When Clusters Meet Shape Priors: A Synergistic Framework for Cluster Infrared Small Target Detection

Infrared small target detection (IRSTD) is critical for low-visibility scenarios, but cluster IRSTD (CIRSTD) faces unique challenges: target ambiguity, background complexity, inter-cluster interference, and inadequate spatial modeling. Current deep learning approaches struggle with dense clusters due to limited fine-grained feature extraction, insufficient contextual constraints, and spatial resolution degradation. This paper proposes the Synergistic Shape-Contextual Fusion Network (SSCFNet) to address these issues. Firstly, the Shape Prior Self-Evolution (SPSE) module employs self-attention to model global spatial dependencies and generate structural priors for consistent detection. Accordingly, the Convolution-Shape Cross-Updating (CSCU) module integrates convolutional operations into shape representations to refine local geometric cues for low-contrast targets. Lastly, the Dynamic Synergistic Detection (DSD) module is designed to adaptively fuse individual and cluster-level features via attention mechanisms to preserve target boundaries and suppress noise. By bridging global structural and local shape refinement, SSCFNet effectively mitigates missed detections, false alarms, and localization errors in CIRSTD. Experimental results demonstrate that the proposed method outperforms state-of-the-art techniques in challenging scenarios. The code will be public at https://github.com/wangtuntun/SSCFNet.