Low-altitude small unmanned aerial vehicle (UAV) frequently invade sensitive areas, which posing a serious threat to national and social security. Aiming at the problems such as high missed detection rate and insufficient detection accuracy for UAV target detection based on thermal imaging, the infrared detection of UAV-YOLO (IDOU-YOLO) algorithm model was proposed. A cross-scale fusion feature pyramid mechanism was constructed to fully explore the feature space information, focus on the cross-scale information fusion and the rich information representation ability of the model, and enhance the target recognition ability. At the same time, the bounding box loss function Scylla IoU (SIoU) was introduced to improve the detection accuracy and accelerate the convergence speed of the model in the training process. The experimental results show that the precision, recall, F₁ score, mAP@0.5 and mAP@0.5:0.95 reach 99.2%, 96.3%, 97.7%, 98.4% and 70.2%, respectively, which indicates that the IDOU-YOLO model improves the detection and recognition ability of UAV targets in various scenarios, and can better meet the application requirements of anti-UAV systems.