Abstract: To address the challenges of complex background interference and multiscale target detection in UAV-based small object detection tasks, this study proposes a network model based on DBB-YOLOv10s. Using YOLOv10s as the baseline model, the proposed network incorporates dilated convolution (DC2f) to expand the receptive field, a bidirectional feature pyramid network (BiFPN) to achieve multiscale feature fusion, and a BAM attention mechanism to enhance the model’s focus on target regions. Experimental results on the VisDrone2021 dataset demonstrate that the proposed algorithm achieves strong performance, with a mean average precision (mAP) of 41.8% and an inference speed of 148 FPS, while maintaining a model size of 6.59 M parameters, ensuring strong practicality. Compared with existing YOLO models and their variants, the proposed model not only maintains high detection accuracy in complex scenarios but also balances real-time performance and computational efficiency, making it suitable for deployment in embedded systems and real-time UAV monitoring tasks. The results indicate that the proposed algorithm significantly enhances detection capability and generalization performance in complex environments.