In high-altitude aerial images, targets usually occupy just a few pixels, rendering them vulnerable to being concealed by complex and cluttered backgrounds. Additionally, factors like motion blur, varying illumination, scale variation, and occlusion further undermine the detection performance. To address these challenges, we propose an efficient small target detection algorithm for UAV - view scenarios, integrating dynamic routing attention and collaborative knowledge distillation mechanisms. Firstly, we construct an Adaptive Slicing Inference (ASI) module based on spatial entropy analysis, which achieves dynamic enhancement of feature density. In addition, we design a Bi-level Routing Attention (BRA) mechanism to adaptively captures multi-scale spatial dependencies and enhances feature representations by dynamically focusing on target-relevant regions. Based on thus, in the model light-weighting process, we further propose a Collaborative Knowledge Distillation (CKD) module. CKD module combines both global context and fine-grained object details from the teacher model to achieve high-fidelity knowledge transfer to the student models. Experiments on VisDrone2021 and UD-Small2025 datasets demonstrate that our algorithm improves the mAP@0.5:0.95 index by 14.8% compared with the baseline models, and the inference speed reaches 153 FPS.