The utilization of compact unmanned aircraft systems (UAS) for military reconnaissance and surveillance is experiencing growth in the intelligence branch. Obtaining a large amounts of data by these means leads to the need for their quick and efficient processing for further use within the commander’s decision-making process (battle management). This paper focuses on the automatic detection of military reconnaissance and surveillance objects, such as vehicles or soldiers, in aerial images by employing the YOLOv8 object detector, a convolutional neural network (CNN) model. To achieve a high detection success rate across diverse military equipment, weather conditions, and geographic locations, a comprehensive dataset comprising thousands of images is essential for training the neural network. However, publicly available datasets of this nature are scarce, presenting a significant challenge. This study utilizes a custom image set with more than ten thousand annotated objects, whereas the data were collected from internet databases, social networks, and military training the authors participated in. The data enabled us to examine distinct preprocessing approaches and model training setups to deliver beneficial findings for future research directions. The model performance results indicate a promising detection success rate according to standard evaluation metrics; however, to ensure the algorithm’s robustness for practical applications, a considerably larger amount of data will be required.