DISTA-Net: Dynamic Closely-Spaced Infrared Small Target Unmixing

TL;DR

This work tackles the challenge of unmixing closely spaced infrared small targets by reframing CSIST unmixing as a dynamic sparse reconstruction problem. It introduces DISTA-Net, a dynamic deep unfolding network that adaptively generates convolution weights and proximal-thresholds conditioned on input data to achieve precise sub-pixel localization and intensity estimation. The authors also establish an open-source CSIST ecosystem—CSIST-100K, CSO-mAP, and GrokCSO—to enable benchmarking and reproducibility. Experimental results on CSIST-100K demonstrate superior localization accuracy and reconstruction quality with efficient computational cost, validating the effectiveness of dynamic proximal mappings in CSIST unmixing.

Abstract

Resolving closely-spaced small targets in dense clusters presents a significant challenge in infrared imaging, as the overlapping signals hinder precise determination of their quantity, sub-pixel positions, and radiation intensities. While deep learning has advanced the field of infrared small target detection, its application to closely-spaced infrared small targets has not yet been explored. This gap exists primarily due to the complexity of separating superimposed characteristics and the lack of an open-source infrastructure. In this work, we propose the Dynamic Iterative Shrinkage Thresholding Network (DISTA-Net), which reconceptualizes traditional sparse reconstruction within a dynamic framework. DISTA-Net adaptively generates convolution weights and thresholding parameters to tailor the reconstruction process in real time. To the best of our knowledge, DISTA-Net is the first deep learning model designed specifically for the unmixing of closely-spaced infrared small targets, achieving superior sub-pixel detection accuracy. Moreover, we have established the first open-source ecosystem to foster further research in this field. This ecosystem comprises three key components: (1) CSIST-100K, a publicly available benchmark dataset; (2) CSO-mAP, a custom evaluation metric for sub-pixel detection; and (3) GrokCSO, an open-source toolkit featuring DISTA-Net and other models. Our code and dataset are available at https://github.com/GrokCV/GrokCSO.

Figures (5)

• Figure 1: Conceptual illustration of imaging and unmixing processes for closely-spaced infrared small targets (CSIST). CSIST unmixing aims to disentangle and accurately estimate the count, positions, and intensities of overlapping targets.
• Figure 2: CSIST Visualization: The top row shows 1 to 5 overlapping targets, and the following rows display unmixing results for sub-pixel division factors of $3 \times$, $5 \times$, and $7 \times$.
• Figure 3: Division of each pixel into an $n\times n$ grid of sub-pixels, representing potential target positions.
• Figure 4: Architecture of the proposed DISTA-Net. The overall framework consists of multiple cascaded stages. Each stage contains three main components: a dual-branch dynamic transform module ($\mathcal{F}_d^{(k)}$) for feature extraction, a dynamic threshold module ($\Theta_d^{(k)}$) for feature refinement, and an inverse transform module ($\tilde{\mathcal{F}}^{(k)}$) for reconstruction.
• Figure 5: Visual comparison of $3 \times$ sub-pixel division reconstruction for scenes containing different numbers of closely-spaced infrared small targets. The red boxes highlight regions where targets exhibit significant sub-pixel characteristics.