To address the challenges of weak feature discriminability and low detection accuracy caused by small object sizes, dense distribution, and complex backgrounds in UAV aerial imagery, this paper proposes a global-local collaborative perception network for object detection. First, inspired by the “overview-then-scrutiny” visual cognition mechanism, we design a Global-Local Collaborative Dynamic Network (GLCD-Net) that leverages global semantic priors to guide dynamic kernel generation, enhancing global context modeling. Second, we construct a C2PSA-MALA module that integrates magnitude-aware linear attention, where a magnitude-aware factor strengthens the focus on weak feature signals while reducing computational complexity without sacrificing detection accuracy. Third, we introduce a Frequency-Aware Background Smoothing Attention (FBSEMA) module, which suppresses high-frequency background noise through spatial-frequency collaborative optimization, thereby improving feature representations for small objects. Experimental results on the VisDrone2019 and TinyPerson datasets demonstrate that the proposed method achieves competitive detection performance.