<p>The detection of key parts of military objects is as important as the full detection of military objects, which is of great significance for improving the precise strike capability of unmanned aerial vehicle (UAV). In order to solve the challenges of low detection precision and high missed-detection rate of military objects and key parts in UAV images, and to meet the requirements of real-time detection and easy deployment on model, we propose a new model for the detection of military objects and key parts, named military object and key part-you only look once (MOKP-YOLO). MOKP-YOLO designs a unified detector, a key part feature integration module, a class-wise feature guidance module, and a cross-task loss function, which can adequately capture contextual, semantic, and spatial dependence between military objects and key parts. Moreover, we create a new MOKP-UAV dataset to expand the diversity of existing military object datasets of UAV images. The experimental results on the self-built dataset and real UAV images show that MOKP-YOLO has a significant advantage in detecting military objects and key parts in UAV images. In addition, extensive ablation studies and generalization performance analysis further validate the effectiveness of MOKP-YOLO.</p>