Abstract: An improved unmanned aerial vehicle (UAV)-YOLO algorithm is proposed to address the problems of low precision and high computation of UAV detection in infrared scenes. First, a weighted bidirectional feature pyramid network (BiFPN) was introduced to enhance the model detection performance by optimizing multi-scale feature fusion. Second, a lightweight shared detail enhanced convolutional detection head (LSDECD) was used to enhance the performance of small-target detection while decreasing the number of parameters, and a convolution and attention fusion module (CAFM) was constructed to strengthen the feature interactions to enhance the robustness. Finally, a Wise–SIoU loss function was used to accelerate model convergence. The experimental results demonstrate that the improved model achieves 91.3% mAP@50, with a 1.7% enhancement compared to the original YOLOv11n. Validation under a weak aircraft target detection and tracking dataset of public infrared images shows that the improved model improves all evaluation indices, proving that it has good generalization and robustness.