An Edge Morphology-Aware Self-Correcting Framework for Precise Infrared UAV Detection

Precise infrared UAV detection is of critical importance for ensuring all-day security surveillance and maintaining the airspace safety. However, existing methods suffer from performance degradation due to the challenges such as the semantic misalignment, the insufficient morphological feature extraction, and the inadequate multi-scale adaptability under low-light conditions. To solve the issues aforementioned, we propose an edge morphology-aware self-correcting framework and a corresponding cross-space self-coordinated network. Specifically, the framework synergizes spatial-semantic alignment with dynamic morphological prior bounding box based on the feature interaction. Moreover, a grouped multi-core collaboration module is introduced to enhance the multi-scale feature fusion through an adaptive weighting process. Furthermore, a position distribution-based morphology modulation model dynamically optimizes the prior bounding box by leveraging the spatial probability estimation. Besides, a morphology-aware key-value coupling module establishes geometric-semantic alignment using learnable virtual feature templates, thereby enabling robust background suppression. Extensive experiments validate the state-of-the-art performance of our method, achieving 2.1% higher detection accuracy and reducing positioning deviation by 3.1% compared to existing methods.