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

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.

💡 Research Summary

The detection and analysis of small infrared targets are critical tasks in various high-stakes domains, including defense, aerospace, and surveillance. A persistent and significant challenge in this field arises when these small targets are located in dense clusters. In such scenarios, the overlapping signals of closely-spaced targets make it extremely difficult to accurately determine the number of targets, their precise sub-pixel positions, and their individual radiation intensities. While deep learning has significantly improved target detection capabilities, the specific problem of “unmixing”—the process of separating superimposed signals from overlapping targets—remains largely unexplored due to the inherent complexity of the task and the lack of standardized infrastructure.

To address this critical gap, this paper introduces DISTA-Net (Dynamic Iterative Shrinkage Thresholding Network), a pioneering deep learning architecture specifically designed for the unmixing of closely-spaced infrared small targets. The core innovation of DISTA-Net lies in its ability to reimagine traditional sparse reconstruction techniques within a dynamic deep learning framework. Unlike conventional Convolutional Neural Networks (CNNs) that rely on static filters, DISTA-Net adaptively generates convolution weights and thresholding parameters in real-time, tailored to the specific characteristics of the input signal. This dynamic approach allows the network to effectively disentangle overlapping features, achieving superior accuracy at the sub-pixel level.

Beyond the architectural innovation, the authors have established a comprehensive open-source ecosystem to catalyze further advancements in the field of infrared target unmixing. This ecosystem consists of three foundational pillars: (1) CSIST-100K, a large-scale, publicly available benchmark dataset designed for training and testing models in dense target environments; (2) CSO-mAP, a novel evaluation metric specifically engineered to assess the precision of sub-pixel detection and separation; and (3) GrokCSO, an integrated open-source toolkit that provides access to DISTA-Net and other state-of-the-art models. By providing the necessary infrastructure—data, metrics, and tools—this research not only presents a superior algorithmic solution but also provides the scientific community with the essential resources to drive future progress in infrared imaging and small target analysis.