Infrared small target detection is a critical capability for remote sensing, wildfire prevention, and surveillance, yet conventional methods often fail when targets are tiny, low-contrast, and lack distinct shape or texture. Researchers from the Research Center for Space Optical Engineering at Harbin Institute of Technology have developed a two-stage deep learning network that combines diffusion-based feature enhancement with state-space modeling (SSM) to suppress background clutter while amplifying target signals. The approach, published on June 30, 2026, in the Journal of Remote Sensing, achieves state-of-the-art detection rates across three public datasets, significantly reducing both missed detections and false alarms in complex imaging environments.
The significance of this work lies in its ability to address a persistent bottleneck in infrared imaging: targets that occupy fewer than 81 pixels, exhibit signal-to-noise ratios around 3, and are easily submerged in background clutter. Existing deep learning methods have improved performance but typically focus only on target features, neglecting the vast background and causing severe class imbalance. The proposed Diffusion-Enhanced Dense Mamba Network (DEDM-Net) tackles this by simultaneously modeling both target and background information. Its first stage uses a dual-path diffusion model with a novel blind processing module that predicts each pixel using only surrounding information—never the pixel itself—preventing extremely small targets from being misclassified as background. The second stage introduces a dense nested Mamba architecture based on SSM, capturing long-range correlations with linear computational complexity, a significant advantage over conventional Transformers. A cross-stage prediction fusion module further integrates features from both stages, improving contour segmentation accuracy.
Evaluated on the NUAA-SIRST, NUDT-SIRST, and IRSTD-1k datasets, DEDM-Net outperformed 11 state-of-the-art methods. On NUDT-SIRST, it achieved 93.40% IoU, 93.28% nIoU, 98.37% detection probability, and a false-alarm rate of just 3.75×10⁻⁶, surpassing DNA-Net and ISTDU-Net. On IRSTD-1k, it reached 73.71% IoU and 93.89% detection probability with only 11.10×10⁻⁶ false alarms. These results underscore the method's robustness in challenging scenarios where missed detections or false alarms can have severe consequences, such as in military threat assessment or early wildfire detection. The research was published (DOI: 10.34133/remotesensing.1046) and supported by the National Natural Science Foundation of China and other funding bodies.
While DEDM-Net achieves superior accuracy, the diffusion-based two-stage design increases inference time compared to single-stage networks. Future work will focus on model distillation, mixed-precision inference, and faster samplers to reduce the required diffusion steps. The approach holds promise for real-time surveillance systems, autonomous drone navigation in low-visibility conditions, and early wildfire detection networks. As corresponding author Dr. Shikai Jiang noted, by modeling both the target-free background and potential target regions simultaneously, the diffusion-enhanced approach amplifies what matters while suppressing what doesn't, and the Mamba architecture provides the global context needed to distinguish true targets from bright clutter. The framework could also inspire new thinking about how generative models and state-space architectures can be combined for other challenging computer vision tasks where target-background separation is critical. For more information on related innovations, visit http://chuanlink-innovations.com.
