity to stripe-irrelevant spatial dependencies, early downsampling may entangle directional stripe components with scene structures, and skip connections in U-shaped architectures can reintroduce shallow residual stripe features into the decoder. Residual stripe artifacts and restoration-induced structural distortions can further impair downstream infrared image analysis, particularly small-target detection. To address these issues, we propose a Spatio-Frequency Synergistic Progressive Restoration Network (SFPRNet) for infrared image destriping. SFPRNet progressively exploits the column-wise statistical characteristics of stripe noise, directional frequency information during early scale transformation, and selective cross-level feature refinement to enhance stripe discrimination and suppression while preserving structural details. Extensive experiments on synthetic and real infrared images demonstrate that the proposed SFPRNet achieves superior or competitive performance across most datasets and degradation settings, providing a favorable balance between destriping quality and detail preservation. Furthermore, downstream evaluation with multiple infrared small-target detectors demonstrates that SFPRNet improves subsequent small-target detection performance.

To address the problem of detecting dim small targets against complex moving backgrounds, based on spatiotemporal saliency features, this paper proposes a novel extraction model: the temporal and spatial saliency feature (TSSF) model. In the temporal domain processing stage, a temporal motion saliency operator is constructed by leveraging the differences in temporal motion responses between the target and the background; this operator simultaneously filters for candidate target points while suppressing large-area background clutter. In the spatial domain processing stage, a max-mean-based grayscale saliency operator is constructed to further validate the candidate target points. Building upon target enhancement processing, target trajectories can be effectively extracted through multi-frame association. Quantitative experiments on ten test sequences with diverse background complexities and signal-to-clutter ratios (SCRs) of around 1 or lower show that the TSSF model achieves optimal background suppression factors (BSFs) on seven sequences and the maximum SCR gain (SCRG) across eight sequences, with an average SCRG of roughly 330. For detection performance assessed via ROC curves, the proposed method obtains the highest AUC values on nine test sequences. The consistently superior performance validates the effectiveness and generalizability of the proposed TSSF model for dim small target enhancement, background suppression, and robust target detection.