Images of military targets typically exhibit characteristics such as camouflage, varying scales, and uneven distribution, making the task of target detection complex and challenging. Furthermore, the limited computing resources of unmanned platforms, such as drones and ground unmanned vehicles, make it difficult to deploy detectors with large parameter counts on these systems. In this study, we enhanced the components of YOLOv8n and proposed a new lightweight military target detection algorithm named YOLO-E. We constructed a military target dataset composed of armed personnel holding various weapons to facilitate the verification of different algorithms. In our designed algorithm, we applied an efficient multi-scale convolution module in the feature extraction network to improve the detection speed of military targets. Additionally, we designed a head network based on weight sharing, significantly reducing the model parameters. We also proposed a novel bounding box loss function, the Normalized Corner Distance IoU, to further enhance the detection accuracy of military targets. We tested YOLO-E on a self-developed military target dataset. Experimental results showed that compared to the original YOLOv8n algorithm, YOLO-E improved detection accuracy by approximately 1.30%, increased detection speed by 1.68%, reduced parameter count by 30.87%, and decreased computational complexity by 37.33%. Furthermore, we compared our method with several advanced object detection algorithms. The results demonstrated that YOLO-E also outperformed in terms of comprehensive parameters, real-time performance, and accuracy. The proposed network model provides effective auxiliary support for analyzing battlefield situations.