MAPPR++-Net: Multi-Path Attention and Progressive Perception Refinement Plus Plus network for Infrared Small Target Detection

Robust detection of diminutive infrared signatures represents a cornerstone capability for contemporary surveillance systems, catastrophe early-warning platforms, and precision-guided weaponry; however, prevailing algorithms demonstrate critical performance degradation when confronting adverse operational conditions characterized by suppressed signal-to-noise ratios, ambiguous target geometries, and dense background interference that collectively precipitate elevated false alarm probabilities and compromised detection fidelity. This research introduces a Multi-Path Attention Residual and Perception Progressive Refinement network specifically architected to mitigate these deficiencies. The proposed framework harnesses a Multi-path Attention Residual Fusion plus plus (MARF++) module employing dynamic attention weighting to orchestrate adaptive local-global feature optimization for enhanced granular discrimination in cluttered environments. Complementing this, the Small Target Key information Enhancement (STKE) and Key-Aware Feature Fusion (KAFF) modules collaboratively amplify shallow-layer representations, while a Cross-Level Feature Progressive Refinement (CFPR) module implements iterative fusion of salient low-level cues with high-level semantic hierarchies to substantially improve contour preservation and detection resilience. Hierarchical refinement is realized through dual specialized components: an Attention-Guided Refinement Module (AGRM) that concentrates computational resources on target-specific middle-level regions, and a Context-Aware Refinement Module (CARM) engineered to capture expansive long-range dependencies within deep feature spaces. Comprehensive evaluation across three standard infrared datasets confirms that the developed methodology achieves superior detection performance relative to existing state-of-the-art approaches, establishing new benchmarks for accuracy and robustness in challenging background scenarios.