Abstract: To address the issues of small target size, low signal-to-noise ratio, complex background, and susceptibility to noise in unmanned aerial vehicle (UAVs) infrared images, we propose an improved lightweight detection model, PNL-YOLO (PConv NWD Lite -You Only Look Once), based on the YOLOv11n architecture. In the feature extraction stage, a partial convolution was introduced to reduce the computational load while maintaining the feature expression ability of the model. A small-target detection layer is added, and the large-target detection layer is removed to optimize the multiscale detection capability. Meanwhile, the detection head was redesigned in a lightweight manner to reduce computational complexity. Additionally, the normalized Wasserstein distance (NWD) was adopted as the similarity metric, effectively solving the problem of sensitivity to the position deviation of small targets and improving the detection accuracy. The experimental results showed that, on the HIT-UAV dataset, compared with the original YOLOv11n model, the average detection accuracy (mAP50) of this model increased by 6.08%, the number of parameters and computational load decreased by 42.24% and 17.46%, respectively, and the model size was only 3.2MB. In summary, this method can significantly improve the detection performance of small targets while maintaining real-time performance, to provide an effective solution for target-detection tasks in complex infrared scenes captured by UAVs.