STDG-FCM: Spatial and Texture Dual-Guided Fuzzy C-Means for Single-Point Supervised Infrared Small-Target Mask Generation

Infrared small-target (IRST) detection is significantly hindered by the high cost and ambiguity of pixel-level annotations, especially for dim targets in low signal-to-clutter ratio (SCR) scenes. To alleviate the annotation burden, we propose a spatial and texture dual-guided fuzzy C-means (STDG-FCM) framework for generating pseudo-masks under single-point supervision. STDG-FCM enhances region coherence through spatially weighted clustering and refines boundaries by employing a novel texture strategy based on the gray-level co-occurrence matrix (GLCM). Experiments conducted on three public datasets validate the superiority of STDG-FCM in generating high-quality pseudo-masks compared to existing methods, with an intersection over union (IoU) of 0.6578 on the SIRST dataset. Furthermore, we employ the generated masks in IRST detectors and achieve performance comparable to fully supervised training. This demonstrates that STDG-FCM is an efficient and practical mask annotation strategy.