The rapid advancement of technology has endowed modern unmanned aerial vehicles (UAVs) with high mobility, flexibility, and user-friendly automation, making them pivotal for small object detection in complex urban environ-ments. However, this task is challenged by frequent occlusions, tiny object sizes, and complex, dynamic backgrounds. To address these issues, this paper takes the YOLOv10b algorithm as the basic framework and proposes an improved UAV multi-target detection scheme named AM-YOLOv10 (Adaptive Multi-scale YOLOv10). This scheme adopts a multi-dimensional optimization approach: optimizing the neck network to enhance the efficiency of target feature extraction; designing an Anti-interference Adaptive Receptive Field Module (AARF) to effectively solve detection problems caused by target occlusion, overlap, and complex backgrounds; introducing the Multi-Feature Enhancement and Fusion Module (MFEF) that integrates channel and position attention to suppress interference from complex backgrounds and improve the recognition of target features; and adding a P2 high-resolution small target detection head to adapt to the detection of small targets in UAV shooting scenarios. Comprehensive experiments on public benchmark VisDrone2021 , demonstrate that our method achieves superior performance and robustness in UAV-based object detection scenarios.