In complex environments such as heavy occlusion and low illumination, unimodal detection methods in anti-UAV tasks suffer from limited accuracy, while existing multimodal fusion methods are vulnerable to noise interference, making it difficult to effectively extract small target features and complementary information. To address these challenges, this paper proposes a Context-Aware Cross-Modal Correction and Fusion Network (CCCFN). Firstly, an Adaptive Context Selection (ACS) module is introduced, which enhances the feature representation of aerial small targets by optimizing multi-scale receptive fields through large-kernel convolution decomposition and capturing long-range spatial dependencies via strip pooling. Secondly, a Cross-Modal Corrective Fusion Module (CCFM) is introduced, which employs a cross-attention mechanism to perform noise correction between RGB and infrared features, further enhancing the consistency and spatial representation of the fused features. Finally, experiments on the self-constructed Det-fly-IRGAN dataset demonstrate that the proposed method effectively integrates complementary multimodal information and outperforms existing approaches in terms of recognition accuracy and robustness.