Wavelet-guided temporal photometric network for multi-frame infrared small target detection

Multi-frame infrared small target detection is challenging due to extremely weak target responses, complex background clutter, and the tendency of tiny targets to be suppressed during deep feature extraction and multi-scale fusion. To address these issues, we propose a wavelet-guided temporal photometric network (WTPNet), a prior-guided three-branch cross-attention framework for moving infrared small target detection. Instead of directly stacking consecutive frames, WTPNet decouples multi-frame information into motion-sensitive temporal cues, context-aware temporal representations, and current-frame spatial appearance. Specifically, a temporal context prior is constructed by jointly using temporal discrete cosine transform and Haar wavelet decomposition, where the former captures motion-sensitive photometric variations across frames and the latter separates low-frequency background structures from high-frequency local details. This prior guides the temporal context branch to emphasize target-related variations while suppressing background-dominated responses. In addition, a directional photometric residual enhancement strategy preserves weak shallow responses in the current frame, and a large-context scale-focused fusion module alleviates target dilution during multi-scale feature aggregation. Experiments on two public datasets demonstrate that WTPNet outperforms representative single-frame and multi-frame detectors. Compared with the baseline, WTPNet improves average precision from 59.10% to 64.06% and from 85.85% to 90.11% on the two datasets, respectively. These results validate the complementarity of temporal-frequency prior learning, photometric response preservation, and scale-focused fusion for robust moving infrared small target detection. The code will be released at https://github.com/vazheaven/WTPNet .